Homoepitaxial GaN micropillar array by plasma-free photo-enhanced metal-assisted chemical etching

Interest in GaN for a wide variety of optoelectronic and electronic devices continues to rise for applications, including μLEDs (Refs. 1 and 2) and high power and high frequency devices.^3,4 As device scaling continues, the capability to produce high aspect ratio (HAR) 3D GaN structures offers benefits that include footprint reduction, enhanced light extraction for light emitters,^5,6 and the development of more efficient fin field-effect transistors (finFETs) and gate-all-around transistors for high performance electronics.^4,7 However, GaN remains challenging to process for HAR micro and nanostructures, which normally requires the use of reactive ion etching (RIE). While RIE could lead to the formation of vertical and smooth side walls, it is well known that RIE causes plasma induced damage on the surface that ranges from changes in stoichiometry to the formation of nonradiative centers, both of which are detrimental to near band edge (NBE) emission and surface and interface controls.^8,9 In the case of μLEDs, RIE induced defects are shown to reduce the external quantum efficiency of devices with performance recovered only with a combination of wet etching and sidewall passivation.^10,11 RIE-based defects also cause leakage for vertical power electronic devices, where damage removal via multistep wet etching processes was necessary.^12,13 Other attempts to mitigate RIE-based damage include the modification of RIE with neutral beam etching,¹⁴ resulting in comparatively low etch rates.

Metal-assisted chemical etching (MacEtch),^15–17 on the other hand, is a plasma-free, metal-catalyzed and local (open-circuit) electrochemical etching method capable of producing HAR semiconductor structures defined by a patterned metal film in a forward (etching beneath metal) or inverse (etching outside metal) process. Silicon nanostructure arrays with pillars¹⁸ of 200:1 aspect ratio and vias¹⁹ from tens of nanometers to millimeters in diameter were enabled by gold (Au) in a forward etch process. CMOS compatible catalyst metals, including TiN and Ru, have been reported.^20,21 Compound semiconductors, including GaAs, InGaAs, and InP, have all been successfully etched using MacEtch.^22–24 For wide bandgap semiconductors, including GaN, SiC, and β-Ga₂O_3,^16,25,26 above bandgap photons generate e^-/h⁺ to initiate MacEtch, which we denote as hν-MacEtch, and is inherently inverse if the metal is not transparent to UV photons. Compared with photoelectrochemical (PEC) etching^27–30 that has been used widely for GaN as a damage-free wet etching method, hν-MacEtch is the open-circuit analog with the carrier distribution controlled by the metal catalyst patterns. While GaN etching was studied using this method (hν-MacEtch) previously,^25,31–33 significant progress is required to meet the ever demanding device needs for HAR and damage-free quality etched surfaces.

In this work, we demonstrate effective and fast etching of metalorganic chemical vapor deposition (MOCVD) homoepitaxial GaN on hydride vapor phase epitaxy (HVPE) GaN substrates using hν-MacEtch, producing large-area uniform arrays of damage-free micropillar structures.

Micropillar arrays were produced from samples of MOCVD epitaxially grown n-type silicon (Si)-doped GaN layer (2.4 μm thick) of various doping concentrations (5 × 10¹⁷, 1 × 10¹⁸, and 5 × 10¹⁸ cm⁻³) on HVPE-grown n-type GaN substrates (1 × 10¹⁹ cm⁻³). The samples were grown in a Veeco D-180 reactor using the following conditions: pressure of 200 Torr, V/III ratio of 3000, and growth temperature of 1030 °C. The sources included trimethylgallium, ammonia, and silane. XRD measurements across ten points on wafers showed an FWHM of ∼59 arcsec for the HVPE substrates, while the MOCVD layers showed an FWHM of ∼46 arcsec.

