NEMS-tunable dielectric chiral metasurfaces

Abstract

Active control of strong chiroptical responses in metasurfaces can offer new opportunities for optical polarization engineering. Plasmonic active chiral metasurfaces have been investigated before, but their tunable chiroptical responses is limited due to inherent loss of plasmonic resonances, thus stimulating research in low loss active dielectric chiral metasurfaces. Among diverse tuning methods, electrically tunable dielectric chiral metasurfaces are promising thanks to their potential for on-chip integration. Here, we experimentally demonstrate nano-electromechanically tunable dielectric chiral metasurfaces with reflective circular dichroism (CD). We show a difference between absolute reflection under circulary polarized incident light with orthogonal polarization of over 0.85 in simulation and over 0.45 experimentally. The devices enable continuous control of CD by induced electrostatic forces from 0.45 to 0.01 with an electrical bias of 3V. This work highlights the potential of nano-electromechanically tunable metasurfaces for scalable optical polarization modulators.

I. INTRODUCTION

Chirality is an asymmetric feature describing structures that are not superimposed onto their mirror images. Chiral structures are known to interact differently with light polarized with different handedness and the optical responses that are sensitive to handedness are called chiroptical responses. Although chirality is ubiquitous in various molecules, the chiroptical effects in natural materials are generally very weak, requiring considerable propagation distances to observe the chiroptical effects such as circular dichroism (CD) or circular birefringence. To overcome the limited amount of the natural chiroptical effects, chiral metamaterials or metasurfaces have been investigated in the past two decades [1–3]. First, plasmonic 3D-printed metamaterials or multilayers of patterned metasurfaces have been explored [4–7]. Despite their strong and broadband chiroptical properties, complicated fabrication procedures limit practical applications and extensions to tunable devices. Planar plasmonic chiral metasurfaces with a single patterned layer have also been investigated. In particular, patterned plasmonic chiral metasurface on top of a flat back metallic mirror has achieved high circular dichroism [8–10]. Unlike conventional metallic mirrors which reflect circular polarized light with reversal of handedness, these plasmonic mirrors selectively reflect one circularly polarized light without changing the handedness while the other circular polarization is absorbed [8–10]. These polarization selective mirrors are often called chiral spin-preserving mirrors and have potential applications in valley exciton-polaritonics [11]. Moreover, the plasmonic chiral metasurfaces can support broadband chiroptcial responses and large CD in absorption. Complementary to plasmonic metastructures, all-dielectric chiral metasurfaces also have shown strong chiroptical effects [12–14]. In particular, it has been recently demonstrated that a single-layer dielectric metasurface is able to realize near-unity CD in reflection [14]. Unlike the aforementioned plasmonic chiral structures with back metallic mirrors [8–10], the single-layer dielectric chiral metasurface selectively reflects one circular polarized light with preserved handedness and transmits the other circular polarized light with flip of handedness [14].

Dynamical control of the chiroptical effects in metamaterials can boost development of devices for novel polarization control. During the last decade, reconfigurable plasmonic chiral metastructures have been extensively investigated in THz and microwave domain [15–18]. For example, diverse active platforms using optical tuning [15], global mechanical deformation [16], micro-electromechanical systems [17], and electrical gating of graphene [18] have been proposed. In addition to devices working in THz or microwave domain, dynamic plasmonic chiral metasurfaces have been also extensively studied in the optical domain. For example, all-optical tuning of phase change materials [19] or DNA structures [20] has enabled switchable chiroptical responses. Also, global environmental tuning of liquid [21], pH [22], strain [23], and magnetic field [24] have been explored. However, these all-optical or global tuning methods often require complicated setups hindering the on-chip integration. Thus, electrical control of the chirality can be more attractive than other methods for practical applications. Recently, the nano-electromechanically tunable chiral plasmonic metasurfaces have been demonstrated, thus enabling active control of polarization and chiroptical responses through nano-electromechanical actuation in vertical direction [25, 26]. Nevertheless, all of the aforementioned tunable plasmonic chiral metasurfaces are inherently lossy which limit optical performance.

