Fast and sensitive terahertz detection with a current-driven epitaxial-graphene asymmetric dual-grating-gate field-effect transistor structure

Terahertz (THz) electromagnetic waves have great potential to be utilized for applications in large-capacity, ultrahigh-speed wireless communication technologies in 6G toward 7G.^1,2 To realize such high-speed communication systems, highly sensitive, fast-response room-temperature detectors operating in the THz and sub-THz ranges are the key elements.^3,4 However, there are many performance limitations for the currently available THz detectors.^5,6 Two-dimensional (2D) plasmons have attracted increasing attention as a promising mechanism for highly sensitive, fast-response THz detection.^6–16 In particular, graphene Dirac plasmons (GDPs)^17–19 are believed to be one of the most promising physical principles for breaking through the technological limit for room-temperature, fast, sensitive THz detection capable of 100 Gbit/s class high-data-rate coding of THz- and sub-THz radiation incidence in next-generation 6G- and 7G-class wireless communications systems.²⁰ Graphene has also been used for fast photothermoelectric (PTE) THz detection^21–25 due to its superior carrier transport and phononic properties.²⁶ In this paper, we design and fabricate an epitaxial-graphene-channel field-effect transistor (EG-FET) featured by the authors’ original asymmetric dual-grating-gate (ADGG) structure²⁷ working as a current-driven terahertz detector with applied nonzero drain–source bias voltages and experimentally demonstrate a high responsivity of 0.3 mA/W (equivalently 12 mV/W under the 50 Ω-loaded condition and 84 mV/W under the high (∼1 MΩ) loaded impedance condition) to 0.95 THz radiation incidence at room temperature. The ADGG and drain–source bias dependencies of the measured photoresponse show a clear transition between plasmonic detection and PTE detection while preserving the fast response speed. The experiments also demonstrate fast temporal photoresponses for plasmonic and PTE detection on the order of 10 ps.

A bird’s-eye view of the ADGG-EG-FET structure is schematically shown in Fig. 1. The channel consists of a few layers of epitaxial graphene that were thermally decomposed from a C-face SiC substrate.^28–32 After device mesa isolation, which defines an active detector area of 20 × 20 µm², source and drain ohmic metallic contacts [Ti (10 nm)/Pd (20 nm)/Au (70 nm)] were formed using standard contact lithography, electron-beam evaporation, and lift-off processes. A gate stack was formed with a 40-nm-thick SiN dielectric layer deposited on the graphene-channel layer using plasma-enhanced chemical vapor deposition (PE-CVD).^32,33 A gate metal electrode [Ti (10 nm)/Pt (20 nm)/Au (70 nm)] was formed in the ADGG structure by using electron-beam lithography, an electron-beam evaporator, and a standard lift-off process.^32,33 A scanning electron microscope (SEM) image of the fabricated ADGG-EG-FET is depicted in Fig. 2(a). The ADGG electrodes, consisting of two interdigitated grating-shaped metals with grating finger widths of 550 nm (L_g1) and 850 nm (L_g2) were laid out with asymmetric distances of 450 nm (d₁) and 750 nm (d₂) to the left-side and right-side adjacent fingers, respectively. The source and drain electrodes were formed on top of the graphene channel with planar³⁴ and edge³⁵ ohmic contacts. The crystallinity of the graphene layer was confirmed by Raman spectroscopy, as shown in Fig. 2(b). The G and G′ bands at 1590 and 2700 cm⁻¹ were clearly identified in the Raman spectra; in contrast, the defect-oriented D band at 1350 cm⁻¹ was as weak as the background noise floor, indicating the favorable high crystallinity of the graphene layer.³⁶ The ratio between the intensities of the G and G′ peaks was ∼3.5, indicating that the samples are few-layer graphene within the three layers.³⁷ The surface morphology of the graphene layer was characterized by atomic force microscopy (AFM), as shown in Fig. 2(c). Crystal domains with a size of several micrometers were identified in the AFM image.³⁸ 

