Terahertz (THz) waves occupy a frequency range between microwave and infrared radiation and have garnered significant research interest in recent years due to their distinct properties and the wealth of spectral information^[1]. As a result, THz technology has found diverse applications in areas such as spectroscopy^[2], imaging^[3], sensing^[4], and communications^[5]. Efficient manipulation of THz waves is crucial for the advancement of THz technology. However, the progress in THz manipulation still faces substantial challenges due to the limited availability of natural materials with high response for THz waves. Recently, scientific and technological interest in metasurfaces^[6–11], which are composed of artificial subwavelength structures, has been gradually increasing for their ability to manipulate electromagnetic waves and facilitate the design of various desired devices, for example, lenses^[12], filters^[13], deflectors^[14], absorbers^[15], and vortex beam generators^[16]. Among them, effectively manipulating the polarization state of electromagnetic waves provides more flexibility and functionality for various applications^[17–22]. A linear polarization (LP) converter^[23–27] effectively rotates the polarization direction of incident linear electromagnetic waves by 90°, thereby expanding the potential applications of electromagnetic waves in various scenarios, from consumer products to high-tech applications.

However, most devices are predominantly static, which limits the utilization that can be achieved. Therefore, it is important to explore a dynamically controlled metasurface. Such a design offers a high degree of flexibility and enables tailored functionalities, thus addressing the diverse requirements of modern optics^[28–31].

Presently, one of the active mechanisms focuses on customizing the geometries or arrangements of meta-atoms through mechanical actuation^[32–45], including stretching, translating, and rotating. In the last few decades, various relevant applications have gained enormous interest, both theoretically and experimentally, in a broad range of electromagnetic spectra, from microwaves to visible light. Due to the reconfigurability of its structural shape and position, a mechanically active metasurface can be flexibly adjusted according to functional requirements for communication, radar, and sensing applications. Compared to electronic modulation technologies, mechanical metasurfaces typically have a lower energy consumption because their regulating components do not usually require continuous energy consumption to maintain their state, reflecting their nonvolatile nature. In addition, due to the faster motion response of the mechanical components, mechanical metasurfaces can quickly respond to changes in the external environment. Recent research has focused on designing and assembling standard mechanical and optical components while optimizing sample processing and testing protocols to enhance adjustment accuracy, accommodating smaller design scales, and ensuring increased reliability. On the basis of the advantages of mechanical modulation, precise spectrally reconfigurable metasurfaces have been fabricated and applied as THz LP converters. For example, Fan et al. demonstrated a freely tunable polarization rotator for broadband THz waves using a three-rotating-layer metallic grating structure with nearly perfect conversion efficiency^[32]. You et al. presented a three-layer metasurface-based planar circular polarizer exhibiting strong and broadband chirality from 0.22 to 0.32 THz^[40]. However, the above two works involve metasurfaces with a three-layer structure, which increases the complexity and cost of mechanical control to a certain extent while introducing higher losses.

To address the limitations associated with existing designs, here we propose a transmissive metasurface design based on gap-tuning in the THz regime. The basic meta-atom consists of two air-gap-separated plasmonic structures with identical patterns rotated by 90^o, intended for operation with orthogonally polarized channels. By precisely adjusting the air-gap spacing by a subwavelength distance from 150 to 800 µm, we achieve an efficient bandwidth-tuned LP converter, whose transmission spectrum undergoes the whole process from dual-frequency conversion to single-frequency conversion and then to single-frequency efficiency modulation. Our device demonstrates a maximum cross-polarization amplitude transmission of 0.87 at the operation frequency of 0.115 THz with a near-unity polarization conversion ratio (PCR) when the air-gap spacing is optimal. A sample is fabricated via thin-film lithography and characterized using a vector network analyzer (VNA)^[46,47] with a homemade sample holder for stretching. This mechanism relies on designing the interactions between localized resonances of multiple structures and adjusting the Fabry–Perot (FP) resonance of the gap. Such strong interactions facilitate flexible and effective modulation at various gap distances, achieved through subwavelength-scale movements. Essentially, this platform enables the realization of gap-tunable wavefront functionalities with significant versatility.