In contrast to the extensive works on tunable plasmonic chiral metasurfaces, investigations related to tunable dielectric chiral metasurfaces have been limited. For instance, optothermal moving of nanoparticles [27] and all-optical tuning of Si nonlinearity [28] have enabled tunable chiroptical responses. However, both systems still require additional optical systems to modulate the device optically, imposing considerable limits on compact integration [27, 28]. More broadly, optomechanical GaAs chiral metasurfaces have been demonstrated using excitation of global mechanical oscillation of the membrane through a bulky piezoelectric actuator [29]. However, the system has not just shown weak tunable response mainly due to the limited movement of the membrane, but also only enabled the oscillation of the output signal [29]. Therefore, to the best of our knowledge, electrical tuning of dielectric chiral metasurfaces has not been explored and can be a promising alternative to that of the plasmonic chiral metasurfaces thanks to the low-loss optical property of dielectric materials. Furthermore, from a perspective of devices, scalable material platforms such as silicon and low electrical bias within CMOS logic level can be important for on-chip integration. In this work, we experimentally demonstrate nano-electromechanically tunable dielectric chiral metasurfaces in the telecom wavelength, fabricated by using standard silicon-on-insulator platforms. The metasurfaces resonantly work as chiral spin-preserving mirrors, exhibiting selective reflection for one circular polarization without external bias. Furthermore, the resonances hosted by the metasurfaces can be continuously tuned by electrostatic forces using voltages less than 3V. In particular, the devices experimentally achieve transition of CD in reflection from 0.45 to 0.01 at the resonant wavelength.

II. MAIN RESULTS

Figure 1 shows the conceptual illustrations of the proposed nano-electromechanically tunable chiral metasurface. In Fig. 1a, the illustrative schematic of the top view of the suspended silicon chiral metasurface is displayed. The metasurface consists of two sets of pairs of doped silicon nanostructures and an electrode is deposited on each set for electrical bias. In Figs. 1a and 1b, different colors are used to explicitly visualize the two sets of the nanostructures. Throughout this paper, all devices are composed of 675 nm thick and 45 μm long silicon nanostructures. The design parameters are shown in Fig. 1b. Period, p, is chosen to be 700 nm so that the period of the pair is smaller than the wavelength of interest to avoid unwanted diffraction under normal incidence. As shown in Fig. 1b, we intentionally break n-fold rotational symmetry for n>2 and any in-plane mirror symmetry to achieve strong chiroptical effects in reflection. It is worth noting here that this design approach has been extensively used for single layer dielectric chiral metasurfaces [14] and plasmonic chiral metasurfaces having back metallic reflectors [8–10]. When electrical bias is applied, all pairs of the neighboring nanostructures that are connected to the different electrodes have voltage difference and become capacitors. Thus, the induced electrostatic forces between the nanostructures enable continuous mechanical actuation as a function of the external bias. In other words, g₁ (g₂) decreases (increases) by applying the bias, where g₁ (g₂) is the gap size between the two neighboring bars in the different (same) set. In Fig. 1c and Fig. 1d, optical functions of the metasurface are schematically illustrated. In Fig. 1c, the structure without the actuation reflects right circular polarized (RCP) light without flip of the handedness while left circular polarized (LCP) light is transmitted with reversal of handedness. When the bias is applied, the suspended nanostructures are actuated and their chiroptical properties are continuously modulated. As shown in Fig. 1d, the metasurface can exhibit negligible chiroptical responses with the actuation so its co- and cross-polarized reflection and transmission become identical for LCP and RCP illuminations at the target wavelength.