When THz radiation is incident on the surface of the device, the ADGG electrodes work as a broadband antenna that can efficiently convert the incident THz photons into GDPs.^27,39 When one grating gate of the ADGG electrodes, Gate 1 (G1), is electrically biased at a high voltage, whereas another gate, Gate 2 (G2), is biased at the Dirac voltage to deplete the carriers, the channel underneath the high-biased G1 gate finger becomes a plasmonic cavity, working as a plasmonic detector producing a rectified direct current (DC) photocurrent due to the hydrodynamic nonlinearity of the GDPs.^15,16,27 The depleted channel underneath the low-biased G2 gate finger becomes a highly resistive load, working as a transducer to produce a photovoltage from the photocurrent.^11,16 Due to the periodic arrangement of such a unit pair of the plasmonic cavity and the resistive load in the “ADGG” structure, the photovoltage generated in each unit pair accumulates in a cascading manner, resulting in highly sensitive THz detection. It is noted that the difference between d₁ (=450 nm) and d₂ (=750 nm) or the asymmetricity between them (d₁/d₂ ≠ 1) is the key to unbalancing the boundary conditions at the left side and right side of the plasmonic cavity so that the plasmonic displacement current flowing from source/drain to drain/source becomes unbalanced, resulting in rectified DC photovoltaic output at the drain terminal. The length of the high-biased gate (L_g1 = 550 nm in the above context) determines the plasmonic resonant mode frequency, whereas the length of the low-biased gate (L_g2 = 850 nm in the above context) determines the load resistance value.

First, we measured the electrical DC and voltage characteristics of the as-fabricated EG-FET using a semiconductor parametric analyzer. The DC drain–source current vs the gate bias V_g1 scanning from/to 0 V to/from −20 V was measured under the condition of V_g2 = 0 V and V_ds = 0.1 V. As shown in Fig. 3, clear ambipolar characteristics near the Dirac voltage (V_Dirac, defined as the charge neutrality voltage point) were observed under a negative gate voltage. DC measurements in the p-type high-current operation region with negative gate voltages larger than V_Dirac could not be conducted due to the gate breakdown limitation. Such a wide shift in V_Dirac is considered to be due to the relatively high unintentional n-type doping of the graphene channel that occurred during the SiN insulator deposition process using PE-CVD.³² Furthermore, the measured hysteresis was weak; the difference in current between the forward and backward bias applications was merely several μAs. It was confirmed that the heterointerfaces of the graphene-SiN-gate stack suppressed undesired defect- and/or strain-induced trap centers, resulting in almost no hysteresis.

Next, we measured the temporal response of the photovoltage output from the drain electrode in response to pulsed quasi continuous-wave (CW) radiation incidence centered at 0.95 THz at room temperature. The THz detection measurement was conducted and implemented with an injection-seeded THz parametric generator (is-TPG)⁴⁰ as the THz radiation incident source (Fig. 4). The is-TPG generated pulsed-CW THz radiation with an envelope pulse width of 155 ps and a repetition rate of 200 Hz.⁴¹ The envelope pulse width of 155 ps was identified by using an optically upconverted cross correlation method with a subnanosecond Nd:YAG infrared pump pulse whose wavelength was centered at 1064 nm, as described in Ref. 41. The THz waves that were output from the is-TPG traveling in free space were focused by a Tsurupica lens with a focal distance of 100 mm and directed via an indium-tin oxide (ITO) mirror to the sample surface placed at the focal point. The radiation incidence energy was 137 nJ/envelope (peak power of ∼911 W). A set of radio frequency (RF) probes were contacted to the ADGG-EG-FET electrode pads to apply the bias voltages (the drain-to-source bias V_ds and two ADGG biases V_g1 and V_g2). To observe the temporal photoresponse waveform without distortions caused by the multireflection between the device output and the far end of the measurement equipment, we used a 50 Ω-impedance measurement setup consisting of a 50 Ω-input-impedance 22 dB-gain wideband preamplifier, a 50 Ω-input-impedance 33 GHz-bandwidth digitizing oscilloscope, and a 1m-long 50Ω-coaxial-cabled transmission line to connect the device output terminal and the oscilloscope. Compared to the high-impedance measurement setup that is frequently utilized for static DC-voltage photoresponse measurements, the measured photovoltage under the 50Ω-loaded condition becomes small by a factor of the voltage divider ratio between the internal channel resistance R_ch (∼300 Ω) and the load resistance R_L (=50 Ω) given by R_L/(R_ch + R_L) ∼ 0.14 in this experiment.

