betaFeedback

WorksPhysics and Astronomy

17 GHz Lossless InP-Membrane Active Metasurface

Abstract

High-speed active metasurfaces enable spatiotemporal control of incident light within an ultra-thin layer, offering new possibilities for optical communication, computing, and sensing. However, a fundamental tradeoff between electrical conductivity and optical absorption of the material has hindered the realization of active metasurfaces that simultaneously achieve broad modulation bandwidth and low optical loss. Here, we experimentally demonstrate a high-speed active metasurface operating in the 1.5-μm wavelength range that realizes a record-high modulation bandwidth of 17.5 GHz, while maintaining a high quality (Q) factor of 102 and an ultra-low optical loss of 0.56 dB. The key enabling technology is the indium-phosphide (InP) membrane platform; an n-type InP offers both high electron mobility and low free-carrier optical absorption, making it an ideal material for active metasurface devices. The high-Q Friedrich-Wintgen quasi-bound-states-in-the-continuum mode inside the InP-membrane high-contrast grating (HCG) is utilized to trap the normally incident light within an organic electro-optic (OEO) material, enabling efficient modulation. InP HCG also serves as an ultralow-resistance interdigitated electrodes for applying high-speed electrical signals to the OEO material, thereby offering 50-fold improvement in modulation bandwidth compared to conventional silicon-based counterparts. Our work paves the way towards high-speed, low-loss active metasurfaces for spatiotemporal control of light beyond the gigahertz regime.

Introduction

Metasurfaces are flat optical devices composed of nanostructures that enable complex manipulation of lightwaves within an ultra-thin device. Over the past decade, various passive metasurface devices, such as flat metalenses[1, 2, 3], polarimeters[4, 5, 6], optical hybrids[7, 8, 9], and color routers[10, 11, 12], have been demonstrated, achieving the functionalities impossible with conventional optics[13, 14, 15]. To further extend the capabilities of metasurfaces to the spatiotemporal regime, development of reconfigurable metasurfaces, also known as active metasurfaces, is desired[16, 17, 18, 19, 20, 21]. Of particular interest is the realization of ultra-high-speed active metasurfaces surpassing the gigahertz (GHz) regime, which hold immense potential to revolutionize diverse fields of optics, including large-capacity optical communication[22, 23] [24], optical computing25–28, and high-speed optical sensing29–31.

Active metasurfaces can be realized either by mechanically displacing the photonic nanostructures31–34 or by modulating the refractive index of materials35–60. To realize a high-speed device operating beyond the GHz regime, the latter strategy is required, utilizing the high-speed electro-optic (EO) effects such as the Pockels effect38–55, free carrier effect56–59, and quantum-confined Stark effect60. However, such high-speed EO effects typically fall short of effectively modulating the incident light within the limited interaction length of a thin metasurface.

Therefore, optical resonating structures are typically utilized in the EO-effect-based active metasurfaces, where the incident light is trapped within a thin metasurface layer to enhance the light-matter interaction. The quality (QQ) factor of the resonator should not be so high that it restricts the modulation bandwidth, but needs to be sufficiently large (Q>100Q > 100) to achieve adequate modulation efficiency. Then, to efficiently apply high-speed electrical signals to the EO material, it is crucial to integrate highly conductive electrodes at the vicinity or within the optical resonating mode to enable efficient modulation. At the same time, they must not introduce optical loss to the resonator. A lossy cavity not only degrades the QQ factor, but also severely restricts the wavefront modulation capability; a lossy cavity generally becomes under-coupled, whereas efficient phase modulation with 2π2\pi tunable range requires an over-coupled cavity61.

Nevertheless, simultaneously satisfying these two requirements is challenging. Metals can serve as ideal low-resistance electrodes, but they suffer from a plasmonic loss, which inevitably reduces the QQ factor of the cavity. On the other hand, doped silicon (Si) has been widely employed in the previous work to achieve high QQ resonance46–50, 56–59. However, Si also suffers from the strict tradeoff between the high electrical conductivity and low optical loss; while it is essential to increase the doping concentration to reduce the electrical resistance, it simultaneously leads to substantial increase in optical absorption through the free carrier plasma effect of Si. As a result, the realization of a low-loss active metasurface with >GHz> \mathrm{GHz} EO bandwidth remains elusive to date.

Here, we demonstrate a novel platform that comprises an indium-phosphide (InP) membrane high-contrast grating (IM-HCG) embedded with organic electro-optic (OEO) material to realize ultra-high-speed active metasurfaces, operating at 1.5-μm wavelength range (Figure 1a-Figure 1d). As shown in Figure 1e-Figure 1g, n-InP outperforms n-Si both in terms of electron mobility and optical absorption, thereby offering substantial improvements over Si-based active metasurfaces. Using n-InP for both the highly conductive electrode and low-loss HCG-based optical resonator, our device efficiently traps the incident light within the OEO material and modulated by applying electric field via InP. The IM-HCG embedded with EO polymer is successfully fabricated to demonstrate a record-high 3-dB modulation bandwidth of 17.5 GHz, optical resonance with a moderately high QQ factor of 102, and ultra-low insertion loss of 0.56 dB. The measured modulation efficiency of the resonant wavelength is 18 pm/V, which can be enhanced by employing an OEO material with higher EO coefficient and by improving the fabrication process.