FIG. 1. Nano-electromechanically tunable all-dielectric chiral metasurfaces.

a Schematic illustration of a top view of the metasurface. The metasurface is composed of two sets of doped silicon nanostructures. Anchors and the gold electrodes are marked. Electrodes are deposited on each silicon layer for mechanical actuation. b Schematic illustration of two pairs of the nanostructures constituting the metasurface. Geometric parameter definitions are shown in the illustration. c and d Illustrations of reflection behavior of the metasurfaces without and with external bias. c: Without external bias, the metasurface selectively reflects RCP light by keeping the handedness and transmits LCP light with change of handedness. d With actuation, the metasurface becomes achiral. The amounts of co- and cross-polarized reflection and transmission become symmetric for both RCP and LCP input lights. In a-d, two different colors are employed to distinguish two sets of the nanostructures and visualize voltage difference between two sets. As shown in b, pink and red colors represent ground, GND, and applied bias, V₀, respectively.

First, the proposed metasurfaces are numerically investigated using a commercial software based on the finite element method, COMSOL^® (see Methods for details). The polarization analysis of reflection from the metasurface is plotted in Fig 2a. The reflection coefficient R_L,L (R_R,R) is defined as the reflection of the LCP (RCP) component from the metasurfaces to the LCP (RCP) input light. Similarly, R_R,L (R_L,R) is defined as the reflection of the RCP (LCP) component from the metasurfaces to the LCP (RCP) input light. In Fig. 2a, the polarization-sensitive reflection is observed near the resonant wavelength of 1478 nm. The metasurface selectively reflects RCP light without flip of the handedness, while the LCP light is mostly transmitted. Specifically, R_L,L, R_R,R, and CD in reflection, defined by |R_L,L-R_R,R| in this paper, are 0.04, 0.89, and 0.85, respectively. Moreover, R_R,L and R_L,R are identical due to the symmetry of the unit cell [30] and it is confirmed by the grey curve plotted in Fig.2a. In Fig. 2b, electric field profiles in the x – y plane at the middle of the nanostructures (i.e. 337.5 nm above from the bottom) are plotted under LCP and RCP illuminations, showing that the metasurfaces interact with LCP and RCP lights differently at the resonance. The chiroptical responses in Fig. 2a and the two distinct electric field profiles in Fig. 2b can be qualitatively explained by the spectral overlap of two leaky guided mode resonances hosted by the dielectric metasurfaces [14]. For example, the input polarization state determines the amplitude and phase of two leaky guided modes and the interference between the resonant modes results in the two distinct field profiles shown in Fig. 2b. The input polarization state also affects the phase and amplitude of the radiations from the two leaky guided modes, so the radiations from the guided modes interfere differently with directly transmitted or reflected light. Therefore, the reflection spectra shown in Fig. 2a highly depend on the handedness of the input polarization.

FIG. 2. Simulated mechanically tunable chiroptical responses.

a Simulated reflection spectra of co- and cross-polarized components under RCP and LCP illuminations. The spectra of R_L,L and R_R,R are plotted by dashed and solid black lines, respectively. R_R,L and R_L,R are identical and plotted together by a grey line. Geometric parameters used in the simulation: p = 700 nm, w = 505nm, g₁ = g₂ = 195 nm, w_p = 85nm, and l_p = 265nm. b Electric field distribution cuts from a middle plane of the nanostructure. The magnitude of the field profiles are plotted under LCP (left) and RCP (right) illuminations at the resonance wavelength of 1478 nm. c Calculated reflection spectra of R_L,L and R_R,R with mechanical movements. The spectra of R_L,L and R_R,R are plotted in dashed and solid lines, respectively. The mechanical displacement, expressed by $\frac{g1-g2}{2}$ , varies from 0 nm to 80 nm. The value of $\frac{g1-g2}{2}$ for each color is shown in the legend. d Spectra of circular dichroism in reflection, |R_L,L – R_R,R|, for the different mechanical displacement. The spectra are calculated from c. The value of $\frac{g1-g2}{2}$ for each color is shown in the legend.

