Thermally tunable microring resonators based on germanium-on-insulator for mid-infrared spectrometer Open Access

In the past decades, the pursuit of the development of photonic integrated circuits (PICs) has propelled the advancement of on-chip technologies. PIC technology^1–3 has provided a paradigm shift by enabling the creation of compact and versatile devices. One of the most valuable platforms for supporting those has been silicon (Si) by exploiting Si-on-insulator (SOI) that enables data communication,⁴ including quantum technology.⁵ Particularly, photonic integrated spectrometers^6–9 with this PIC technology with Si play a key role across diverse applications ranging from environmental monitoring¹⁰ to medical diagnostics.¹¹ Unlike a traditional spectrometer, which has drawbacks with bulky components and high cost due to the mechanical parts, the PIC-based spectrometers leverage the principles of on-chip integration, enabling the consolidation of essential optical components including a laser¹² and detector^13,14 onto a single chip. For the on-chip spectrometer, typical structural configurations, such as arrayed waveguide gratings (AWGs),^9,15 Bragg-grating structures,¹⁶ and microring resonators (MRRs),¹⁷ have been adopted. Among them, MRRs can achieve a higher spectral resolution owing to the narrow linewidth of the resonant peaks in MRRs, enabling precise spectral analysis. In particular, the Fourier transform spectrometer with a heater (TiN)-integrated MRR presented a very narrow linewidth with a sub-nanometer (∼0.15 nm) scale at 1528 nm wavelength.¹⁸ Thus, the narrow linewidth offered by these PICs ultimately contributes to significantly reducing crosstalk in spectrometers while achieving high resolution.

With recent advances in mid-infrared (IR) sensing technology,^19–21 there is an increased emphasis on utilizing spectrometers based on PIC technology in the mid-IR region. The specific focus on the mid-IR wavelengths in on-chip spectrometer research is for providing valuable spectral information associated with molecular vibrations, enabling precise identification and analysis of optical absorption. Exploring the potential of the mid-IR application on the PIC platform has emerged as a crucial avenue for advancing spectroscopic sensing capabilities. The precision afforded by the PICs in manipulating mid-IR light on-chip has unveiled new possibilities for spectroscopic sensing.²² In addition, implementing the device in mid-IR spectrometers offers a notable advantage with reduced fabrication tolerance compared to the visible or near-IR bands, attributable to the wavelength scaling effect. Thus, providing the sensor with high performance is possible by manufacturing the device with precise dimensions. However, unlike the C-band operating devices, mid-IR spectrometers might have challenges in realizing narrow 3 dB linewidth at the resonance peak, if assuming the uniform propagation losses in a conventional strip-waveguide-based ring resonator in all wavelengths. Furthermore, when integrating a heater on the top cladding layer of a mid-IR spectrometer, it is crucial that the cladding layer remains lossless in the mid-IR range and that its thickness is sufficient to ensure effective mode confinement within the waveguide. To address these challenges in the mid-IR range, employing a platform with extremely low propagation loss appears to be the most promising approach.

In this regard, recent research focuses on the implementation of the germanium (Ge)-based platforms,^23–26 employing the advantages of both PIC technology and the unique properties of the Ge. This union paves the way for the development of on-chip spectrometers that are compact and efficient as well as tailored for applications requiring mid-IR spectral analysis. Ge, as one of the promising platform materials for the mid-IR band, has been showing strong motivation for the PICs with significant properties with broadband transparency and high refractive index in the mid-IR range. Those are well-suited for efficient light–matter interaction in the mid-IR range, making it an ideal candidate for the realization of mid-IR photonic devices. In addition, as a semiconductor, the Ge exhibits high compatibility with the CMOS process,²⁷ which enables mass production and large-scale integration such as Si. In terms of the spectrometer, Ge has a strong nonlinear coefficient with the Kerr index n₂ of 25.5 nm²/W²⁸ and a high thermo-optic (TO) coefficient of 4.269 × 10⁻⁴ K⁻¹,²⁹ whereas each value of Si was 3.29 nm² and 1.7 × 10⁻⁴ K⁻¹, respectively. Due to Ge having a greater TO coefficient, various applications, such as band-stop filter³⁰ and phase shifter^31,32 on Ge-based MRRs,^22,33,34 have been reported. Especially, the Fourier-transform spectrometer developed on the Ge-rich SiGe platform³⁵ represents a significant milestone as the first on-chip mid-infrared spectrometer. Despite initial challenges regarding linewidth and heating efficiency, this innovative platform shows promising potential. Operating within the mid-infrared band, it offers a linewidth ranging from 5 to 15 cm⁻¹ and an average heater efficiency of 2 K/W at 7.7 µm, showing progress in thermal tuning on the SiGe waveguide platform. Hence, the very efficient TO effect of our Ge-OI platform could contribute to a broad tuning range of the mid-IR band for the spectrometer. Thanks to the property with low propagation loss in the Ge-OI waveguide³⁶ with a Y₂O₃ insulator, which is highly transparent in the mid-IR range (extinction coefficient, κ ∼ 10⁻⁵ over 3–13 µm wavelength), over 120k of intrinsic Q factor (Q_int) for the ring resonator and 176k of Q_int for the racetrack resonator have been demonstrated .³⁷ Furthermore, the Y₂O₃ (n_Y2O3 ∼ 1.85) layer could provide a high-index contrast to Ge (n_Ge ∼ 4.02), leading to great optical confinement in Ge.