As illustrated in Fig. 1(a), the proposed tunable transmissive metasurface polarizer consists of two periodic metallic arrays printed separately on two dielectric film substrates. A gap between the bottom and top layers is established, generating an air cavity. Both resonators possess identical thicknesses of the dielectric layers $t_1=t_4=50\mathrm{\mu m}$ and the metallic layers $t_2=t_3=200\mathrm{nm}$. Precise and independent control over the top and bottom layers facilitates their upward and downward movements toward/away from each other. The dimension of the air cavity $d$, separating the two metallic resonators, can be altered and continuously tuned by external stimuli, thereby enabling mechanically tunable metasurfaces.

It is well understood that the design and arrangement of the subwavelength unit cell structures play an essential role in the realization of desired electromagnetic responses. A unit cell comprising a split-ring slit (SRS) structure is selected to build the metallic patterns, as illustrated in Fig. 1(b). The top and bottom metallic layers share identical patterns, but the bottom layer is rotated by 90° around the $z$-axis relative to the top layer. The SRS structures on the top and bottom layers exhibit orthogonal polarization selection characteristics, representing bright and dark modes, respectively. Aluminum (Al) is chosen for its cost-effectiveness and high absorption, being a suitable material for metal components. Meanwhile, a cyclic olefin copolymer (COC) film, characterized by extremely low loss and good mechanical stability at THz frequencies, is selected as the dielectric spacer, further enhancing the performance of the system in the THz regime.

Similar to typical subwavelength slit resonators, the SRS can be stimulated by an electric component perpendicular to its base arm, inducing an electric resonance. The resonance frequency is determined mainly by the overall slit length. On the other hand, the resonance fields are predominantly localized at the base arm. Consequently, by optimizing both the base and side arm lengths, one can simultaneously adjust the resonance frequency and near-field coupling of the coupled SRS structures, thereby achieving high efficiency. Figure 1(c) shows the physically realized unit cell geometry of the proposed LP converter and presents the detailed physical dimensions of the SRS pattern, where $a=509\mathrm{\mu m}$, $b=509\mathrm{\mu m}$, $w=157\mathrm{\mu m}$, $P_x=1600\mathrm{\mu m}$, and $P_y=1600\mathrm{\mu m}$.

The three-dimensional full-wave simulation software CST Microwave Studio is used for numerical analysis. In the simulation, the lossy metal Al is loaded from the CST material library (${\sigma }_{\mathrm{Al}}=3.56\times {10}^7\mathrm{S}/\mathrm{m}$), and the dielectric spacer is COC with a relative permittivity of 2.32 and a loss tangent of 0.0009 over the frequency range of interest, as determined experimentally using photoconductive-antenna-based THz time-domain spectroscopy. Due to the periodicity of the metasurfaces, the entire structure can be analyzed effectively by simulating a unit cell, while retaining comprehensive design information and optimizing time and computational resources. Periodic boundary conditions were applied along the $x$- and $y$- directions, while an open boundary condition was set along the $z$-direction. The source was defined as a normally incident plane wave with an electric field aligned along the $x$-axis and propagating along the $z$-axis from an $xy$-plane situated 1500 µm above the top layer. A receiver was positioned at the center of the $xy$-plane at 6000 µm above the bottom layer to capture transmission signals.

The incident Ex-wave was linearly polarized and normal to the sample surface, facilitating direct excitation of the SRS patterns on the top layer by the incident electric field. In contrast, the SRS patterns on the bottom layer remained inactive. For all simulation results in our paper, the probe signal parameters, $E_x$- and $E_y$-fields, were utilized to ascertain the amplitude transmissions of the co- and cross-polarization components, respectively. To model the effect of the gap modulation, position-movable top and bottom layers were incorporated into the model with micron-level adjustment capability and without the inclusion of mechanical control components.