To simulate the mechanical tuning of the chiroptical effects, the spectra of R_L,L and R_R,R are calculated by varying the nanobeam displacement, $\frac{g2-g1}{2}$, from 0 to 80 nm and plotted in Fig. 2c. In Fig. 2c, an increase of $\frac{g2-g1}{2}$ causes red shift of peaks in the spectra of R_R,R and blue shift of dips in the spectra of R_L,L. Furthermore, the amount of the spectral shift of the R_L,L is larger than that of R_R,R. To visualize the change of the chiroptical effects clearly, |R_L,L – R_R,R| is evaluated from Fig. 2c and plotted in Fig. 2d. In Fig. 2d, it is clearly shown that the mechanical displacement in lateral direction causes the strong change of |R_L,L – R_R,R| around the resonant wavelength of 1478 nm in Fig.2d. The 80 nm displacement leads to a change of |R_L,L – R_R,R| from 0.85 to 8×10⁻⁴ at 1478 nm in Fig. 2d. Namely, the mechanical actuation in the lateral direction results in the transition from strong to negligible chiroptical response even with the presence of chirality in the structure.

To experimentally verify nano-electromechanically tunable chiroptical responses, the devices are fabricated using a conventional nanofabrication process for silicon-on-insulator substrates (see Methods for details). We should mention here that g₁ and g₂ are adjusted in the fabrication process such that g₁ is 60 nm smaller than g₂. This adjustment allows for a large shift of the resonance with nano-electromechanical tuning of the gaps. Figure 3a shows scanning electron microscopy images, confirming good agreement with the illustrative schematics in Figs. 1a and 1b. To characterize chiroptical responses of the devices, spectra of R_L,L and R_R,R are measured using the setup in Supporting Fig. 1. The measured spectra shown in this paper are normalized by the reflection from 65 nm gold electrode in order to estimate the absolute reflection efficiency and remove fluctuations resulting from variations in polarization states of the input tunable laser (see Methods and Supporting Figure 1 for details about measurement procedures). Two different devices are measured and the corresponding spectra of R_L,L and R_R,R are plotted in Figs. 3b and 3c (see Supporting Table 1 for the measured design parameters of the devices). In Fig. 3b, the dip in the spectrum of R_L,L and the peak in the spectrum of R_R,R are overlapped at ~1475 nm, showing good agreement with the shape of the simulated spectra in Fig. 2a. However, the dips of the spectra of R_L,L and R_R,R in Fig. 3c are overlapped at ~1492 nm. This deviation mainly results from the non-perfect match of the design parameters. Specifically, the value of l_p is slightly larger than its optimal value (see Supporting Figure 2 for details). The spectra in Figs. 3b and 3c show maximal |R_L,L – R_R,R| of 0.45 and 0.37 at the resonance wavelengths of 1475 nm and 1493 nm, respectively. Even with a few percentage reflection loss of the gold electrode used for the normalization, the maximal values calculated from Figs. 3b and 3c are still higher than 0.44 and 0.36, respectively. On the other hand, the ratios between R_L,L and R_R,R reach 3.58:1 and 8.19:1 in Figs. 3b and 3c, respectively. The large ratio between R_L,L and R_R,R directly indicates the potential for electromechanically tunable circular polarization filters. The measured values of |R_L,L – R_R,R| are smaller than the simulated value shown in Fig. 2a. We believe that the deviation results from the finite length of the resonators, which may cause limited coupling between the resonance and the input light. Furthermore, the imperfect fabrication and the high sensitivity of the design shown in Fig. S2 possibly lead to the difference between the measurement and the simulation. However, we should note here that optimized single-layer dielectric metasurfaces are able to reach near unity CD in reflection [14].

FIG. 3. Dielectric chiral metasurfaces and measurements of chiroptical responses in reflection.

a Scanning electron microscope images of the fabricated metasurfaces. Left: An array of the nanostructures. Right: Zoom-in scanning electron microscope image of the 2 pairs of the nanostructures. Scale bars in left and right denote 5 μm and 1 μm, respectively. b and c Measured reflection spectra of R_L,L, and R_R,R for two different structures (see Supporting Table 1 for the measured design parameters). The spectra of R_L,L and R_R,R are plotted in dashed and solid lines, respectively.