Herein, we demonstrated a thermally tunable Ge-OI MRR for the on-chip spectrometer in the mid-IR range. Through the optical and thermal simulations, the transmission curves of the Ge-OI MRRs with different refractive indices according to the TO coefficient and temperature (T) change of the Ge were characterized. Following that, we integrated a ring-shaped heater on the top of the Ge-OI MRR with an intermediate layer of the top cladding layer of the Y₂O₃. Here, the Ge-OI wafer was processed by direct wafer bonding. Before the optical measurement for verifying the resonant shift by Ge’s effective index change, practical T change in the Ge-OI MRR was figured out by using thermo-reflectance microscopy, while applying bias onto the device. Consequently, a high Q factor of the Ge-OI MRR despite the optical loss from the metal heater was obtained and the resonant-shifted transmission curves at 4.2 µm wavelength due to the TO effect of the Ge were characterized. Thus, our thermally tunable Ge-OI MRR would be key in future spectrometer sensors, such as a multi-channel gas sensor. It will selectively filter mid-IR wavelengths for on-chip spectrometer sensors, integrating light sources, sensing elements, and detectors. For the spectrometer in the mid-IR range, our thermally tunable Ge-OI MRR could be one of the best candidates for providing a high resolution.

Figure 1(a) shows the proposed thermally tunable ring filter, which was fabricated by forming a ring-shaped Au heater on the Y₂O₃ top cladding layer of the Ge-OI MRR. The geometry parameters, including the thickness of Ge (H_Ge), the width of the ring (W_ring), the width of the bus waveguide (W_bus), and the ring-to-bus gap (G), were optimized to ensure critical coupling conditions and achieve high Q factor and extinction ratio (ER) at the 4.2 µm wavelength band.³⁷ The ring radius (R) was set to 63.5 µm to resonate at 4.2 µm wavelength. The width (W_ht) and thickness (H_ht) of the Au heater on the top cladding Y₂O₃ layer (thickness, H_tc: 1.2 µm) were 8 and 0.1 µm, respectively. The effective index (n_eff) and group index (n_g) of the Au heater integrated Ge-OI MRR were calculated to be 2.87 and 7.06 (supplementary material A), respectively, at 4.2 µm wavelength and 300 K temperature using a finite difference eigenmode solver (Ansys Lumerical) with our database of the dispersion curve of the refractive indices for the Y₂O₃.^36, Figure 1(b) shows the electrical field distribution of the Ge-OI MRR operating in the TE mode. To efficiently transfer heat to the Ge waveguide while maintaining low optical losses, a Y₂O₃ layer with a thickness of 1.2 µm was chosen, resulting in a calculated propagation loss of 3.08 dB/cm, considering all optical losses including free-carrier absorption (FCA) in Ge.³⁸ Although the mode overlap in the Au heater was small at 3.5 ppm, the presence of the heater attributed to the propagation loss of 2.06 dB/cm because of the huge extinction coefficient value of Au in the mid-IR region from our loss analysis. Assuming no loss component in the device except for the Au loss, the theoretical Q factor from this loss becomes 79k at 4.2 µm. The radiation and scattering losses, which are very small, were considered negligible. The loss due to FCA in the Ge was 0.69 dB/cm, and the absorption loss from the Y₂O₃ layer was 0.32 dB/cm (7.4% loss). In contrast, the propagation loss of the waveguide without the top cladding layer was measured as 0.92 dB/cm,³⁷ where the Y₂O₃ loss was around 0.1 dB/cm (2.3% loss). Therefore, the observed increase in loss to 3.08 dB/cm (a difference of 2.16 dB/cm) is attributed to the material losses associated with the Au and the Y₂O₃ cladding. Notably, the FCA loss slightly decreased by 0.09 dB/cm in the thermally tunable Ge-OI MRR due to reduced mode confinement in the waveguide. The calculated propagation loss, including bending loss, of the heater-integrated Ge-OI MRR decreased and reached saturation at ∼3 dB/cm for ring radii exceeding 15 µm. Moreover, to ensure the prevention of heater breakdown under the applied voltage to the electrode, a width of 8 µm was deemed appropriate for obtaining stable IV characteristics.