First, we conducted experiments at a sufficiently doped voltage of +15 V for V_g1 and at the Dirac voltage (=V_Dirac, the charge neutrality point) for V_g2 to ensure that the channel regions underneath G1 (G2) were sufficiently doped (depleted); thus, in theory, only the plasmonic rectification should be responsible for THz detection. In addition, V_ds was biased at 1 V to drive the current in the channel. As shown under the condition V_g2 = −15.5 V in Fig. 5, we confirmed a clear photovoltaic response at room temperature, indicating that the ADGG-EG-FET works properly as a plasmonic THz detector. The observed tail-free pulse width full-width at half-maximum (FWHM) value of 199 ps was slightly wider than the envelope width of the is-TPG radiation incidence, which was characterized to be 155 ps. The discrepancy between them might be due to several systematic factors caused by the different routes of pulse-width characterization. To the best of the authors’ knowledge, this is the first experimental demonstration of a fast temporal photoresponse of a graphene-channel FET THz detector at room temperature.

Second, we measured the photoresponse by increasing V_g2 from −15.5 V (=V_Dirac) to +15 V (=)V_g1) under fixed V_g1 (=+15.5 V) and V_ds (=+1.0 V) conditions. As shown in Fig. 5, the measured temporal photovoltage output increased with increasing V_g2 and corresponding electron densities underneath G2 while preserving the high-speed response. The peak values of the photovoltage vs V_g2 are plotted in Fig. 6(a). With increasing V_g2, the photoresponse increased and started to saturate at ∼−5 V. We claim that the observed increase in the photoresponsivity is due to a new type of unipolar PTE effect assisted by electrostatic carrier drift/diffusion, which will be discussed in the section titled Discussion.

The gate bias conditions for G1 and G2, which correspond to the plasmonic rectification and the PTE effect, are shown in Figs. 6(b) and 6(c), respectively. To confirm the behavior of photoelectrons due to the PTE process, we fixed V_g1 = V_g2 = 0 V, which ensured that the graphene-channel area was entirely sufficiently doped, and then increased V_ds from 0 to +1.5 V. The temporal photoresponse waveforms under typical nonzero V_ds bias voltage conditions were measured, as plotted in Fig. 7(a). The temporal photoresponse preserved its waveform with an FWHM value of 199 ps, independent of the applied nonzero V_ds bias voltages.

The peak values of the temporal photoresponse under the 50-Ω-loaded condition shown in Fig. 7(a) are plotted in Fig. 7(b) as a function of V_ds. The output photovoltage increased linearly with increasing V_ds. More importantly, when V_ds = 0 V, preventing the PTE rectification operation, no photoresponse was observed. This is clear evidence that the photoresponse observed in Figs. 7(a) and 7(b) under the fully doped conditions resulted from the PTE rectification effect.

In addition to the mechanism of THz detection by the GDPs, the experimental results suggest that the ADGG-EG-FET is also able to work as a current-driven PTE THz detector^21–25 due to the photo-Seebeck effect. When V_g2 is biased near the Dirac point, the carrier density in the channel is not sufficient to drive the PTE effect; plasmon rectification dominates the detection, which is shown in the blue part of Fig. 6(a). With increasing V_g2, the density of electrons under V_g2 increases toward the highly doped level under V_g1. This makes generating a photovoltaic response through plasmonic rectification difficult. Therefore, one needs to consider different mechanisms to interpret the observed result of increasing photovoltage with increasing V_g2, as shown in the orange part of Fig. 6(a).

When the drain terminal is DC-biased, the electric potential forms a slope along the channel, resulting in asymmetric thermodiffusion of photogenerated hot electrons under THz radiation incidence due to the photo-Seebeck effect along the channel with the help of field-induced electrostatic drift/diffusion; the hot electrons photoexcited by THz electromagnetic radiation diffuse and become biased in the direction of the potential slope. Such a PTE detection mechanism can well support the photoresponse vs V_g2 tendency in our observation when all gate electrodes are zero biased. In this regard, this mechanism might be said to be electrostatic-drift/diffusion-assisted PTE detection. When the drain electrode is zero biased, in contrast, thermodiffusion occurs isotropically, and there is no specific fraction of the diffusion direction going either to the source or to the drain, resulting in no photoresponse. It is worth mentioning that one merit of PTE detection is the possibility of conducting zero gate-bias operations. If the THz radiation spot size is sufficiently small with respect to the channel area to realize local THz irradiation, photothermodiffusion is dominant in either direction toward the source or drain electrode, depending on which electrode is closer to the spot. In this case, a nonzero photoresponse can be obtained. Our experiments fall into the former case; the spot covers the entire channel region, so no photoresponse can be obtained under zero drain-bias conditions.