To demonstrate nano-electromechanical tuning of the chiroptical responses, static electrical bias is applied to the electrodes and the induced changes in the optical reflection are characterized. The spectra of |R_L,L – R_R,R| are measured under several electrical biases and plotted in Figs. 4a and 4b (see Supporting Figure 3 for the measured spectra of R_L,L and R_R,R). The devices used in Figs. 3b and 3c are used for Figs. 4a and 4b, respectively. In both spectra, the CD in reflection is varied from the maximum value to nearly zero. For example, an external bias of 2.75 V (2.8 V) causes a change in CD from 0.45 (0.37) to 0.01 (1×10⁻⁴) in Fig. 4a (Fig. 4b). The required electrical bias for the maximal change of the chiroptical response is smaller than 3V, which is already within CMOS logic level. Also, the static bias causes blue-shifts of the main peaks of the |R_L,L – R_R,R|, which mainly results from the dominant blue-shift of R_L,L shown in Fig. 2c. The peak shifts of |R_L,L – R_R,R| shown in Fig. 4a and Fig. 4b are as large as −2 nm and −6 nm under the bias of 2.75 V and 2.8 V, respectively. The large spectral shifts up to 6 nm indicate that the low-Q leaky guided resonances are sufficient to achieve considerable chiroptical tunablity. The large spectral shift is advantageous in terms of bandwidth and robustness, which are important from practical considerations. Furthermore, the measured spectral shift is the lower bound of the limit, as the induced voltage up to 2.8V is smaller than the pull-in voltage.

FIG. 4. Nano-electromechanical tuning of chiroptical responses.

a and b Measured circular dichroism in reflection, |R_L,L – R_R,R|, under different external biases. The devices exploited in Fig. 3b and Fig. 3c are measured and plotted in a and b, respectively. The applied bias for each color is shown in legends. c Measured temporal response of the metasurfaces. Top: Input square wave signal of which duty cycle, frequency, and amplitude are 0.5, 100 Hz, and 2V, respectively. Bottom: Measured output signals of R_L,L by a photodetector. Raw and filtered reflection signals are plotted by grey and black curves, respectively.

Finally, the dynamic response of the chiral metasurfaces is investigated in air. A periodic square-wave signal with a modulation frequency of 100 Hz, amplitude of 2V, and duty cycle of 50% is applied (see Methods for details). The device used for Fig. 4b is measured with the input light at the wavelength of 1493 nm. The input electrical signals and the measured raw signals of R_L,L are plotted in Fig. 4c. It clearly demonstrates that the optical responses are fully reconfigurable. In Fig. 4c, the measured rise time (up to 90% power) and fall time (down to 10% power) are 1.48 ms and 460 μs, respectively. We believe that the low speed mainly results from the low-doping density of the Si layer of the silicon-on-insulator wafer.

III. DISCUSSION AND SUMMARY

Many aspects of the proposed devices can be improved with further investigations. First of all, the modulation depths of CD in reflection shown in Figs. 4a and 4b are not limited by the spectral shifts but by the measured maximal CD in Figs. 3b and 3c. Considering the low loss of the dielectric metasurface, a large sweep of the design parameter spaces may result in near unity CD in reflection [14]. Furthermore, inverse-design of the tunable chiral metasurfaces can be of interest [31]. The inverse-design may not only improve CD in reflection, but also find other guided modes that are more robust against imperfect fabrication or more efficient for nano-electromechanical tuning. Specifically, the inverse design may find two leaky guided mode resonances which similarly react to variations in design parameters. Thus, the large chiroptical responses resulting from the interference between two modes possibly become tolerant against imperfect fabrication. For efficient nano-electromechanical tuning in lateral direction, the optical modes can be optimized to store considerable electromagnetic energy at the gap between the nanostructures instead of inside of the nanostructures. In addition to the optical aspects, considering the scale of the devices, we envision that high switching speed up to a few MHz is achievable with a highly-doped Si layer, co-optimization from both mechanical and optical perspectives, and proper packaging [32, 33]. With mechanical resonances supported by the metasurfaces, the proposed devices might provide efficient electromechanical platforms for polarization controlled optomechanical transduction [29].