Utilizing the temperature-dependent Sellmeier equation and calibrated coefficients,²⁸ we determined the temperature coefficient (dn_Ge/dT) of 4.27 × 10⁻⁴ K⁻¹ for Ge at 4.2 µm wavelength within the temperature range of 300–360 K, as shown in Fig. 2(a). In this analysis, we disregarded the thermal optic coefficient of the Y₂O₃ due to its significantly lower value (∼10⁻⁶ K⁻¹) compared to that of Ge.³⁹  Figure 2(b) shows the temperature-dependent effective indices of the cross-sectional waveguide of our thermally tunable Ge-OI MRR obtained through eigenmode solver simulations. Consequently, the increased effective refractive indices of Ge resulting from temperature elevation contributed to the resonance shift toward longer wavelengths. This correlation was validated through simulated transmission curves (3D-FDTD; Ansys Lumerical) of the Ge-OI MRRs to determine the precise resonant characteristics incorporating temperature-dependent indices in the mid-IR band, as shown in Fig. 2(c). Consequently, the wavelength shift due to temperature variation was determined to be 0.61 nm/K, with the free spectral range (FSR) of the spectra ∼5 nm.

Given the geometrical parameters of our Ge-OI MRR [Fig. 3(a)], we analyzed the temperature distribution within the Ge ring through heat transfer simulations (Ansys Lumerical), applying power (P) to the heater. Thermal properties of all materials, including density, specific heat, and thermal conductivity, were meticulously considered based on the parameters from Refs. 40 and 41. Especially, the thermal conductivity of Y₂O₃ was regarded as 12.2 W/(m K),⁴² acknowledging its low crystallinity attributed to the sputtering deposition process. With the range of 20–200 mW of P, as shown in Fig. 3(b), 2D T mapping plots of the MRRs at the center of the Ge (z = 0.25 µm) linearly increased. The overlapped black solid lines in the 2D-contour graphs indicated the Ge waveguides. Effective thermal transfer from the Au heater to the Ge ring structures was observed, with heat being localized adjacent to the Ge, maintaining the thermal profile dictated by the shape of the Au heater. However, the observed heat conduction toward the -z direction can be attributed to the relatively lower thermal conductivity of Y₂O₃, as shown in Fig. 3(d). The heating efficiency, which is the temperature variation with applied power change (∆T/∆P), was ∼96 K/W, as shown in Fig. 3(c). Accordingly, the resonance change (∆λ/∆P) considering ∆λ/∆T [Fig. 2(c)] of 0.61 nm/K was calculated to be 58.56 nm/W.