We also consider whether the plasmonic rectification mechanism could be responsible for the zero gate-bias detection in the EG-FET with the Dirac voltage deeply shifted to the negative region. Since the spatial carrier density distribution along the channel under zero gate-bias conditions is rather monotonic and less periodically modulated, the photoresponse is saturated at V_g2 ∼ 0 V and beyond [see Fig. 6(a)]. Thus, plasmonic detection cannot be dominated. As a consequence, as long as the drain is nonzero biased, the photovoltaic response is associated with the PTE rectification mechanism. This is not similar to the standard PTE rectification in bipolar p–n junction diode structures^21–25 with metals of the anode and cathode electrodes with different work functions in which both electrons and holes contribute to the rectification function. Our current-driven ADGG-EG-FET contributes only unipolar carriers of hot electrons or hot holes excited by THz radiation incidence. Therefore, the observed THz radiation rectification mechanism is regarded as a new type of unipolar PTE detection assisted by field-induced electrostatic drift/diffusion. The results shown in Fig. 6 suggest that plasmonic rectification and/or PTE rectification take place in the ADGG-EG-FET under THz radiation incidence depending on the ADGG bias conditions, and that these two effects coexist under a wide range of V_g1 and V_g2 conditions. The theoretical estimation of the contributions of these two mechanisms to photocurrent generation revealed that they are on the same order of magnitude (see Sec. II of the supplementary material) and supported the experimental results.

In general, there can be additional contributions from ratchet mechanisms^42–46 in graphene with spatially modulated carrier density other than the plasmonic ratchet mechanism⁴⁷ included in our model. The reported ratchet mechanisms work even under the zero drain-bias condition without dc channel current flow, whereas no photoresponse was observed with our ADGG-EG-FET detector in the zero drain-bias condition, as shown in Fig. 7(b). As those prior works on ratchet mechanisms^42–46 clearly suggested that the geometrically asymmetric dual-grating-gate layout breaks the symmetry of the ratchet photocurrent flow, there might be contributions from ratchet mechanisms in our case. However, since we measured the detection photoresponse at the sensitivity level of plasmonic detection under periodic charge density modulations as the reference and observed an increase in the photoresponse with increasing V_g2, as the charge density increased from the charge neutrality point in the initially depleted region up to the entirely fully doped condition, the level of detection responsivity due to the ratchet mechanisms was considered to be too small to be identified in our measured range of the scale. Further investigation of the ratchet mechanisms in our detector will be performed in future work.

Next, we investigate the response speed of our ADGG-EG-FET by fitting the response peaks using Gaussian functions, as shown in Fig. 8. For each pulse of the applied gate bias (V_g2) from −15.5 to 15 V, which corresponds to the transition from the plasmonic detection region to the plasmonic/PTE hybridized detection and PTE detection regions, the FWHM increases from ∼190–200 ps, as shown in Fig. 8(a). The plasmonic detection mechanism is very fast, on the order of 1 ps.⁴⁸ However, as discussed in Sec. II of the supplementary material, the response time of the PTE detection mechanism in our ADGG-EG-FET with a very long channel length is intrinsically limited by the intraband optical-phonon emission, which has an energy relaxation time on the order of 10 ps.⁴⁹ This qualitatively explains the observed slight increase of ∼10 ps in the FWHM due to the contribution of the PTE detection mechanism. In addition, the FWHM slightly decreases from 204 to 200 ps with increasing applied drain bias voltage, as shown in Fig. 8(b), suggesting the contribution of the PTE detection mechanism. All the thermodiffusion, electrostatic diffusion, and drift velocities increase with increasing drain bias voltage, shortening the detector response time. The difference between the FWHM of the output photoresponse, which has a minimum value of 190 ps, and the FWHM of the incident THz radiation, which is 155 ps, is attributed to the response speed of the ADGG-EG-FET detector as well as perturbative waveform distortions caused by systematic factors in interconnection environments to the oscilloscope. Considering that the rise and fall times are each dulled by 17.5 ps, the response time of our ADGG-EG-FET detector is estimated to be on the order of 10 ps.

Plasmonic THz detection works well under the zero drain-bias condition with zero-power consumption,^13–16 whereas the new type of unipolar PTE detection works only under the nonzero drain-bias condition with nonzero power consumption, as demonstrated in Fig. 6(b). In this regard, the ADGG-EG-FET can work in both cases, with and without any power supply, via current-driven? detection and zero drain-bias plasmonic detection.