Square arrays of circular dots of approximately 3 μm in diameter and 6 μm in pitch were patterned via photolithography, followed by e-beam evaporation of 10 nm Pt with a 5 nm Ti adhesion layer and then lift-off. Samples were then etched under UV illumination (Dymax BlueWave^® 200) with a peak intensity of 26 W/cm² across 300–450 nm, with the UV source situated approximately 6.5 cm above the sample. The etch solution comprised of KOH, K₂S₂O₈, and Na₃PO₄ with varying molar ratios of 1:1:1, 1:6:6, and 1:12:12. The solution was heated to 85 °C and etched for 5–10 min. Samples were characterized using a Hitachi S4800 SEM for morphology and a Keyence VK-X1000 3D laser (405 nm) scanning confocal microscope for depth profiles. PL measurements were performed via a home-built deep UV PL system with 266 nm excitation, 40× objective of 0.5 NA (spot size ∼0.7 μm), and a Princeton Instruments PIXIS:2KBUV detector.

Figures 1(a)–1(c) show the effect of the doping concentration on surface morphology, for GaN samples with Si-doping of 5 × 10¹⁷, 1 × 10¹⁸, and 5 × 10¹⁸ cm⁻³, respectively, MacEtched in a 1:1 KOH:K₂S₂O₈ solution for 10 min. The etching stopped within the MOCVD homoepitaxial layer, and no HVPE GaN surface is revealed. The insets show the corresponding single pillar views. The roughness of the surface in-between the pillars decreases with increasing doping concentration. Figure 1(d) plots the surface roughness measured by a 3D optical profilometer, confirming that the etch surface roughness improves from ∼38 to ∼5 nm RMS as the doping concentration increases from 5 × 10¹⁷ to 5 × 10¹⁸cm⁻³.

To quantify the etch rate, optically scanned surface profiles are obtained. Figures 2(a)–2(c) show the surface profiles for the 1 × 10¹⁸cm⁻³ Si-doped GaN sample under three different etching solutions, with the insets depicting the corresponding SEM images of single pillars. Figures 2(d)–2(f) show the corresponding line scans from representative arrays of pillars. While the 1:1 etch condition produces micropillar arrays in the MOCVD GaN layer only, the 1:6 and 1:12 conditions reveal the underlying HVPE substrate, and the MOCVD/HVPE interfaces are highlighted by red arrows in the insets of Figs. 2(b) and 2(c) and dashed lines in Figs. 2(e) and 2(f). Figure 3(a) plots the extracted etch rates from the measured depth profiles as a function of etchant condition (1:1, 1:6, and 1:12 KOH:K₂S₂O₈) for all MOCVD samples (5 × 10¹⁷, 1 × 10¹⁸, and 5 × 10¹⁸ cm⁻³ Si-doped) and the standalone HVPE substrate. For the etching conditions (1:6 and 1:12 for 10 min) that etched through the MOCVD epilayer and HVPE substrate interface, the etch rate for the epilayer is extracted by subtracting the corresponding time used for the HVPE layer based on the separately calibrated HVPE etch rate. The etch rate increases with increasing oxidant molar ratio, which can be attributed to increased hole injection that enhances oxidation at the etchant–semiconductor interface. The etch rate of the HVPE sample is much higher than that of the MOCVD epilayers, while the etch rate of the epilayers increases with the doping concentration in general. To compare the MacEtch rate with that of RIE, Fig. 3(b) plots previously reported RIE (Refs. 34–37) and PEC (Refs. 27–30) etch rates of GaN samples with available doping information as indicated. Our work clearly demonstrates that the hν-MacEtch process has etch rates comparable with those commonly observed for RIE. This is of particular importance when comparing surface morphologies shown in other studies. For example, smooth sidewalls were reported with RIE of MOCVD GaN material at an etch rate of approximately 120 nm/min;³⁵ in contrast, etch rates ranging from 150 to 270 nm/min dependent on doping concentration are already achieved using hν-MacEtch. The highest etch rate demonstrated in this study approaching 780 nm/min for the standalone HVPE substrate with a porous shell is comparable with the visibly rough morphologies etched by RIE at an etch rate of 550 nm/min,³⁵ and that with etch rates up to 845.3 nm/min³⁷ being reported. GaN etch rates using PEC etching from the literature are also plotted for comparison. Note that most of the data points obtained from the literature do not have the doping level systematically varied or defect information and etched surface roughness specified, so direct benchmarking would not be strictly possible. Nonetheless, it is clear that hν-MacEtch, a remarkably simpler process, has already achieved etch rates comparable or exceeding those commonly observed for RIE or PEC.