To better understand the polarization conversion of the proposed mechanically tunable metasurfaces, we define $t_{xx}=E_{xt}/E_{xi}$ and $t_{yx}=E_{yt}/E_{xi}$ to represent the transmissions of $x\text{-}\mathrm{to}\text{-}x$ and $x\text{-}\mathrm{to}\text{-}y$ polarization conversions, respectively. The subscripts $i$ and $t$ indicate the incident and transmitted waves, respectively. The subscripts $x$ and $y$ denote the polarization components of the electromagnetic waves. Thus, in the above two expressions, the former represents co-polarization and the latter cross-polarization. We use the PCR to describe the performance of frequency-dependent polarization conversion for linearly polarized waves. The PCR for an $x$-polarized incident plane wave can be defined as $$\mathrm{PCR}=\frac{{|t_{yx}|}^2}{({|t_{yx}|}^2+{|t_{xx}|}^2)}.$$(1)

Figures 2(a) and 2(b) illustrate the simulated amplitude transmissions of the cross- and co-polarizations as functions of $d$ from 150 to 800 µm under $x$-polarized normal incidence within the frequency range from 0.1 to 0.13 THz, respectively. For air gaps below 460 µm, a splitting phenomenon in amplitude occurs at the center frequency of 0.115 THz. At this time, the system is over-coupled, and efficient dual-frequency orthogonal polarization conversion distributes on both sides of the center frequency. As the spacing increases, this splitting phenomenon gradually diminishes. For spacings greater than 460 µm, the curves display only efficient single-frequency orthogonal polarization transition, which then decreases as the spacing continues to increase. Hence, we designate the optimal value of $d$ as 460 µm, and this air-gap distance is associated with the maximum cross-polarization amplitude transmission (0.874) at the center frequency. The distribution of co-polarization transmission exhibits similar characteristics, albeit with significantly lower values across the entire spectrum, remaining below 0.1 at the optimal $d$ value.

The PCR map exhibits a quasi-symmetrical distribution around the center frequency of 0.115 THz. At 0.115 THz, the PCR remains consistently high (over 0.92) regardless of variations in the air-gap spacing, as depicted in Fig. 2(c). The frequency range exhibiting high PCR levels gradually narrows as the spacing increases. The bandwidth of cross-polarization conversion decreases gradually from broadband to narrowband (approximately 25 to less than 5 GHz) with increasing spacing, as illustrated in Fig. 2(d). When the gap interval is less than the optimal value, due to the phenomenon of efficient dual-frequency orthogonal polarization conversion, the efficiency at 0.115 THz is split and instead forms a local minimum because the system is over-coupled. The transmission at 0.115 THz reaches a peak of 0.874 as the air gap d increases to the optimal value of 460 µm and subsequently decreases to 0.38 when the spacing further increases to 800 µm.

Hence, our proposed mechanically tunable metasurfaces establish a spacing-variable FP air cavity, modulating the operation efficiency and bandwidth of polarization conversion. For larger air gaps, there is increasingly stronger coupling/cross-talk due to scattering and multiple reflections in the FP cavity, resulting in a gradual decrease in the metasurface efficiency. Compared with the inductive–capacitive resonance formed by the SRS metallic array, the FP resonance generated by the cavity composed of two resonators has a broader bandwidth. As the gap spacing increases, the FP resonance decreases much more than the inductive–capacitive resonance, so the entire bandwidth gradually decreases. Furthermore, it showcases the capability to convert the incident linearly polarized wave into its orthogonal component in the transmitted wave, achieving a high PCR at the center frequency and maintaining this level within a 650 µm range of variation in the air-gap spacing. This tunability enables the device to accommodate a range of practical applications.

To gain further insight into the influence of the air-gap distance on the performance of the designed LP converter based on this theory, we simulated the orthogonally polarized $E_y$-field amplitude ($|E_y|$) distributions in the top and bottom layers at the center frequency of 0.115 THz. Moreover, for a given thickness of the dielectric spacer, the near-field coupling can only be tuned by changing the relative positions of the bright and dark resonances. The air-gap distances $d$ are chosen to be 260, 460, and 660 µm, which represent gap spacings smaller than, equal to, and greater than the optimal $d$ value, respectively. The incident wave impinges on the top layer and transmits through the bottom layer. It can be observed in Fig. 3 that the resonance modes around the bottom layer interacting with the top layer should be responsible for the further enhancement of the localized electric field in the bottom layer across all $d$ values, albeit to varying extents. For $d=260\mathrm{\mu m}$, the localized electric field energy is mainly concentrated in the bottom layer, while the energy in the top layer is significantly suppressed, indicating the strongest interaction and coupling between the two metallic layers in this case. In contrast, for $d=460\mathrm{\mu m}$, both the top and bottom layers exhibit considerable localized electric field energy, leading to pronounced out-of-plane resonance with significant interlayer coupling between the two metallic planes and thus resulting in a prominent peak in the spectral response. Conversely, for $d=660\mathrm{\mu m}$, the localized electric field energy resides primarily in the top layer rather than the bottom layer, suggesting weak resonance involving a single metallic plane due to the scattering of the incident electromagnetic wave by the periodically arranged array, which dominates in this scenario.