From optical and thermal analyses, we fabricated the thermally tunable Ge-OI MRR after implementing the Ge-OI wafer. Figure 4(a) shows the fabrication procedure for the device. First, the Y₂O₃ layer with a thickness of 1.5 µm was sputtered on the prepared Ge epitaxial wafer and Si substrate. The Ge epitaxial wafer contained the growth layers with the core layer (Ge; 500 nm), selective etch layer (Si_0.5Ge_0.5; 10 nm), and buffer layer (Ge; 900 nm) to relax the lattice mismatch. To bond these two substrates, the Y₂O₃ films on both substrates were chemically polished to reach the very smooth roughness of the surface. After, the chemical cleaning process and O₂ plasma treatment were applied on the surfaces, followed by bonding together for over 5 h at 50 °C in the vacuum chamber. Next, we mechanically thinned down the Si layer of the Ge epi. substrate and chemically removed it by using a heated tetramethylammonium hydroxide (TMAH) solution (∼90 °C). The Ge buffer layer and Si_0.5Ge_0.5 were etched by SC-1 solution (H₂O + H₂O₂ + NH₄OH) and TMAH, respectively. We defined the pattern of the ring resonators and grating couplers on the Ge-OI wafer with the process of electron-beam lithography. Then, the Ge was etched by ICP-RIE in C₄F₈ and SF₆ [Fig. 4(b); Top]. By using a shadow mask, we deposited a 1.2 μm thick Y₂O₃ on the part of the MRRs on the fabricated Ge-OI waveguides with the RF sputter to form the Au/Cr metal heater on the ring region except the grating couplers [Fig. 4(b); Middle]. Subsequently, photolithography with a negative photoresist (DNR) was carried out to define the heater area, followed by Cr/Au metal deposition (10/100 nm) with the E-beam evaporator. Here, we slightly coated Cr under the Au layer to enhance adhesion. Finally, the ring-shaped Au heater was formed by using the lift-off process [Fig. 4(b); bottom].

Figure 5(a) shows the magnified optical microscope image for the heater integrated on the Ge-OI MRR. The 8 μm wide Au heater was well processed on the Ge-OI MRR. To mitigate the input power by changing the applied voltage (V), we conducted current–voltage (IV) characteristic measurements within the ohmic regime spanning from −3 to 3 V, as shown in Fig. 5(b). The inset table presents the corresponding power values for applied biases of 1, 2, and 3 V, corresponding to 36.69, 139.62, and 313.96 mW, respectively. In addition, to verify heat generation on the Au heater, we performed analysis using thermo-reflectance microscopy (TRM), as shown in Figs. 5(c) and 5(d), employing a TRM measurement setup⁴³ (Nanoscope Systems, TRM250). TRM is a non-destructive technique that employs 2-D temperature mapping by utilizing T variations in surface reflectivity (R) induced by ∆T to generate high-resolution two-dimensional thermal images with sub-micron scales. The fundamental principle underlying TRM lies in its ability to detect variations in surface reflectivity caused by temperature changes. The thermo-reflectance coefficient (κ) calibration involved determining the linear relationship between the T-induced change in surface reflectivity (ΔR/R) and the corresponding ΔT, expressed as ΔR/R = κΔT. Considering the challenge of recording temperature changes due to no reflectivity change at 0 V, we applied a near-zero voltage to the electrode, as shown in Fig. 5(c). To enhance the detection of temperature variations on the Ge waveguide side, we thoroughly aligned the microscope’s focus with the Ge-OI waveguide plane. Under a bias of 2.3 V (P = 181.7 mW), the TRM image shown in Fig. 5(d) revealed distinctive thermal profiles. Negligible heat generation was observed near the bus waveguide region, while a significant thermal response (∆T = 17 K) was specifically identified in the ring structure, demonstrating excellent agreement with the heat simulation at P = 180 mW [Fig. 3(b)]. Furthermore, pronounced heat generation (∆T = 36 K) occurred in the narrowed region from the metal pad to the ring heater. These findings underscore the localized heating effect and efficient energy transfer within the ring structure. The absence of heat generation in the bus waveguide region indicates effective thermal isolation and confinement within the ring structure, showing the thermal functionality and efficiency of the implemented metal heater. Given the temperature-power linearity within the applied bias range of 0–3 V, the experimental heating efficiency was determined to be 93.6 K/W, demonstrating good agreement with the predicted value (96 K/W) from the simulation results.