Figure 1. Concept of the proposed InP-membrane active metasurface. a Schematic of the proposed device. IM-HCG is realized by forming periodic InP-membrane bars embedded with OEO material on quartz substrate. InP bars simultaneously function as a resonator to trap the normally incident input light, and as an interdigitated electrode to apply voltage to the OEO material. By shifting the resonance wavelength, both the reflectance and transmittance of the IM-HCG can be modulated. b Cross-sectional schematic of IM-HCG. A normally incident yy-polarized light is input to the device and couples to a resonating mode. By applying electric field to the EO polymer via the InP bars, the resonance wavelength is shifted. c Top-view schematic of IM-HCG. The InP bars form interdigitated electrodes to apply voltage signal to the OEO material. The modulation bandwidth of the device is limited by the RC time constant of the circuit, and thus the reduction of the resistivity of the InP bars is essential for high speed operation. d The electric field intensity (∣Ey∣2)(|E_y|^2) distribution of the resonating mode of the IM-HCG (λ=1550\lambda= 1550 nm, Λ=760\Lambda= 760 nm, w=380w = 380 nm, d=630d = 630 nm). The light is strongly confined in the OEO material, enabling efficient modulation. e,f Doping concentration dependence of the electron mobility (e) and absorption coefficient (f) for n-InP73–75, 77 and n-Si76, 78. n-InP is superior to n-Si in both mobility and absorption across the entire range of doping concentration. g The tradeoff between the electrical resistivity and optical absorption for n-InP and n-Si.

Results

Concept and operating principle of IM-HCG active metasurface

Figures 1a-c show the schematic of the proposed IM-HCG active metasurface. The IM-HCG is composed of arrayed InP membrane bars on a quartz (SiO2_2) substrate, which are embedded with an OEO material. An HCG generally exhibits a wide variety of optical characteristics by judiciously designing its thickness dd, period Λ\Lambda, and bar width ww46, 62 (see Supplementary Fig. S1). Notably, at the dual-transverse-mode region (1.27 <λ/Λ<< \lambda/\Lambda< 2.2 in supplementary Figure S1), high-QQ resonances can be obtained at specific values of dd through the bimodal resonance effect46, 62, 63, which are also recognized as Friedrich-Wintgen quasi-bound-states-in-the-continuum (FW-qBICs)64. In particular, a high-QQ resonating mode is obtained when d/Λ∼0.8d/\Lambda\sim0.8 (Supplementary Fig. S1 inset). The electric field intensity profile of this mode for λ=1550\lambda= 1550 nm is depicted in Figure 1d. The field is strongly confined inside the narrow gap of HCG, enabling efficient modulation of light through the refractive index change of the OEO material.

Figure S1. Resonance spectrum of IM-HCG. The reflectance contour map of the IM-HCG as a function of wavelength λ\lambda and grating thickness dd normalized by the grating period Λ\Lambda (Λ=760\Lambda= 760 nm, w=380w = 380 nm). By judiciously designing Λ\Lambda and dd, a sharp resonance can be obtained through the bimodal resonance effect. The inset shows the magnified view of the contour map near the resonant mode employed in our device.

The InP bars serve not only as an optical resonator but also as interdigitated electrodes for applying electric field to the OEO material (Figure 1b), so that its refractive index is modulated through the Pockels effect46, 47. As a result of the resonance wavelength shift, the reflectance and transmittance of the active metasurface are complementarily modulated as depicted in Figure 1a inset. The proposed device is lossless if the material absorption is negligible. In such a case, the sum of the reflectance and the transmittance becomes unity; the device functions as a unitary beam splitter, whose splitting ratio can be modulated at a high speed.

The modulation bandwidth of the device is determined by the RC time constant of the InP bars (Figure 1c). Therefore, it is crucial to reduce the electrical resistance of the InP bars, typically by increasing the doping concentration NDN_D. However, higher doping concentration generally induces free carrier absorption. Thus, the doping concentration needs to be kept below a certain level where the effect of optical absorption is negligible. To this end, n-InP is superior to n-Si due to its excellent characteristics in terms of optical absorption and electrical conductivity65–67. First, the electron mobility μ\mu of n-InP outperforms that of n-Si by nearly an order of magnitude at the same doping concentration (Figure 1e). Second, the doping-induced optical absorption of n-InP is one order smaller than that of n-Si (Figure 1f). Under the same constraint on the maximum allowed optical absorption, therefore, significantly larger doping concentration can be employed for n-InP compared to n-Si. As a result, the resistivity of n-InP bars, which is inversely proportional to μND\mu N_D, can be one to two orders of magnitude smaller than that in the case of n-Si (Figure 1g). Indeed, for an absorption coefficient of α=10\alpha= 10 cm−1^{-1}, the resistivity of n-InP is 50-fold smaller than n-Si.

Numerical analysis

Figure 2a shows the numerically simulated reflectance spectrum of IM-HCG (Λ=760\Lambda= 760 nm, d=630d = 630 nm, and w=380w = 380 nm) with various doping concentrations NDN_D (see the Methods for details). Even at a large doping concentration of ND=1×1019N_D = 1 \times10^{19} cm−3^{-3}, the reflectance peak at the resonant wavelength is kept high above 95%. For comparison, Figure 2b shows the case for n-doped Si-membrane (SM)-HCG (Λ=748\Lambda= 748 nm, d=540d = 540 nm, w=374w = 374 nm) for various values of NDN_D. Unlike IM-HCG, the reflectance peak drops rapidly as NDN_D exceeds 1×10181 \times10^{18} cm−3^{-3}. These results directly reflect the one-order-of-magnitude larger free-carrier absorption of Si compared with InP (Figure 1f).

Figure 2. Numerical analysis. a,b The reflectance spectra of IM-HCG (a) and SM-HCG (b). c The QQ factor and optical loss at the resonance as a function of doping concentration for both IM-HCG and SM-HCG. d,e The reflectance spectra of IM-HCG (d) and SM-HCG (e) at ND=1×1019 cm−3N_{\mathrm{D}} = 1 \times10^{19}\ \mathrm{cm}^{-3} for various nEOn_{\mathrm{EO}}. f Characteristics of IM-HCG and SM-HCG (ND=1×1019 cm−3N_{\mathrm{D}} = 1 \times10^{19}\ \mathrm{cm}^{-3}) when used as reflective modulators. g The 3-dB EO bandwidth of IM-HCG and SM-HCG active metasurfaces as a function of doping concentration.