The estimated intrinsic current responsivity of the ADGG-EG-FET detector, defined as the ratio of the photocurrent to the power of the incident THz wave at the active detector area, was ∼0.3 mA/W. Correspondingly, the intrinsic voltage responsivity under a high loaded impedance condition, which is given by the product of the current responsivity R_I and the channel resistance R_ch, was characterized to be 84 mV/W. To further investigate the level of responsivity in comparison with a recently published result with a maximal responsivity of 1.9 mA/W for an ADGG-EG-FET detector using the highest quality exfoliated hBN/graphene/hBN van der Waals heterostructures at 0.3 THz and 300 K,¹⁶ the observed responsivity of 0.3 mA/W at 0.95 THz and 300 K is ∼1/6 of the value reported at 0.3 THz and 300 K. The first factor that we should consider is the difference in detection frequency. As reported experimentally¹¹ and theoretically,⁴⁷ the plasmonic ratchet effect and drag effect have different frequency dependences on the responsivity, and the overall trend shows a monotonic decrease with increasing frequency. In the case of InGaAs/InAlAs/InP ADGG-HEMT detectors,¹¹ the responsivity at 1 THz weakens by one order of magnitude from that observed at 0.3 THz. Our result obtained with the ADGG-EG-FET showed less attenuation of ∼1/6, which is thought to be due to the superior transport properties of graphene Dirac fermions.

The second factor is considered to be the lower crystallinity of epitaxial graphene damaged throughout standard semiconductor integrated device processes, including the PE-CVD gate stack process used in this work, than that for the highest quality exfoliated graphene with low-damage exfoliation/transfer device processes reported in Refs. 15 and 16. According to previously published results,^49,50 the relaxation time (τ) of electrons can be estimated by the linear relationship between τ and the intensity ratio of Raman G and D peaks, which was calculated in the range of 8.25–25.30 for our epitaxial graphene, as shown in Fig. 2(b), resulting in τ values in the range of 1–2.5 ps. The corresponding carrier mobilities were identified to be $∼16,900∼43,000cm2/Vs$ (the derivation is given in Sec. I of the supplementary material). Our graphene sample was fabricated using a thermal decomposition method from the (⁠$0001̄$⁠) surface of a chemically and mechanically polished SiC wafer. This method is able to produce high-quality graphene with a carrier mobility of 100 000 cm²/Vs even at room temperature, which was experimentally observed by using time- and angle-resolved photoemission spectroscopy (T-ARPES).⁵¹ In this reported experiment, the in situ measured nonequilibrium carrier energy relaxation delay time was properly fitted using parameters including the Fermi velocity and carrier mobility.⁵¹ Moreover, in another EG-FET device fabricated on this kind of graphene membrane in our previously published report,⁵² the mobility is derived from the electric properties of EG-FET as 50 600–63 300 cm²/Vs. According to these results, our estimated value of mobility in this paper is reasonable. Compared with the mobility of ∼38 000 cm²/Vs for the ADGG-graphene-channel FET¹⁶ fabricated using mechanical exfoliation and a transfer process with an encapsulation sandwiched by two h-BN thin layers, the carrier mobility of the epitaxial graphene in this work is fairly comparable. This result shows that such a process-dependent degradation of the quality factor will be minor. Further quantitative investigation will be given after future experimental works.

The PTE effect could enhance the detection sensitivity in the current-driven ADGG-EG-FET detector from the level of the plasmonic rectification effect without any deterioration of the fast response speed. The obtained 10 ps-order fast detector response meets the requirements for a fast, sensitive receiver to be applicable in 6G/7G-class next-generation THz wireless communication systems. In addition, all of the device fabrication processes are suitable for mass production implemented with current semiconductor integrated device processing technologies.

Finally, we estimated the noise equivalent power (NEP) taking into account both the thermal noise and the shot noise as follows:¹¹ 

where N_ThV and N_ShV are the thermal noise and shot noise factors in units of $A/Hz$⁠, where k_B is the Boltzmann constant, T_e is the electron temperature, e is the elementary charge, I_ch is the channel current, and R_ch is the channel resistance. In these experiments, the output photovoltaic response was smaller than the true signal by a factor of the voltage divider ratio of R_L/(R_ch + R_L), where R_L is a load resistance of 50 Ω. With the internal channel resistance R_ch ∼ 300 Ω, the photovoltages of the true signals are expected to be ∼7 times higher than the observed outputs. To use our detector in high-speed wireless transmissions, the highest signal integrity with the least distortion is required. The high-impedance-loaded condition in our detector can work if the interconnection length between the detector output and the input of the next-stage circuitry, such as a preamplifier, can be managed within 100 μm (less than 10% of the electrical wavelength under consideration). Such a circuit configuration is routinely obtainable in monolithic integrated circuit technology.