It is well known that MacEtch relies on the localized generation of e⁻/h⁺ pairs via catalysis of the oxidant in the solution that causes the adjoining semiconductor material at the metal/semiconductor interface to oxidize and subsequently etch away. The metal catalyst then proceeds to sink and “graft” into the semiconductor (i.e., Si or GaAs) and the general cycle is continuously repeated for the duration of etching, in what we define as a “forward-etching” process. In the case of wide bandgap semiconductors including GaN, because of the low hole mobility, carriers necessary for MacEtch must be generated via external UV illumination with above bandgap photons. While the reduction of the oxidant is still catalyzed by the metal, the GaN directly beneath is shadowed, preventing light absorption, as illustrated in Figs. 4(a) and 4(b), and so the metal pattern serves as a catalytic mask. Once the e⁻/h⁺ pairs are generated outside of the metal covered area, the photogenerated e⁻ are collected by the metal catalyst (e⁻ sink) to reduce the oxidant, while the semiconductor serves effectively as the counter-electrode injected with h⁺. As a result, the catalyzed reduction of K₂S₂O₈ serves as a spatially localized e⁻ sink, exposed GaN is oxidized, KOH serves as a chemical etchant to remove the oxidized material, and Na₃PO₄ serves as a buffer to maintain pH for consistent etch behavior. As a result, increasing the doping concentration in GaN or the oxidant concentration in the etch solution corresponds to an increase in the local photocurrent, and, hence, to an increase in the etch rate, consistent with the observation in Fig. 3(a). Incidentally, the faster etch rate with increasing doping concentration in GaN also leads to a smoother surface, as observed in Fig. 1.

It should be noted that the etch mechanism consists of both a photocurrent driven MacEtch component and a pure chemically driven component due to the etching of GaN by KOH,³⁸ and the ratio of the two directly affects the surface morphology and topography. As highlighted in Fig. S1,⁴³ the chemically driven component of hν-MacEtch is particularly evident for the GaN pillars formed from the HVPE substrate exhibiting an ∼22% undercut of the Pt catalyst after 5 min of etching. In contrast, the MOCVD homoepitaxial GaN pillars have an undercut of ∼16% after 10 min of etching under the same solution, which is ∼2.8× smaller when the etch time is normalized, assuming a constant etch rate with time. This strongly suggests that the HVPE substrate is more susceptible to chemical attack than the MOCVD homoepitaxial GaN. In addition, the KOH chemical etch is known to be dependent on crystal orientation, resulting in the hexagonal crystal facets on the sidewalls of the formed pillars beneath the Pt catalyst [see, for example, Fig. 4(g)].

Similar to PEC etching, hν-MacEtch leads to the formation of whiskers in GaN [Figs. 4(b) and 4(e)], which are known to be associated with the vertically propagating dislocations.^39,40 This is because the dislocations act as carrier recombination centers, preventing their extraction to the electrodes. Hence, they are left unetched and removal commences only via chemical attack from KOH over time. Once the MOCVD homoepitaxial GaN is etched through, there is continued whisker formation in HVPE GaN. However, the overall morphology of the etched HVPE surface is dramatically rougher with more pronounced whiskers. Since this is true for a standalone HVPE GaN substrate independent of the MOCVD epitaxial layer (Fig. S2),⁴³ we believe that the etch morphology difference is associated with the innate material differences from the two different growth methods, including the type and distribution of impurities and vacancies, as well as extended defects, which are exacerbated by the hν-MacEtch mechanism. On the other hand, voidlike holes observed for the HVPE sample and further revealed by FIB processing, as shown in Fig. S3,⁴³ can be attributed to chemical etching of the discrete defects in HVPE GaN. This is clearly shown in both the composite structure and the standalone substrate, as indicated by Fig. S1.⁴³ The detailed correlation with the structural properties will be systematically characterized in future studies when a larger set of samples become available.