Figure 2c summarizes the impact of the doping concentration on the QQ factor as well as overall optical loss of the modulator for both cases. Here, the optical loss (dB) is defined as −10log⁡10(R+T)-10\log_{10}(R+T), where RR and TT denote the reflectance and transmittance at the resonant wavelength, respectively. For IM-HCG, the QQ factor is larger than 900 and the optical loss is as small as 0.24 dB even at a relatively high doping concentration of ND=1×1019 cm−3N_D = 1 \times10^{19}\ \mathrm{cm}^{-3}, thanks to the small free carrier absorption of n-InP. In contrast, the QQ factor degrades and the optical loss increases rapidly at ND>1×1018 cm−3N_D > 1 \times10^{18}\ \mathrm{cm}^{-3} for SM-HCG.

Figures 2d and 2e show how the reflectance spectrum changes as we modulate the refractive index of the OEO material nEOn_{\mathrm{EO}} for the IM-HCG and SM-HCG, respectively. We assume ND=1×1019 cm−3N_D = 1 \times10^{19}\ \mathrm{cm}^{-3} in both cases. A large resonant wavelength shift of Δλ/ΔnEO≃450 nm/RI\Delta\lambda/\Delta n_{\mathrm{EO}} \simeq450\ \mathrm{nm/RI} is obtained for both IM-HCG and SM-HCG, thanks to the strong confinement of light within the OEO material. Figure 2f compares the characteristics of the IM-HCG and SM-HCG when they are used as reflective optical intensity modulators. The operating wavelength is selected to be 1547.4 nm for IM-HCG and 1548.6 nm for SM-HCG. A small insertion loss of 0.21 dB is obtained with IM-HCG, while that of SM-HCG is as large as 4.3 dB. Moreover, thanks to the large QQ factor of IM-HCG, 5-dB modulation can be obtained at a small index change of ΔnEO=2.2×10−3\Delta n_{\mathrm{EO}} = 2.2 \times10^{-3}. This corresponds to a required driving voltage of Vpp=5.6V_{\mathrm{pp}} = 5.6 V, assuming a simple parallel-plate model and the EO coefficient of the state-of-the-art OEO materials (r33=200 pm/Vr_{33} = 200\ \mathrm{pm/V})68–70. In contrast, larger index change of ΔnEO=3.5×10−3\Delta n_{\mathrm{EO}} = 3.5 \times10^{-3} is required to achieve 5-dB modulation for SM-HCG due to its degraded QQ factor. This corresponds to a required voltage of Vpp=8.7V_{\mathrm{pp}} = 8.7 V.

Figure 2g shows the simulated 3-dB EO bandwidth as a function of the doping concentration for IM-HCG and SM-HCG. When the device size is 40×40 μm240 \times40\ \mathrm{\mu m}^{2} and ND=1×1019 cm−3N_D = 1 \times10^{19}\ \mathrm{cm}^{-3}, the 3-dB bandwidth of IM-HCG active metasurface can exceed 40 GHz, which is comparable to that of waveguide-based EO modulators deployed in current optical communication systems. On the other hand, the 3-dB bandwidth of SM-HCG is limited to around 3 GHz at ND=1×1019 cm−3N_D = 1 \times10^{19}\ \mathrm{cm}^{-3}. Moreover, at such a large NDN_D, the optical absorption is non-negligible and exceeds 3 dB for SM-HCG (Figure 2c). At ND=1×1018 cm−3N_D = 1 \times10^{18}\ \mathrm{cm}^{-3}, where the optical absorption is less than 0.34 dB and comparable to that of IM-HCG at ND=1×1019 cm−3N_D = 1 \times10^{19}\ \mathrm{cm}^{-3}, the 3-dB bandwidth is further reduced to 800 MHz. Thus, we can achieve 50-fold improvement in the EO bandwidth by using IM-HCG, reaching tens of gigahertz, without increasing the insertion loss.

Device fabrication and characterization

Figure 3 shows the IM-HCG device, which was fabricated by bonding a 590-nm-thick n-InP membrane layer on a quartz (SiO2_2) substrate (see Methods for details of the device fabrication). As the OEO material, we employed the side-chain EO polymer71 with a glass transition temperature of 132.5 °C. Ti/Ni/Au electrode with ground-signal-ground (GSG) pads were integrated to enable high-speed characterization. The active metasurface region had a footprint of 40×40 μm240 \times40\ \mathrm{\mu m}^{2}. The electron mobility and doping concentration of the n-InP membrane layer were measured to be 1670 cm2^{2}/Vs and 3.4×1018 cm−33.4 \times10^{18}\ \mathrm{cm}^{-3}, respectively, which are consistent with the values reported in the literature. The electrical capacitance of the device was 227 fF, which also agrees well with the FEM analysis (see Supplementary Note 3).

Figure 4a shows the measured reflectance and transmittance spectra of the IM-HCG with Λ=750\Lambda= 750 nm and w=405w = 405 nm. The experimental setup is depicted in Figure 4b (see Methods for details of characterization). We can see a clear optical resonance at λ=1510\lambda= 1510 nm with a QQ factor of 102. We can confirm that the sum of the reflectance RR and transmittance TT is more than 0.88 (−0.56-0.56 dB) over the entire wavelength range, indicating that our device is almost lossless. By applying a direct current (DC) voltage to the device, a clear shift of resonance wavelength at an efficiency of 18 pm/V is observed through the Pockels effect. Figure 4c shows the reflected and transmitted optical power observed when we apply 20-VppV_{\mathrm{pp}} sinusoidal voltage signal to the device. We can see that the transmittance and reflectance are complementarily modulated at the opposite phases. The modulation depths of the reflectance and transmittance are ΔR/R=3.1%\Delta R/R = 3.1\% and ΔT/T=3.35%\Delta T/T = 3.35\%, respectively.

Figure 4. DC and low-frequency characterization of the fabricated device. a The reflectance (RR) and transmittance (TT) spectra of the device under different voltages (+30 V, 0 V, −30 V). The sum of RR and TT is also plotted, which exceeds 0.88 across the entire measured wavelength range. The magnified view is depicted in the right inset, which shows clear spectral shift under modulation. b Experimental setup. TLD: tunable laser diode. PC: polarization controller. Coll.: collimator. Pol.: polarizer. BS: beam splitter. Obj.: objective lens. DUT: device under test. PA: picoammeter. FG: function generator. PM: power meter. Rx: receiver. APD: avalanche photodiode. TIA: trans-impedance amplifier. Osc.: oscilloscope. c The reflected (red) and transmitted (blue) optical power measured when a 200-Hz 20-Vpp_{\mathrm{pp}} sinusoidal voltage is applied to the device.