The NEP and the intrinsic current responsivity R_i vs V_ds are plotted in Fig. 9. According to Eqs. (1)–(3), we obtained a minimum NEP value of $166nW/Hz$ at the maximum R_i (R_v) of 0.3 mA/W (84 mV/W) at room temperature. These responsivity and noise performances are comparably high among those reported for graphene THz detectors but lower than those reported for InGaAs/InAlAs/InP ADGG-HEMTs. This is because gapless graphene prevents complete carrier depletion, giving rise to a relatively low channel resistance reflecting the low responsivity. The introduction of bilayer graphene to open the bandgap with the application of the vertical electric field intensity will drastically improve the responsivity to a certain extent without severe degradation of the speed performance.

We designed and fabricated our original ADGG-EG-FET structure for a current-driven terahertz detector and experimentally demonstrated a 10-ps-order fast temporal response, a high responsivity of 0.3 mA/W, and an NEP of 166 nW/√Hz over wide ranges of gate and drain bias voltages at room temperature. When one gate voltage V_g2 was biased at −15.5 V near the Dirac voltage and the other gate voltage V_g1 was fixed at +15 V higher than the Dirac voltage under the condition of a nonzero drain voltage V_ds = +1 V, the plasmonic rectification dominated the detection, and the photoresponse increased with increasing V_g2 from −15.5 to +15 V. We identified the detection mechanism that we observed as a new type of unipolar-type PTE detection in which only electrons or holes contribute to rectifying the THz radiation, assisted by anisotropic electrostatic drift/diffusion under current-driven conditions. The ADGG bias dependence of the measured photoresponse showed a clear transition between plasmonic and PTE detection via a rather wide intermediate range where these two detection mechanisms coexisted while preserving the fast response speed. Within the applied bias voltage range for the ADGG, the current-driven PTE detection exhibited a superior responsivity almost twice as high as that for current-driven plasmonic detection while preserving the fast response speed. Moreover, the EG-FET using epitaxial graphene on a SiC substrate was demonstrated to be a suitable device for mass production integrated with device process technology. Recently, a novel technique for synthesizing high-quality few-layer epitaxial graphene on a single-crystalline SiC thin film grown on a Si wafer was developed.³¹ The precise control of epitaxially grown graphene layers is also a remaining issue, which could be managed by introducing a microfabricating SiC substrate technique that could spatially confine the epitaxy area.³⁰ In conclusion, the obtained results indicate that the ADGG-EG-FET THz detector has a promising potential for applications in 6G- and 7G-class high-speed wireless communication systems.

See the supplementary material for the derivation of the field-effect mobility of graphene carriers.

The device was fabricated at the Nanoelectronics and Spintronics Laboratory, RIEC, Tohoku University, Japan. This research was partly carried out at the Fundamental Technology Center, Research Institute of Electrical Communication, Tohoku University. This work was supported by JSPS KAKENHI Grant Nos. 16H06361, 18H05331, 20K20349, and 21H01380, Japan. These research results were obtained from the Commissioned Research by NICT Grant No. 01301, Japan.

The authors have no conflicts to disclose.

Koichi Tamura: Data curation (lead); Investigation (equal); Validation (equal); Visualization (equal); Writing – original draft (lead). Chao Tang: Data curation (equal); Visualization (equal); Writing – original draft (supporting). Daichi Ogiura: Investigation (equal); Validation (equal). Kento Suwa: Resources (equal). Hirokazu Fukidome: Resources (equal). Yuma Takida: Methodology (supporting); Resources (equal). Hiroaki Minamide: Methodology (supporting); Resources (equal). Tetsuya Suemitsu: Resources (equal). Taiichi Otsuji: Conceptualization (supporting); Funding acquisition (lead); Project administration (equal); Supervision (supporting). Akira Satou: Conceptualization (lead); Investigation (supporting); Methodology (lead); Project administration (equal); Supervision (lead); Writing – review & editing (equal).

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