In conclusion, we demonstrate nano-electromechanical tuning of the suspended silicon chiral metasurfaces. With an external bias below 3V, the devices experimentally achieve continuous tuning of CD in reflection from 0.45 to 0. This work paves the way of nano-electromechanically tunable dielectric chiral metasurfaces towards scalable and novel optical modulators, which can be used in diverse applications such as dynamic polarization engineering, stereoscopy, valleytronics, polarization optomechanics, and chiral sensing.

IV. METHODS

Simulation and design

The reflected spectra of the metasurface were calculated using a commercial software based on the finite element method, COMSOL^®. Assuming an infinite periodic array, the 3D silicon structure is simulated in air with normal incident light. The design parameters used in the simulation are shown in Fig. 2.

Device fabrication

The devices are fabricated using a silicon-on-insulator SOI wafer. The detailed fabrication process can be found in Ref. [34]. We use a wafer having a device layer of 675 nm and a buffered oxide layer of 3 μm on a 1 mm thick silicon wafer. The fabrication includes two sequential e-beam lithography steps. First, the metasurfaces were fabricated with e-beam lithography and reactive ion etching. Subsequently, the electrodes are defined by the e-beam lithography, e-beam deposition of 65 nm gold/5 nm chromium layers, and lift-off process. Buffered hydrofluoric acid was used to etch the buffered oxide layer under the silicon layer. The time of the under-cut process is carefully controlled so that the metasurfaces are fully suspended while the anchors are supported by the SiO₂. In other words, one end of every suspended nanostructure is connected to anchors and the other end is connected to the silicon layer. That is because the connections on both sides prevent breaking during the under-cut. To prevent destruction of the suspended device, the device is dried by a critical point dryer after the under-cut process. The device is bonded to a custom printed circuit board using a wire bonder (WestBond 7476D).

Measurement procedure

All of the reflection spectra presented in this paper are characterized using the set-ups shown schematically in Supporting Fig. 1. We use a tunable laser (Photonetics, TUNICS-Plus) as the light source and the wavelength of the light is tuned from 1450 nm to 1580 nm. A beam splitter is placed in front of the fiber collimator (Thorlabs, F260FC-1550) to capture the power from the source and send the light to the sample. For reference, the power from the source is captured by a InGaAs detector (Thorlabs, PDA10CS). A linear polarizer and a quarter waveplate (QWP) are inserted between a polarization beam splitter and a 20× infinity-corrected objective lens (Mitutoyo, M Plan Apo NIR) to set the polarized state of the incident light. The QWP is mounted on a rotation stage to set the input polarization state to LCP or RCP. The sample at the object plane is imaged by the objective lens and a tube lens with a focal length of 200 mm. At the image plane, a pinhole with a diameter of 400 μm is inserted to select a region of interest with a diameter of 20 μm in the object plane. The spatially filtered light was simultaneously focused onto another InGaAs detector for the measurement of the spectra, or imaged on an InGaAs SWIR camera (Goodrich, SU320HX-1.7RT) using relay optics. All spectra in this paper were obtained by dividing the signal from the sample by the signal from the sources. To estimate absolute reflection and remove fluctuation resulting from variation in polarization states of the input laser, the spectra are normalized by the reflection from the 65 nm gold electrode. Considering that the reflection of the 65 nm of gold layer is ~98% in simulation, the actual reflection can be a few percentages smaller than the values presented in Figs. 3 and 4. For the measured dynamic responses shown in Fig. 4, we use a function generator (FeelTech, FY6600-60M).

DATA AVAILABILITY

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