Finally, Figure 5a shows the frequency response of the device, measured using the setup shown in Figure 5b. The device exhibits a 3-dB bandwidth of 17.5 GHz, which agrees excellently with the FEM analysis (see Figure 2g). To the best of our knowledge, this represents the largest 3-dB bandwidth among all experimentally demonstrated active metasurfaces reported to date (see Supplementary Table 1).

Figure 5. High-speed characterization of the fabricated device. a Measured EO frequency response of the device. A magnified view is depicted in the right inset. The 3-dB bandwidth is 17.5 GHz. b Experimental setup. The reflected light from the device is coupled to a fiber, amplified, and converted to an electrical signal by a biased photodiode (PD). VNA: vector network analyzer. EDFA: erbium-doped fiber amplifier. OBPF: optical bandpass filter.

Discussion

We have demonstrated an ultra-high-speed IM-HCG active metasurface device. Thanks to the large electron mobility of n-InP, our device exhibits a record-high 3-dB EO bandwidth of 17.5 GHz. This is nearly 6-fold improvement from the previously demonstrated active metasurfaces of all kinds, including plasmonic metasurfaces with metallic electrodes (see Supplementary tbl:1 for the comparison). At the same time, owing to the small free-carrier absorption of n-InP, our device efficiently traps the incident light within a thin OEO layer without incurring significant optical absorption, so that R+T>0.88R + T > 0.88, which corresponds to an ultra-low optical loss of 0.56 dB. This is also the record-low value among the active metasurfaces reported in the literature, while it exhibits a decently large QQ factor, exceeding 100.

The demonstrated device has a plenty of room for improvement in terms of modulation efficiency. First, the measured QQ factor of the resonator is ∼102\sim102, which is substantially smaller than that (Q∼900Q \sim900) expected from numerical simulation. This inconsistency is attributed to the imperfections in forming perfect IM-HCG structure, such as the sidewall roughness (Figure 3c,Figure 3d), which can be improved by optimizing the fabrication conditions. In addition, the EO coefficient r33r_{33} of our fabricated device is estimated to be ∼20\sim20 pm/V, assuming r33∼3r13r_{33} \sim3r_{13}72. We expect that r33r_{33} can be increased by employing an OEO material with a higher r33r_{33} coefficient (as large as 500 pm/V70) and by improving the poling conditions. Assuming that the QQ factor and the r33r_{33} coefficient are enhanced to 930 and 200 pm/V, respectively, we numerically estimate that a 5-dB modulation can be achieved with a driving voltage of approximately 5.6 Vpp_{\mathrm{pp}}.

Fabricated device with microscopic, magnified, and scanning electron microscope images

Figure 3. Fabricated device. a Microscopic image of the fabricated device. b Magnified view of the IM-HCG before coating the EO polymer. c,d Scanning electron microscope (SEM) images of the InP bars before coating the EO polymer.

Furthermore, the EO modulation bandwidth can be extended by increasing the doping concentration of n-InP. While the actual doping concentration of n-InP in our fabricated device was approximately 3.4×10183.4 \times10^{18} cm−3^{-3}, we have numerically demonstrated that it could be increased to 1×10191 \times10^{19} cm−3^{-3} without inducing a substantial optical loss (Figure 2c). Additionally, reducing the device size should be effective in extending the modulation bandwidth further due to smaller overall capacitance. We numerically estimate that an ultra-broad bandwidth exceeding 100 GHz should be attained at a device size of 24×2424 \times24 μm2^{2} (see Supplementary Fig. S7). Our work, therefore, provides the pathway towards the realization of ultra-high-speed spatiotemporal control of lightwaves using active metasurfaces.

Methods

Numerical analysis of the optical characteristics

The optical characteristics of the IM-HCG and SM-HCG shown in Figure 1d and Figure 2a-Figure 2f were obtained by two-dimensional finite-difference time-domain (FDTD) simulation (Ansys Lumerical FDTD) in the zxzx plane. For simplicity, the HCGs were assumed to be infinitely long in the yy direction and the periodic boundary condition was employed in the xx direction. A normally incident yy-polarized plane wave was input to the HCG. The refractive indices at λ=1550\lambda= 1550 nm were used for all materials and the chromatic dispersion of the materials was neglected for simplicity. The duty ratio of the grating was set to w/Λ=0.5w/\Lambda= 0.5 for all cases.

The absorption coefficients of n-InP and n-Si for various levels of doping concentration were obtained by using the values from the literature73–76 to fit an empirical expression of α(ND)=A⋅(ND)B\alpha(N_D) = A \cdot(N_D)^B, where AA and BB are the fitting parameters. The effect of absorption was then taken into the FDTD simulation by modifying the imaginary part of the complex refractive index of either n-InP or n-Si. The doping-induced changes in the real part of the refractive index was neglected to avoid the shift in the resonant wavelength, allowing direct comparison among different doping conditions at a same wavelength. For the optical loss plotted in Figure 2c, we selected the largest value across the entire wavelength range shown in Figure 2a and Figure 2b.

Numerical analysis of the electrical characteristics

The electrical characteristics of the device was simulated using the three-dimensional finite element method (FEM) (Ansys Lumerical CHARGE). The doping-concentration-dependent mobility was obtained from the Caugh-Thom as-like model77 for n-InP and from the Masetti model78 for n-Si (see Supplementary Note 5 for details).