Figure 6(a) shows a high-resolution measurement of the transmission at 4.2 µm wavelength for extracting the Q factor of the heater-integrated Ge-OI MRR under a 0 V applied bias. The characterization of the Ge-OI MRR was conducted using our mid-IR fiber (InF₃)-based waveguide measurement setup integrated with an electrical characterization system (Keithley 2400). The mid-IR light was sourced from an external cavity quantum cascade laser (EC-QCL; MIRcat-QT, Daylight Solutions) and detected by an MCT detector (PVI-4TE, VIGO Photonics). Here, the linewidth of EC-QCL was under 100 MHz, and the scanning step of wavelength was 0.05 nm with our timing trigger function to resolve the data by tuning a specific time constant and recording time. We used the same system in our previous work³⁷ for the optical part of the measurement setup. Since our Ge-OI grating coupler was designed and fabricated to have maximum efficiency at 4.2 µm wavelength in the TE mode, the mid-IR light was coupled to the waveguide with the TE polarization through recording the highest transmission. The transmission spectrum (red square) was fitted using a Lorentzian curve with a nonlinear least-square (NLSQ) method (black dotted line). The loaded Q factor (Q_load) was determined to be 21 084, corresponding to an intrinsic Q factor (Q_int) of 31 338, while the extinction ratio (ER) showed 9.23 dB, as shown in the inset of Fig. 6(a). The presence of the Au heater on top of the Y₂O₃ cladding region led to a degradation in the Q factor due to increased waveguide loss compared to the previous results of the Ge-OI MRR (Q_int ∼ 120k).³⁷ In addition, the coupling strength between the ring and bus waveguide strengthened as the gap was filled with Y₂O₃, resulting in an over-coupling condition with a reduced ER compared to the previous results, attributed to the decrease in coupling loss between the ring and bus waveguide. Including that reason, the pair of the metal pads that was formed on top of the cladding layer degraded the coupling efficiency. As a result of this coupling loss effect, the measured Q factor was reduced by approximately half (Q_int = 31 338) compared to the theoretical Q factor of 60 534 obtained from the equation considering only waveguide loss. Nevertheless, the device exhibited sub-nanometer linewidth with 0.199 nm despite its optical loss caused by the metal line. Transmission curves of the thermally tunable Ge-OI MRR, measured at different biases (1, 2, and 3 V), are shown in Fig. 6(b), with each dataset normalized to its maximum value on the spectrum. A clear spectral shift is evident with increasing applied voltage, corresponding to increased refractive indices due to the thermo-optic effect of Ge. Figure 6(c) magnifies the spectra in a linear scale to confirm the resonance shift for three different biases, revealing shifts (∆λ) of 1.94, 4.06, and 6.36 nm for applied voltages of 1, 2, and 3 V, respectively. These shifts arise from the heating induced by joule heating on the Ge-OI MRR due to the applied power values obtained from the IV curve. This results in a linear fitted tunability of 33.7 nm/W, which corresponds to 0.36 nm/K. The observed discrepancy in tunability between experimental and calculated values (0.61 nm/K) may arise from variations, such as the limitations of the Sellmeier model analyzed within room temperature and adjustments in the TO coefficient for sputtered Y₂O₃. Due to the relatively large size of our MRR’s ring radius (63.5 µm), significant tunability may not be achieved. However, reducing the ring radius allows for a smaller heater size, thereby reducing the applied bias required for thermal tuning-induced resonance shifts at the same temperature, provided that the bending loss degradation does not compromise the loss characteristics. Figure 6(d) shows the normalized transmission at a wavelength of 4.2 µm during thermal tuning from 0 to 3 V. Since the resonance peak shifts to higher wavelengths when applying the positive bias, peaks one order lower than resonance around the 4.2 µm wavelength were observed to shift. The behavior of this thermal tuning, based on measured transmission, aligns well with the absolute value of a sinusoidal function with a period of 1 V up to 2.5 V, but the subsequent emergence of the next peak in the spectrum occurs after applying 2.5 V. This effective thermal tuning at a single wavelength of our MRRs suggests the device's responsiveness to temperature changes, facilitating precise wavelength adjustments and reliable spectrometry. Thus, our thermally adjustable Ge-OI MRRs provides precise control and manipulation of the optical resonance for the mid-IR spectrometer, as demonstrated by the observed resonance shifts per unit power, opening up novel applications, such as phase modulation and optical sensing.