Interestingly, the interface between the MOCVD epi and the HVPE substrate shows a clear gap where the remaining MOCVD homoepitaxial GaN upper region is connected to the HVPE stem via a layer consisting of inverted nanopyramidlike (INP) structures, as illustrated in Fig. 4(c) and shown in Fig. 4(e). We attribute the INP structures to termination points of dislocations,⁴¹ and this is highlighted by the red outline in Fig. 4(g). The aggressive lateral etching at the epilayer/substrate interface is likely due to a surface layer with extremely high Si-doping (∼1 × 10¹⁹cm⁻³),⁴² which is susceptible to attack from both hν-MacEtch and KOH chemical etching. With continued etching into the HVPE substrate, continued undercut of the metal catalyst eventually causes it to lift-off GaN, as depicted in Fig. 4(d), while the INP connected interface further deteriorates from KOH etching, resulting in the structure shown in Fig. 4(f). As a result, the photocurrent driven component of hν-MacEtch reduces over time because of the reduced contact area in conjunction with a longer distance from the catalyst. The native chemically driven portion of the etch from KOH dominates and polishes away the whiskers. Simultaneously, porosity from defects in the GaN material is revealed and the existing porosity is widened, leaving behind a porous HVPE “stem” that is observed in some samples such as those shown in Fig. S1.⁴³ 

To evaluate the optical quality of the etched structures, Fig. 5(a) shows representative spatially resolved room temperature PL spectra on a semilog scale from 300 to 650 nm from three regions (pillar top, sidewall, and spacing in-between) on the 5 × 10¹⁷ cm⁻³ doped GaN sample etched under various KOH:K₂S₂O₈ molar ratios as indicated, measured after Pt catalyst removal in aqua regia (unless noted otherwise), along with those taken from the unetched sample. Figures 5(b) and 5(c) separately highlight the PL intensity comparison on a linear scale in the NBE and the yellow band (YB) emission regions, respectively, for the 1:1 KOH:K₂S₂O₈ etching condition where the etching does not expose the HVPE substrate. It can be seen that the NBE intensity is the highest in the spacing region, with the intensity from sidewalls (or the edges) and pillar tops incrementally lower, all of which are higher than that from the unetched sample. The PL map taken at the NBE peak (363 nm) shown in Fig. 5(d) confirms this trend by scanning across a 5 × 5 array of pillars. The PL enhancement can be attributed to the morphology differences previously described, where the interpillar region exhibits the highest roughness, which serves as antireflection texturing to enhance PL collection. The removal of surface states and possibly Ga vacancy complexes as a result of etching could also contribute to the slight enhancement of the PL intensity. On the other hand, Fig. 5(c) indicates that YB emission from the corresponding regions consistently shows a lower intensity than in the unetched sample overall, although the difference is almost negligible. These observations are in strong contrast to the commonly observed NBE PL reduction for RIE samples,^8,9 which indicates most importantly that hν-MacEtch does not damage the GaN surface. The same trend holds true for all GaN samples etched regardless of doping levels and etch conditions. As an example, representative PL spectra from different regions on the 5 × 10¹⁸ cm⁻³ doped sample are presented in Fig. S4 in the supplementary material.⁴³ 

In conclusion, we show that hν-MacEtch can produce large-area ordered arrays of highly vertical GaN micropillars with etch rate comparable or higher than RIE. We found that increasing the Si-doping concentration leads to increased etch rates and smoother etch surfaces for MOCVD homoepitaxial GaN, while HVPE GaN shows a dramatically different etch morphology and a faster etch rate. Remarkably, etched GaN structures are devoid of nonradiative damage found in plasma-based RIE. Compared with the plasma-free PEC methods that have been used widely for GaN, hν-MacEtch is simpler because of its open-circuit nature, an accelerated etch rate because of the metal catalytic effect, and a higher aspect ratio because of the localized carrier distribution, all while maintaining the inherent damage-free advantage of wet chemical etch. It can be envisioned that by tuning the carrier distribution locally with the metal catalyst pattern, local control of etch rate and aspect ratio can be achieved for various device applications requiring 3D architectures with damage-free surfaces. We believe that the same approach works for InGaN and AlGaN, as well as heterojunctions of III-nitrides, by tuning the etch chemistry and the excitation wavelength.

The authors would like to thank Hsien-Chih Huang for his help with FIB cross-sectional analysis. The work at the UIUC was partially supported by NSF ECCS No. 18-09946 and a grant from the ZJUI. The work at the NRL was supported by the Office of Naval Research (ONR).

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