Device fabrication

The fabrication flow is shown in Supplementary Fig. S2. First, an InP epitaxial substrate [n-InP (590 nm), InGaAs (50 nm), InP (50 nm), InGaAs (100 nm), InP (substrate)] was prepared by the metal organic vapor phase epitaxy (MOVPE) growth. Thin Al2_2O3_3 adhesive layers were formed on the InP epitaxial wafer and the quartz wafer by the atomic layer deposition (ALD) before they were bonded79. The bulk InP substrate and the sacrificial epitaxial layers were removed through the wet-etching processes using hydrochloric acid (for InP) and phosphoric acid (for InGaAs), leaving only the 590-nm-thick n-InP membrane on the quartz substrate. After sputtering SiO2_2 as a hard mask layer, electron beam (EB) lithography was performed to define the IM-HCG pattern. The SiO2_2 hard mask and InP were then etched by the reactive-ion etching (RIE) processes (CHF3_3 and C4_4F8_8/SF6_6 for SiO2_2, cyclic process of CHF3_3/H2_2 and O2_2 for InP). After removal of the SiO2_2 hard mask, Al2_2O3_3 passivation layer is formed by ALD. Subsequently, Ni/Ti/Au electrodes were formed through EB lithography, EB evaporation, and lift-off process. Then, the EO polymer is deposited by spin coating process, followed by the sputtering of SiO2_2 layer to protect the EO polymer. The IM-HCG region is defined by photolithography, and RIE is performed to etch the SiO2_2 (CHF3_3) and EO polymer (CHF3_3/O2_2). Finally, the EO polymer was poled by applying an electric field of 73 V/μm to the EO polymer at an elevated temperature of 131 °C, followed by rapid cooling while maintaining the poling field.

Electrical characterization

The sheet resistance and contact resistance of the device were characterized by the transfer length method (TLM). The resistivity of the 590-nm-thick n-InP membrane was measured to be 1.1×10−31.1 \times10^{-3} Ω · cm. The contact resistance of the n-InP was as low as 9.5×10−69.5 \times10^{-6} Ω · cm2^{2}. The electron mobility and the carrier concentration of n-InP were measured using the Hall measurement method. The capacitance of the device was measured using a semiconductor parameter analyzer. See Supplementary Note 4 for further details of the test patterns and the characterization result.

Electro-optic characterization

The DC and low-frequency measurement of the EO characteristics of the device was performed using the setup depicted in Figure 4b. Continuous-wave (CW) light was launched from the tunable laser diode (TLD), controlled to yy polarization, and emitted to free space via a fiber collimator. The light was then delivered to the IM-HCG device via a 3-dB beam splitter (BS) and an objective lens. The reflected light was collected by the same objective lens and coupled to a fiber via the beam splitter and a fiber collimator. The reflected optical power was measured by a power meter (PM) for the reflectance spectrum measurement and by an optical receiver consisting of an avalanche photodiode (APD) and trans-impedance amplifier (TIA) for the low-frequency modulation measurement. The transmitted light was measured by a bucket power meter. The reflectance and transmittance were calibrated by normalizing them to the measured values of reflection from a reference gold mirror and transmittance without any device, respectively. The voltage signal was applied to the device via DC probes. We used a picoammeter (PA) for applying a DC voltage and a function generator for applying a sinusoidal voltage. The QQ factor of the device was extracted by fitting the reflection spectra to the Fano resonance model (see Supplementary Note 6).

The high-speed characterization of the device was carried out using the setup shown in Figure 5b. The fiber-coupled reflected light was amplified by an erbium-doped fiber amplifier (EDFA), filtered by an optical band pass filter (OBPF), power-controlled by a variable optical attenuator (VOA), and detected by a high-speed photodiode. The frequency response of the device was measured using a vector network analyzer (VNA). The signal launched from the VNA was applied to the device via an RF probe and the electrical signal from the photodiode was delivered to the VNA.

Data availability

The data that support the findings in this study are included in the article and its supplementary information. Other data are available from the corresponding authors upon reasonable request.

Acknowledgements

This work was supported in part by the commissioned research (JPJ012368C03601, JPJ012368C08801) by National Institute of Information and Communications Technology (NICT) and Japan Society for the Promotion of Science (JSPS KAKENHI JP23H05444, JP23H00272, JP21J11982, JP21J10272). T.F. and K.S. acknowledge supports from Materials Education program for the future leaders in Research, Industry, and Technology (MERIT) of the University of Tokyo. Portion of the device fabrication was conducted at the cleanroom facilities of d.lab in the University of Tokyo, supported by MEXT Nanotechnology Platform, Japan. The authors acknowledge Shingo Kaneta-Takada and Makoto Fujiwara for their technical supports.

Author contributions

T.F. and T.T. conceived the concept. T.F. performed the numerical simulation with some assistance from H.M. and G.S.. K.S. and T.F. conceived the device fabrication flow. W.Y. and T.F. performed the epitaxial wafer growth with assistance from E.K.. T.F. and K.S. fabricated the device with assistance from H.M., E.K., G.S., R.T, and J.L.. K.S. and T.F. performed the electrical characterization. T.F. conceived and performed the optical and electro-optic characterization with some assistance from H.M. and G.S.. T.Y. and A.O. developed the EO polymer. K.T., M.T., S.T., and Y.N contributed to overall discussion and provided experiment facilities. T.T. supervised the project. T.F. and T.T wrote the manuscript with the inputs from all authors.

Competing interests

The authors declare no competing interests.

Additional information

Supplementary information The online version of this paper contains supplementary material. Correspondence and requests for materials should be addressed to Taichiro Fukui and Takuo Tanemura.

Bimodal resonance in HCG

Figure S1 shows the reflectance contour map of the IM-HCG as a function of the wavelength λ\lambda and the thickness of the grating dd for a normal-incident yy-polarized light, having the electric field oriented parallel to the grating. The inset shows the magnified view near the high-QQ resonant mode employed in our device. In these plots, we set Λ=760\Lambda= 760 nm and the width of the InP bars to w=380w = 380 nm, so that the duty ratio is w/Λ=0.5w/\Lambda= 0.5. We assume that the refractive indices of all materials are constant: 3.17, 1.65, and 1.44 for InP, OEO material, and SiO2_2, respectively. In the dual transverse mode region, high-QQ resonance can be achieved by judiciously tuning the thickness of the grating dd[1, 2, 3]. Here, the transverse mode refers to the optical mode that propagates in the zz-direction through an infinitely thick grating, subject to a periodic boundary condition in the xx-direction.