The sample, comprising two metallic structures, underwent a sequential fabrication process. Initially, a 50-µm-thick COC substrate was affixed to a 1-mm-thick silicon wafer, which served as a rigid holder. Subsequently, a thin layer of approximately 3-µm-thick photoresist (PR4000) was spin-coated onto the COC substrate. Following this, UV lithography and image development were utilized to pattern the photoresist through a pre-prepared photomask and a developing solution (RZX3038). Afterward, a plasma degluing machine was employed to eliminate a uniform thin photoresist layer from the sample surface. This ensured the complete removal of the photoresist in the structural area while preserving the sharp gradient of the photoresist at the edge between the structural and non-structural regions. After this step, a 200-nm-thick layer of Al was thermally evaporated onto the top layer of COC. Following deposition, the Al layer was selectively removed using an acetone solution, leading to the formation of patterned structures. Last, the COC film was stripped from the silicon wafer. Subsequently, the aforementioned steps were reiterated to create another metallic layer on another COC film. The sample was obtained according to the above fabrication steps, and the macroscopic and microscopic images are given in Figs. 4(a) and 4(b), respectively. [See Fig. S1(a) of the Supplementary Document for the flow chart of the sample fabrication procedure.]

To confirm the numerically simulated results mentioned earlier and further validate the performance of the device, experimental characterization of the sample was conducted. As illustrated in Fig. 4(c), a VNA (Ceyear 3672B, Keysight) together with two frequency doubling modules, two waveguides (WR8), and two antenna horns was employed to investigate the performance of the manufactured LP converter within the frequency range from 0.1 to 0.13 THz. The VNA transmitted a 20 GHz microwave signal through a frequency doubling module, and the signal was subsequently converted into a THz wave within the 90–140 GHz range. The waveguides were then fixed to the horn antennas using flanges, facilitating conversion of the incident signal from a waveguide mode to a free-space light mode and effective coupling to the sample. To capture a well-defined orthogonally polarized signal, a twisted waveguide was fixed in front of the receiving port. Consequently, the transmitted wave underwent a 90° polarization rotation, while the co-polarization wave experienced significant attenuation. Additionally, absorbers shielded the relevant modules to eliminate spurious reflections.

The design of the sample size needs meticulous consideration, incorporating both processing operability and experimental accuracy. Variations in the sample size are expected to exert a favorable influence on the test outcomes. While the peak intensity of the outgoing wave typically occurs at the center of the antenna horn, referred to as the main beam position, it should be noted that the substantial aperture of the antenna horn within the test system leads to non-negligible radiation in the peripheral regions. Consequently, despite a major part of the outgoing wave making contact with the sample, the metallic nature of the sample holder ensures that the outgoing wave generates standing waves in the edge region when coming into contact with the holder. Expanding the sample size mitigates this interaction, thereby reducing the adverse effects on measurement precision. However, enlarging the sample size concurrently introduces complexities in the design and fabrication processes. Consequently, the processed metasurface sample adopts a circular shape with a 48 mm diameter, corresponding to approximately $30\lambda \times 30\lambda$ periodic arrays.

Figure 4(d) depicts the sample stretching device, comprising two 2D displacement platforms, two lifting platforms, and two 3-inch beam-splitting prisms arranged in a bottom-up configuration, all affixed onto one solid Al optical breadboard and utilizing identical standard parts [see Fig. S1(b) of the Supplementary Document for the photo of the measurement system]. The precision of the 2D displacement platforms is 10 µm, facilitating linear and precise adjustment of the air-gap spacing by the same margin. In the design of the mask, cross-alignment marks are incorporated at the peripheries of the four directions beyond the structural area of the sample.