To enhance the performance of the mid-IR spectrometer based on our proposed Ge-OI MRR with thermal tuning, achieving a high-resolution and low-crosstalk spectrometer necessitates improvements in the Q factor. Addressing this challenge can be effectively tackled by implementing a thicker Y₂O₃ layer, as shown in Fig. 7(a). The calculated propagation loss of the heater-integrated waveguide significantly decreases with increasing thickness of H_tc. Consequently, the expected Q_int of the device (if assuming the ring resonator operates at critical coupling condition), with H_tc set at 1.5 µm, is projected to reach 190k based on theoretical predictions, as shown in Fig. 7(b). However, a critical concern arises regarding the potential reduction in heat conduction from the Au heater to the Ge. From this point, our 1.2 μm thick top cladding layer in the heater-integrated Ge-OI MRR was appropriate to achieve a good quality factor and thermal tunability. Fortunately, preliminary investigations reveal that the decrease in heat transfer remains manageable, particularly under P of 150 mW, as shown in Fig. 7(c). Hence, we believe that a thermally tunable spectrometer with a significantly narrower linewidth compared to the existing 0.199 nm could be realized, thereby enhancing its performance.

While there have been a few reports on mid-IR operating MRRs capable of thermal tuning based on the Ge platform, we compared this result to the previous studies,^{30,33,35,44,45} which provided data with the tunability with respect to temperature change and the linewidth for the spectrometer technology, as shown in Fig. 8. Demonstrating high tunability and excellent linewidth solely with a simple ring structure in the mid-IR range, we anticipate further enhancements through increased coupling strength or additional FSR tuning with structures such as a racetrack or Vernier filter. For instance, in Ref. 33, they achieved a tenfold increase in tunability (blue circle; Fig. 8) with a Vernier amplification factor of around ten compared to a single MRR. Moreover, in Ref. 44, the tunable Vernier filter with the racetrack resonator based on the Ge-on-SOI platform operating in the 5 µm wavelength showed the Q_load with around 20 000 and tunability of 0.5 nm/K. The Q factor is close value to our work, and it has the greater tunability. In this regard, the Vernier technique is promising to achieve excellent linewidth of resonator in terms of the mid-IR spectrometer. Thus, assuming a fivefold increase in tunability with the implementation of a Vernier filter, we could potentially achieve a tunability level of around 1.8 nm/K with a ring size of 63.5 µm. In addition, a recent development in mid-IR sensing technologies employing PICs has been prompted by the integration of bolometers onto the Ge-OI platform for mid-IR detection.¹⁴ This development reveals a promising future for mid-IR PIC technologies, paving the way for compact on-chip spectrometer sensors based on our thermally tunable Ge-OI MRRs.

Therefore, we present an experimental demonstration of a thermally tunable mid-IR spectrometer utilizing our Ge-OI microring resonators (MRRs), achieved by integrating a simple Au heater atop the Y₂O₃ cladding layer. Through comprehensive simulations of transmission and thermal responses of the MRRs, we successfully fabricated the heater-integrated Ge-OI MRR using the direct wafer bonding process. Subsequently, to validate the practical temperature increase in the Au heater, we employed our thermo-reflectance microscopy (TRM) measurement setup to confirm the temperature change of the Ge-OI MRR. Despite the MRR exhibiting slightly high propagation loss (3.08 dB/cm) due to its proximity to the metal, the Q factor of the Ge-OI MRR remained high, with Q_load of 21k at a wavelength of 4.2 µm wavelength, coupled with a tunability of 33.7 nm/W (0.36 nm/K). From further development by optimizing the structure, significantly enhanced performance of the mid-IR spectrometer could be achieved in the aspects of spectral resolution and heating efficiency. We contend that this thermal tuning technique based on the Ge-OI platform holds significant promise as a leading candidate for mid-IR optical spectrometers.

See the supplementary material for calculating effective indices and group indices of the heater-integrated Ge-OI.

This work was supported by the National Research Foundation of Korea (NRF) (Grant No. 2023R1A2C2002777, RS-2024-00407767), the KIST Institutional Program (Atmospheric Environment Research Program, Project No. 2E32402), and BK FOUR.

The authors have no conflicts to disclose.

J. Lim: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). J. Shim: Formal analysis (supporting); Methodology (supporting); Software (supporting); Writing – review & editing (equal). I. Kim: Data curation (supporting); Formal analysis (supporting); Methodology (supporting); Software (supporting). S. K. Kim: Formal analysis (supporting); Investigation (supporting); Methodology (supporting). D.-M. Geum: Investigation (supporting); Methodology (supporting); Writing – review & editing (equal). S. Kim: Conceptualization (equal); Funding acquisition (lead); Supervision (lead); Writing – original draft (supporting); Writing – review & editing (lead).

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