Fabrication process

Figure S2 shows the fabrication flow of the device. Details of the flow are described in the Methods section.

Figure S2. Fabrication flow of the device. a Epitaxial layers grown on InP substrate. b The fabrication flow of the device.

Electrical capacitance of the device

Figure S3a shows the CV measurement result (measurement frequency: 1 MHz) of the devices with two different sizes, 40×40 μm240 \times40\ \mu\mathrm{m}^{2} and 80×80 μm280 \times80\ \mu\mathrm{m}^{2}. The capacitance of the device is almost independent on the bias voltage. Figure S3b shows the simulated capacitance of the device for various size using finite element methods (FEM). Here, the doping concentration is set to ND=3.4×1018 cm−3N_{\mathrm{D}} = 3.4 \times10^{18}\ \mathrm{cm}^{-3}. The measured capacitance of the device shows good agreement with the numerical analysis, where the capacitance scales quadratically with the device size LL. Figure S3c shows the simulated frequency-dependent impedance of the device. At low frequency, the lumped capacitance model (∣Z∣=1/ωC\lvert Z \rvert= 1/\omega C) describes the impedance characteristics well. At higher frequency, the impedance saturates, limited by the resistivity.

Figure S3. Figure S3. Capacitance of the device. a Experimentally measured capacitance-voltage characteristics of the fabricated devices with two different sizes, 40×40 μm240 \times40\ \mu\mathrm{m}^{2} and 80×80 μm280 \times80\ \mu\mathrm{m}^{2}. b Simulated capacitance of the device as a function of size, LL. Experimental results are also plotted, showing excellent agreements. c Simulated frequency-dependent impedance of the device.

Electrical characterization of n-InP membrane

Figure S4 shows the schematic of the test pattern for the transfer length method (TLM). By measuring the resistivity with different gap length LgL_g while keeping the electrode pattern same (width WW), the sheet resistance RsR_s of the InP membrane and the contact resistance RcR_c of the Ni/Ti/Au contact to InP can be extracted. Figure S4 shows the TLM result. By fitting the data to the equation

Figure S4. Transmission line measurement (TLM). a Schematic of the test pattern. b Measured results.

R=2Rc+RsLg/W,(1)R = 2R_c + R_s L_g/W, \tag*{(1)}

RsR_s and RcR_c are extracted to be 18.9 Ω\Omega□ and 1.68 Ω\Omega, respectively. Given the InP membrane thickness d=590d = 590 nm, the resistivity ρs\rho_s of InP is given as ρs=Rs×d=1.1×10−3\rho_s = R_s \times d = 1.1 \times10^{-3} Ω⋅\Omega\cdot cm. On the other hand, according to the transmission line model, the contact resistivity ρc\rho_c of n-InP/Ni/Ti/Au contact can be derived as[4]

ρc=Rc2W2Rs=9.5×10−6 Ω⋅cm2.(2)\rho_c = \frac{R_c^2 W^2}{R_s} = 9.5 \times10^{-6}\ \Omega\cdot\mathrm{cm}^2. \tag*{(2)}

This small contact resistivity is attributed to the formation of InP/Ni alloy.[5]

Figure S5 shows the schematic of the test pattern for the Hall measurement. By measuring the Hall voltage VHV_{\mathrm{H}} under a constant injection current II while varying the applied magnetic flux BB, the carrier concentration can be extracted. Figure S5 shows the Hall measurement result (offset voltage is removed for clarity). VHV_{\mathrm{H}} varies linearly with BB. By fitting the data to the equation

Figure S5. Hall measurement. a Schematic of the test pattern. b Measured results at I=1I = 1 mA.

VH=−BIqNDd(3)V_{\mathrm{H}} = -\frac{BI}{qN_{\mathrm{D}}d} \tag*{(3)}

NDN_{\mathrm{D}} is derived as 3.4×1018 cm−33.4 \times10^{18}\ \mathrm{cm}^{-3}. The electron mobility μ\mu is then derived as μ=1/(qρsND)=1670 cm2/V⋅s\mu= 1/(q\rho_{\mathrm{s}}N_{\mathrm{D}}) = 1670\ \mathrm{cm}^{2}/\mathrm{V} \cdot\mathrm{s}.

Electron mobility model

In order to calculate the doping-concentration-dependent electron mobility of n-InP and n-Si, we employed empirical models. For n-InP, we used the Caughey-Thomas-like model, where the electron mobility is expressed as[6]

μ(ND)=μmin⁡+μmax⁡−μmin⁡1+(NDN0)0.47(4)\mu(N_D) = \mu_{\min} + \frac{\mu_{\max} - \mu_{\min}}{1 + \left(\frac{N_D}{N_0}\right)^{0.47}} \tag*{(4)}

Here, N0=3×1017 cm−3N_0 = 3 \times10^{17}\ \mathrm{cm}^{-3}, μmax⁡=5200 cm2/V⋅s\mu_{\max} = 5200\ \mathrm{cm}^2/\mathrm{V}\cdot\mathrm{s}, and μmin⁡=420 cm2/V⋅s\mu_{\min} = 420\ \mathrm{cm}^2/\mathrm{V}\cdot\mathrm{s}. Figure S6 shows the doping concentration-dependent electron mobility of n-InP obtained from the model and compares it with the experimentally obtained value. Good agreement between the model and the experiment is obtained.

Figure S6. Figure S6. Electron mobility of n-InP. Doping concentration dependence of the electron mobility of n-InP based on the Caughey-Thomas model. The experimentally obtained value agrees well with the model.