The VNA was preheated to ensure that the internal electronic components reached a stable operating temperature. After preheating, the instrument was calibrated to compensate for various system errors during the test. Before measurement, the stretching device was rotated to adjust the angle of incidence toward normal incidence. Then, the two pieces (that is, the two metallic layers) of the sample were affixed separately onto the two beam-splitting prisms. The alignment marks of both pieces were aligned horizontally and vertically by adjusting the transverse knobs of the 2D displacement platforms and the control knobs of the lifting platforms, ensuring high precision in center alignment. (See Figs. S2–S4 of the Supplementary Document for the effects of the relative in-plane movement of one metasurface on the LP converter performance.)

During the experiment, the device ensured that the centers of the two pieces were aligned and in complete contact at the start. Considering the unavoidable small crevices between the two pieces of the sample for our assembled stretching device, the initial minimum air spacing was set to 150 µm by adjusting the longitudinal knobs of the 2D displacement platforms. The air gap was precisely determined using the indicators of displacement platform. The transmission of the cross-polarized signal was measured in 10 µm increments until the spacing increased to 800 µm.

Figures 5(a)–5(c) depict the measurement results for both cross- and co-polarization amplitude transmissions, and in particular, the PCR performance of the LP converter within the frequency range from 0.1 to 0.13 THz. Figure 5(d) illustrates the bandwidth and amplitude transmission of the cross-polarization component at the center frequency of 0.115 THz as a function of the air-gap distance ranging from 150 to 800 µm. A satisfactory agreement between measurements and simulations is observed, and the discrepancies can be attributed to two primary factors. First, inevitable variations in the geometric dimensions of the metal array during the fabrication of the metasurfaces, as well as in the dielectric constant and loss of the thin-film material, can lead to differences in the resonance frequency and response of the resonators. Second, during experimental measurements, deviations in the center alignment accuracy and surface parallelism of the two pieces of the sample, along with differences in the optical path design and the incident excitation source of the system, may also contribute to these discrepancies. The experimental findings demonstrate the efficacy of a transmissive mechanically tunable metasurface design acting as an efficiency- (ranging from 0.838 to 0.31) and bandwidth-modulated (ranging from 23.7 to 2.4 GHz) LP converter ($\mathrm{PCR}>90\%$) for THz waves. The experimental results corroborate well with simulations. Notably, this modulation is attained without a discernible frequency shift, facilitated by the utilization of a mechanically tunable air gap.

In addition, our design does not use complex electromechanical control systems, which reduces the cost of manufacturing and maintenance and is less susceptible to such factors as material fatigue and mechanical failure. More importantly, using the coupling effect, the three-layer grating array used in previous designs to achieve polarization conversion control of THz radiation is simplified into a dual-layer bright–dark mode coupling structure. Such improvements render the proposed structure applicable in engineering contexts, particularly in wireless communications scenarios where minimal cross-talk between orthogonally polarized channels is desired.

In summary, we have proposed a metasurface design that can be mechanically adjusted to manipulate the polarization state of THz waves. This manipulation is achieved by altering the spacing between two arrays of metallic resonators. The bandwidth and efficiency of the cross-polarization transmission are adjusted, thus achieving spectral reconfigurability while maintaining a high PCR. Increasing the air-gap spacing gradually diminishes the coupling coefficient. Subsequently, we fabricated the metasurfaces on a thin film rather than on a solid block wafer, which can solve the problems of poor mechanical flexibility, fragility, and difficulty in stretching and cutting. A sample stretching device was utilized to precisely fine-tune the air-gap spacing by subwavelength increments. This enables the effective operation of the mechanically tunable metasurfaces in the THz band. Experimental measurements confirm that the developed metasurfaces function as an efficient LP converter. The efficiency and bandwidth of transmission are modulated with adjustable air gap spacing. Moreover, the air-gap spacing controls the degree of coupling between the resonators, allowing for the transition from efficient dual-frequency to single-frequency orthogonal polarization conversion. We qualitatively elucidate the polarization conversion of the gap-tuned meta-atoms using a multilayer film interaction mechanism, which offers a more straightforward and effective explanation.

Our study presents a practical and efficient solution for active control of THz polarization via a metasurface-based microsystem with mechanical modulation. We believe that this concept of mechanical tunability also offers a valuable framework for designing reconfigurable intelligent surfaces. Future integration with microelectromechanical devices is anticipated to enable rapid modulation. Importantly, this approach transcends the THz range, holding promise for applications at lower or higher frequencies. Such a design could find potential applications in high-speed communications and other information technologies.