For n-Si, we used the Masetti model[7], where the electron mobility is expressed as

μ(ND)=μmin⁡+μmax⁡−μmin⁡1+(NDCr)α−μ11+(NDCs)β.(5)\mu(N_D) = \mu_{\min} + \frac{\mu_{\max} - \mu_{\min}}{1 + \left(\frac{N_D}{C_r}\right)^\alpha} - \frac{\mu_1}{1 + \left(\frac{N_D}{C_s}\right)^\beta}. \tag*{(5)}

Here, Cr=9.68×1016 cm−3C_r = 9.68 \times10^{16}\ \mathrm{cm}^{-3}, Cs=3.43×1020 cm−3C_s = 3.43 \times10^{20}\ \mathrm{cm}^{-3}, α=0.68\alpha= 0.68, β=2\beta= 2, μmax⁡=1471 cm2/V⋅s\mu_{\max} = 1471\ \mathrm{cm}^2/\mathrm{V}\cdot\mathrm{s}, μmin⁡=52.2 cm2/V⋅s\mu_{\min} = 52.2\ \mathrm{cm}^2/\mathrm{V}\cdot\mathrm{s}, and μ1=43.4 cm2/V⋅s\mu_1 = 43.4\ \mathrm{cm}^2/\mathrm{V}\cdot\mathrm{s}.

Bandwidth simulation

Figure S7 shows the simulated 3-dB modulation bandwidth as a function of device size LL. At a device size of 24 μm×24 μm24\,\mu\mathrm{m} \times24\,\mu\mathrm{m}, the 3-dB modulation bandwidth exceeds 100 GHz. Note that the 3 dB bandwidth roughly scales with 1/L21/L^2, where LL denotes the device size (or the bar length). This is because both the resistivity and capacitance of each InP bar scale linearly with LL. The slight deviation from the 1/L21/L^2 trend is attributed to the effect at the edges of the device.

Figure S7. Size dependence of the modulation bandwidth. Simulated 3-dB modulation bandwidth of the device (with ND=1×1019 cm−3N_{\mathrm{D}} = 1 \times10^{19}\,\mathrm{cm}^{-3}) as a function of the device size LL.

Fano resonance fitting

To determine the QQ factor and resonance wavelength of our device, we fitted the measured reflectance spectrum to the Fano resonance model, described as [8]

R(ω)=∣rd+a1γ1γ1+i(ω−ω1)+a2γ2γ2+i(ω−ω2)∣2,(6)R(\omega)=\left|r_d+\frac{a_1\gamma_1}{\gamma_1+i(\omega-\omega_1)}+\frac{a_2\gamma_2}{\gamma_2+i(\omega-\omega_2)}\right|^2, \tag*{(6)}

where the first term rdr_d denotes the direct reflection, the second term corresponds to the high-QQ resonance, the third term corresponds to the low-QQ background resonance, aia_i denotes the amplitude of the i-th resonance, γi\gamma_i denotes the decay rate of the i-th resonance, and ωi\omega_i denotes the resonance frequency of the i-th resonance, respectively. Note that rdr_d is generally complex due to the phase change induced by direct reflection. Figure S8 shows the resonance spectrum of the device at 0 V with the fitting result. The experimental result shows good agreement with the Fano resonance model. Here, ∣rd∣=0.0456|r_d| = 0.0456, arg⁡(rd)=0.7769\arg(r_d) = 0.7769 rad, a1=0.6097a_1 = 0.6097, γ1=6.0936×1012\gamma_1 = 6.0936 \times10^{12} s−1^{-1}, ω1=1.2474×1015\omega_1 = 1.2474 \times10^{15} rad/s, a2=0.2063a_2 = 0.2063, γ2=3.5938×1013\gamma_2 = 3.5938 \times10^{13} s−1^{-1}, and ω2=1.240×1015\omega_2 = 1.240 \times10^{15} rad/s.

Figure S8. Fitting the reflectance spectrum to Fano resonance model. a Reflectance spectrum of the device fitted to the Fano resonance spectrum. b The QQ factor of the device as a function of applied DC voltage. c The resonance wavelength of the device as a function of applied DC voltage.

From the fitting result, the QQ factor can be derived as [9]

Q=ω12γ1.(7)Q=\frac{\omega_1}{2\gamma_1}. \tag*{(7)}

The extracted QQ factor is depicted in Figure S8, which is almost independent on the applied voltage. Figure S8 shows the extracted resonance wavelength λ1=2πc/ω1\lambda_1 = 2\pi c/\omega_1 (cc is the speed of light) as a function of the applied voltage. The resonance wavelength shift per voltage is derived to be 18 pm/V.

Benchmark

Table S1 compares the performance metrics of our device with those of the active metasurfaces experimentally demonstrated in the literature.

Our device experimentally exhibits the record-high 3-dB bandwidth of 17.5 GHz, which is nearly a 6-fold improvement over previously demonstrated active metasurfaces of all kinds reported to date, including plasmonic metasurfaces with metallic electrodes. This is attributed to the low resistivity of n-InP and optimized device geometry to reduce the parasitic capacitance. The excess optical loss (0.56 dB) of our device is also a record low loss among various active metasurfaces reported to date, thanks to the small optical absorption of n-InP.

Active materialModulation schemeResonator materialElectrode materialWavelength (nm)QQ factor3 dB bandwidth f3dBf_{3\mathrm{dB}} (GHz)Optical loss#^{\#} (dB)Efficiency Δλ/V\Delta\lambda/V (pm/V)Reference
OEOPEInPInP151010217.50.5618This work (Experiment)
OEOPEInPInP1550930400.24180This work (Numerical)
OEOPESiSi151075N/Aa^{a}N/A†^{\dagger}16[2]
OEOPESiSi/Au151013103×10−33 \times10^{-3}N/A†^{\dagger}105[10]
OEOPESiAu1540550b550^{b}34.1c4.1^{c}55[11]
OEOPESiAu/ITO13101530.411.4∗11.4^{*}22[12]
OEOPEAuAu1400700.05N/A††^{\dagger\dagger}70[13]
OEOPEAuAu/ITO12802000.19.7∗9.7^{*}28[14]
OEOPEAuAu1290770.0812.9∗12.9^{*}7.4[15]
OEOPEAuAu16501131.2527∗27^{*}40[16]
LNPELNAu7701292.5×10−32.5 \times10^{-3}N/A††^{\dagger\dagger}N/A‡^{\ddagger}[17]
LNPEAlITO/Al/Cr/Au900307.78×10−47.78 \times10^{-4}12.2∗12.2^{*}60[18]
LNPEAuAu900301.35×10−21.35 \times10^{-2}17∗17^{*}60‡‡60^{\ddagger\ddagger}[19]
LNPEAuAu15501800.1256.2∗6.2^{*}40‡‡40^{\ddagger\ddagger}[20]
BTOPEBTOITO6302002×10−32 \times10^{-3}N/A††^{\dagger\dagger}50‡‡50^{\ddagger\ddagger}[21]
SiFCESiSi15503500d3500^{d}N/Ae^{e}N/Af^{f}750g750^{g}[22]
InGaAsPQCSEAlGaAsAu920N/AN/Ah^{h}12.1890[23]

Table S1. Comparison of performance metrics of active EO metasurfaces. In the table, the active material, resonator material, and electrode material refer to the material whose refractive index is modulated, the nanostructured material that functions as the resonator, and the material used to apply signal to the active material, respectively. LN: LiNbO3_3. BTO: BaTiO3_3. ITO: indium tin oxide. PE: Pockels effect. FCE: free carrier effect. QCSE: quantum confined Stark effect.

References

References

  1. [1]Kosugi, Y., Yamada, T., Otomo, A., Nakano, Y. & Tanemura, T. Surface-normal electro-optic-polymer modulator with silicon subwavelength grating. IEICE Electron. Express 13, 20160595 (2016).DOI
  2. [2]Ogasawara, M. et al. Electro-optic polymer surface-normal modulator using silicon high-contrast grating resonator. in Conference on Lasers and Electro-Optics (OSA, Washington, D.C., 2019).
  3. [3]Chang-Hasnain, C. J. & Yang, W. High-contrast gratings for integrated optoelectronics. Adv. Opt. Photon. 4, 379–440 (2012).
  4. [4]Schroder, D. K. Semiconductor Material and Device Characterization (John Wiley & Sons, 2015).
  5. [5]Kim, S. et al. Self-aligned metal source/drain InP n-metal-oxide-semiconductor field-effect transistors using Ni–InP metallic alloy. Appl. Phys. Lett. 98, 243501 (2011).
  6. [6]Sotoodoh, M., Khalid, A. H. & Rezazadeh, A. A. Empirical low-field mobility model for III–V compounds applicable in device simulation codes. J. Appl. Phys. 87, 2890–2900 (2000).
  7. [7]Masetti, G., Severi, M. & Solmi, S. Modeling of carrier mobility against carrier concentration in arsenic-, phosphorus-, and boron-doped silicon. IEEE Trans. Electron Devices 30, 764–769 (1983).
  8. [8]Fan, S., Suh, W. & Joannopoulos, J. D. Temporal coupled-mode theory for the fano resonance in optical resonators. J. Opt. Soc. Am. A Opt. Image Sci. Vis. 20, 569–572 (2003).
  9. [9]Joannopoulos, J. D., Johnson, S. G., Winn, J. N. & Maede, R. D. Photonic Crystals: Molding the Flow of Light - Second Edition (Princeton University Press, 2008).
  10. [10]Zheng, T., Gu, Y., Kwon, H., Roberts, G. & Faraon, A. Dynamic light manipulation via silicon-organic slot metasurfaces. Nat. Commun. 15, 1557 (2024).
  11. [11]Benea-Chelmus, I.-C. et al. Gigahertz free-space electro-optic modulators based on Mie resonances. Nat. Commun. 13, 3170 (2022).
  12. [12]Sun, X. & Qiu, F. Polarization independent high-speed spatial modulators based on an electro-optic polymer and silicon hybrid metasurface. Photonics Res. 10, 2893 (2022).
  13. [13]Benea-Chelmus, I.-C. et al. Electro-optic spatial light modulator from an engineered organic layer. Nat. Commun. 12, 5928 (2021).
  14. [14]Sun, X. et al. Manipulating dual bound states in the continuum for efficient spatial light modulator. Nano Lett. 22, 9982–9989 (2022).
  15. [15]Zhang, L. et al. Plasmonic metafibers electro-optic modulators. Light. Sci. Appl. 12, 198 (2023).DOI
  16. [16]Zhang, J. et al. High-speed metasurface modulator using perfectly absorptive bimodal plasmonic resonance. APL Photonics 8, 121304 (2023).
  17. [17]Weigand, H. et al. Enhanced electro-optic modulation in resonant metasurfaces of lithium niobate. ACS Photonics 8, 3004–3009 (2021).
  18. [18]Weiss, A. et al. Tunable metasurface using thin-film lithium niobate in the telecom regime. ACS Photonics 9, 605–612 (2022).
  19. [19]Damgaard-Carstensen, C. & Bozhevolnyi, S. I. Nonlocal electro-optic metasurfaces for free-space light modulation. Nanophotonics 12, 2953–2962 (2023).
  20. [20]Damgaard-Carstensen, C., Yezekyan, T., Brongersma, M. L. & Bozhevolnyi, S. I. Highly efficient, tunable, electro-optic, reflective metasurfaces based on quasi-bound states in the continuum. ACS Nano 19, 11999–12006 (2025).
  21. [21]Weigand, H. C. et al. Nanoimprinting solution-derived barium titanate for electro-optic metasurfaces. Nano Lett. 24, 5536–5542 (2024).
  22. [22]Qiu, C., Chen, J., Xia, Y. & Xu, Q. Active dielectric antenna on chip for spatial light modulation. Sci. Rep. 2, 855 (2012).
  23. [23]Wu, P. C. et al. Dynamic beam steering with all-dielectric electro-optic III–V multiple-quantum-well metasurfaces. Nat. Commun. 10, 3654 (2019).

Paper details

vlnc:W_8zuyf87x
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents