Surface plasmons (SPs) are collective electron oscillations occurring at the interfaces between dielectric and metallic materials [1]. Their unique characteristics, encompassing localized field enhancement and subwavelength confinement, have spurred applications in various fields, including super-resolution imaging [2,3], surface-enhanced Raman spectroscopy [4,5], and high-sensitivity chemical and biological sensing [6,7]. As they propagate along the surface, SPs interact robustly over a substantial length and can traverse line-guided metallic structures such as stripes, grooves, and gaps, thus showing significant potential in integrated plasmonic components for signal waveguiding and processing [8–10], especially in the domains of communications and photonic circuits. Consequently, extensive investigation has been undertaken across the visible, infrared, and terahertz (THz) frequency bands due to the critical need for seamlessly coupling free-space propagating waves into SPs. In particular, THz SP couplers, characterized by well-confined fields facilitating efficient interaction with analytes in surface sensing and non-linear modulation applications, also simultaneously permit chip-scale THz wave manipulation, propagation, and processing. However, conventional SP couplers based on prisms or gratings are either bulky or inflexible, hindering the further development of this field [11–16].

In the last two decades, the development of metamaterials and metasurfaces has revolutionized the design concepts of traditional optical devices, offering new opportunities for efficient and flexible coupling of SPs [8,10]. In particular, phase gradient metasurfaces have been proven as efficient SP metacouplers [17], since the momentum difference between free-space light and SPs can be matched by properly arranging the phase gradient at the interface. Such a paradigm was first proposed at microwave frequencies and has been rapidly developed in different frequency regimes, fostering various metacouplers of different configurations and different functions [17–38]. However, due to the non-uniformity of a phase gradient metasurface, unlike SPs that constitute the eigenmode of the system, the driven surface waves in this case suffer from significant scattering loss [20]. In order to solve this problem, a recently reported work was proposed to guide out and support the eigenmode SPs by placing an additional plasmonic layer, achieving efficiencies of up to 94% in simulations and 73% in experiments [33]. However, the strict requirements of the separation and parallel configuration of the SP coupler and the plasmonic layer complicate the fabrication, which makes it difficult to extend this scheme to THz frequencies. Hence, there is a pressing need to explore new design strategies and structures to enhance the performance of THz SP metacouplers and expand their potential applications.

In this paper, we propose and experimentally demonstrate a novel approach that employs bright–dark mode coupling in a microstructure composed of bilayer sandwiched metallic patterns to efficiently excite broadband SPs (see Table 1 in Appendix A for a detailed comparison with other SP couplers). By fine-tuning the near-field coupling strength between adjacent meta-atoms and adjusting the resonance frequencies, we can achieve transmitted cross-polarized components with nearly 100% efficiency. To demonstrate the effectiveness of our approach, a THz metacoupler for SP excitation is designed and simulated. Our proposed structure offers several advantages, including high SP conversion efficiency (approximately 95.01% at 0.73 THz under the ideal conditions of no material absorption), wide bandwidth (0.7–0.76 THz with an efficiency exceeding 80%), and a degree of robustness against changes in vertical and lateral coupling. The compact, cost-effective, and high-coupling performance of our SP excitation scheme holds the promise for advancing integrated plasmonics by enabling high-specificity chip-scale sensing [39] and high-speed communications [40].Table 1.

Comparison of Key Metrics between Our Metacoupler and Reported Designs^a

A coupled system for cross-polarization conversion should have a broken mirror symmetry along the incident polarization and the axis perpendicular to the incident polarization. Moreover, for multilayer metallic thin-film metamaterials, near-field intercoupling can be tuned not only by the relative positions of the bright and dark resonances but also by the thickness of the dielectric spacer, which means more design freedoms and high flexibility for applications in subwavelength optics. With these conditions taken into consideration, a unit cell composed of a bilayer coupled split-ring slit (SRS) and a cut-wire slit (CWS) is chosen as the metal pattern and illustrated in Fig. 1(a). The CWS and SRS at the top and bottom layers with orthogonal polarization-selection properties act as the bright and dark modes, respectively. Polyimide (PI) film has a relatively low loss at THz frequencies and therefore is particularly suitable to serve as a spacer in the THz regime. Moreover, the PI was chosen due to its good mechanical stability, which can protect the structural completeness of the processing. Thus, a 22-μm-thick PI film is sandwiched between two 200-nm-thick metal structures. The periods of the meta-atoms are 200 μm along the $x$ direction and 100 μm along the $y$ direction. Since metals behave close to perfect conductors in the THz regime, the SP dispersion relation is very close to that of the free-space wave, and thus the SPs cannot be confined as tightly at the metal surfaces as in the visible regime. To increase the confinement, the two layers of metals are separately coated with 20-μm-thick PI. Figure 1(b) shows two configurations of the meta-atom with the bright resonators oppositely positioned. Intriguingly, by performing a mirror-symmetry operation of the CWS to change the configuration, the transmission amplitude remains unchanged while the phase acquires a $\pi$ shift. This can be explained by the fact that by performing a mirror-symmetry operation to the meta-atom with respect to the plane along the polarization direction of the output radiation, the wave is reversed and thus acquires $\pi$ phase retardation [41].

Similar to common subwavelength slit resonators, the SRS and CWS can be excited by an electric component perpendicular to their base arms and then produce an electric resonance. The resonance frequency and coupling strength are mainly determined by the overall slit length and the relative in-plane and inter-plane spacings. Thus, by changing the above geometric parameters, we can obtain the desired transmission performance at specific frequencies. To materialize the realistic meta-atom design, the spectral response was simulated with the commercial software CST Microwave Studio. By means of numerical optimization, we fix the structural parameters of the meta-atom as follows (see Appendix B for the parameter settings in the simulation): $a_1=157\mathrm{\mu m}$, $b_1=12\mathrm{\mu m}$, $g=10\mathrm{\mu m}$, $a_2=44\mathrm{\mu m}$, $b_2=44\mathrm{\mu m}$, $r=10\mathrm{\mu m}$, $l_1=27\mathrm{\mu m}$, and $l_2=56.5\mathrm{\mu m}$. Thus, the CWS can be excited directly by the incident electric field, while the SRS cannot. The $E_x$-field component was detected to obtain the cross-polarization transmission for all simulations in this work. To verify the feasibility of the design scheme, the inherent material loss was not considered in the simulations. The metal was set as a perfect electric conductor with a conductivity of infinity due to its similar performance in the THz band to that of a common metal like aluminum (Al). The dielectric material PI was set as loss-free with relative permittivities ${\epsilon }_{\mathrm{PI}}=3.26$ at the frequency of interest (0.73 THz), respectively. The relative permittivities of the dielectric materials were experimentally obtained by photoconductive-antenna-based THz time-domain spectroscopy (THz-TDS) (see the detailed descriptions in Appendix C) and consistent with previous reports in the THz frequency range [42,43]. Herein, we only consider the real parts of the dielectric constants. Figure 1(c) shows the simulated amplitudes of the transmitted cross-polarized component for the two different configurations [Fig. 1(b)] with the above parameters. It is clear that the designed meta-atoms exhibit the same and high-efficiency transmission of around 96%. As shown in Fig. 1(d), the phase difference between the two configurations is maintained at $\mathrm{\Delta }\psi =\pi$ over the entire frequency range.

By systematically scaling the geometric parameters of the dark and bright mode resonators, as depicted respectively in Figs. 1(e) and 1(f), it is observed that the collective response of the dual-layer bright–dark mode resonator system undergoes substantial frequency shifts and amplitude reductions. The optimal system response is achieved only when the scale factor is set to 1, representing the adoption of the original optimized geometric parameters. To gain further insight into the performance of the polarization conversion and bright–dark resonance mode interaction, we have simulated the cross-polarized electric field distributions (see Appendix D for the results and detailed discussion).

Before fabrication and measurement of the meta-atom samples, we take into account the loss of the actual materials. Consequently, the geometric parameters of the structure are re-optimized, incorporating the imaginary parts of the dielectric constants and the finite conductivity of metal Al: ${\sigma }_{\mathrm{Al}}=3.56\times {10}^7\mathrm{S}/\mathrm{m}$ and ${\epsilon }_{\mathrm{PI}}=3.26\times (1+0.011i)$. (For non-ideal geometric parameters and corresponding simulation results, see the detailed descriptions in Appendix E; see Appendix F for the steps of sample fabrication.)

To experimentally validate the accuracy of numerical simulations, we characterized the fabricated sample using far-field THz-TDS to measure the transmittance of the unit cells under normal incidence. During the experiment, the chamber was purged with dry nitrogen to avoid noise from water vapor absorption. Since both the incident and detection modules of our THz-TDS system were sensitive to the $x$-polarized electric field, the sample needed to be rotated 90° and tightly attached to the sample holder during measurement. To ensure high polarization-directivity of the incident wave, we placed an $x$-direction polarizer in front of the samples and measured the cross-polarization transmittance by sequentially placing a 45°- and an $x$-direction polarizer behind the samples. A schematic diagram of the system and microscopy images of the fabricated meta-atom samples are shown in Fig. 2(a). The results, depicted in Figs. 2(b) and 2(c), show that the measured outcomes align reasonably well with the transmission and phase spectra of the simulations. Specifically, the response peaks and step changes in phase in the experiment exhibit small differences compared to the simulations. There are a slight frequency shift and some intensity changes, potentially caused by the inherent loss of the materials used in the actual samples and unavoidable dimensional and alignment errors of the top and bottom layers during sample processing.

The supercell of the metacoupler in the horizontal direction is constituted by two meta-atoms featuring the two different configurations shown in Fig. 1(b) with a supercell period $\mathrm{\Lambda }=400\mathrm{\mu m}$ to cover the $2\pi$ phase range and is then periodically replicated in two directions to obtain the high-efficiency metacoupler. Figure 3(a) shows a three-dimensional diagram of the overall design. As mentioned earlier, all samples are spin-coated with a 20-μm-thick PI layer in our design. The solid and dashed lines in Fig. 3(b) illustrate the dispersion relations for the fundamental SP mode in the first Brillouin zone and the free-space wave, respectively (see Appendix G for the settings in the simulation). To show the difference more clearly, we display only the dispersion relations within the frequency range near the working frequency of 0.73 THz. Specifically, $k_{\mathrm{SP}}=1.02k_0$ at 0.73 THz.

SPs arise from the interaction between the incident electromagnetic field and the free electrons on a metal surface. This interaction leads to the formation of a rapidly decaying field within the metal, enabling the propagation of SPs along the metallic plate as an eigen transverse magnetic (TM) mode. Concentrated primarily at the interface, the electric field of the SPs gradually diminishes along the propagation direction due to the absorption of the metal and the inherent loss within the spin-coated PI confinement layer. To ensure that the emergent waves are confined to propagate at the interface, where almost all the emergent electric fields exist as $E_z$ components, the phase gradient should equal the SP wavenumber, namely, $\mathrm{d}\phi /\mathrm{d}x=\pi /200=k_{\mathrm{SP}}$ in our design. Figure 3(c) illustrates the simulated real part of the $E_z$-field distribution, obtained by defining electric field monitors in the $xz$-plane at the center ($y=0$), with the bottom layer of metacoupler located at $z=-31.2\mathrm{\mu m}$. It can be seen that the excitation exhibits bidirectional launching behavior, and the electric field increases obviously due to the constructive interference caused by the periodic pattern.

Figure 3(d) illustrates the $E_z$-field intensities of the SPs with the probes placed 50 μm away from the bottom layer and five wavelengths away from the excitation range of the metacoupler. The positions of the probes were the same on the positive and negative sides of the $x$-axis. Upon comparison, it is found that the $E_z$-field intensities obtained by the two probes are almost the same across the entire frequency range, indicating that the SP excitation is not only bidirectional but also symmetrical. For a simplified analysis, each row of the metasurface that satisfies the phase matching condition $\mathrm{d}\mathit{\phi }/\mathrm{d}x=k_{\mathrm{SP}}$ can be treated as a source of secondary SPs (the detailed descriptions of the generation mechanism of SPs based on phase discontinuities are given in Appendix H). In the case where the phase gradient along the $y$ direction is zero, namely $\mathrm{d}\phi /\mathrm{d}y=0$, the emergent SPs are in phase with each other and collectively produce a normal wavefront. As shown in Fig. 3(e), the simulated distribution of the real part of the $E_z$-field in both the excitation and propagation regions is obtained in the $xy$-plane at 50 μm above the metacoupler. The excitation range is marked by a dashed rectangle, and it can be seen that the excited SPs exhibit a normally launched wavefront with a parallel wavelength ${\lambda }_{\mathrm{SP}}=392\mathrm{\mu m}$. According to the definition of the SP propagation distance, we can obtain the propagation distance as 23 mm. Furthermore, we have simulated the influence of the coupling changes both laterally [44] and vertically [45] on the excitation results, and a degree of insensitivity is observed (the detailed descriptions are given in Appendix I).

To confirm the numerically simulated results discussed earlier, we performed experiments employing scanning near-field THz microscopy to measure the SP electric field of the metacoupler under linearly polarized normal incidence, as depicted in Fig. 4(a). The fabrication process was similar to the case for the meta-atom samples. This measurement was achieved by employing a pair of electrodes sensitive to the $E_z$ component. For a comprehensive 2D scan of the electric field, the sample was mounted on a 2D translation stage, and the probe could be moved precisely by electrical machinery. The distance between the metasurface and the probe was fixed at 50 μm, consistent with the simulation conditions. The overall scanning area covered $9\mathrm{mm}\times 3\mathrm{mm}$, with 0.1 mm and 0.05 mm intervals along the horizontal and vertical directions between consecutive scans, respectively. The distribution of the real part of the $E_z$-field at 0.73 THz in both the excitation and propagation regions of the metacoupler depicted in Fig. 4(b) represents a well-defined surface wave with a wavelength of ${\lambda }_{\mathrm{SP}}=0.405\mathrm{mm}$.

Figure 4(c) illustrates the $E_z$-field intensities of the SPs at identical distances of 1.5 mm away from the left and right sides of the excitation range of the metacoupler. The measurements distinctly demonstrate bidirectional and symmetrical SP coupling properties that broadly align with the corresponding simulations (the simulated results under non-ideal conditions are shown in Appendix J). Thus, we have experimentally confirmed the proposed coupling scheme. We attribute the observed discrepancy to the fabrication imperfections arising from the proximity effects of non-uniform illumination and misalignment of the bilayer metallic patterns in the lithography process. Additionally, non-exact replication of the excitation conditions and slight variations in the dielectric constants and conductivity between the real and simulated values may also contribute to the differences.

It is almost impossible to collect all the near-field energy in the experiment. Therefore, we will indirectly quantify the efficiency by comparing the intensity of the SPs excited by our metacoupler with that of an optimized slit coupler with the same excitation region. Additionally, their average intensity values were obtained at the same position in the propagation region. The experimental results in Fig. 4(d) show that the average strength of the metacoupler at 0.73 THz is 5.6 times that of the slit coupler, which indirectly proves the higher efficiency of our design.

Having demonstrated the capability of our metacoupler, we quantitatively evaluated the working efficiency of our scheme. In our calculations, we numerically integrate the total powers carried by the excited SP beam and the linearly polarized input beam and then define the ratio between them as the working efficiency of our metacoupler. As the schematic diagram in Fig. 5(a) and the results for ideal parameters in Fig. 5(b) show, the efficiency of a sample with 10 supercells reaches 95.01% at 0.73 THz. It is slightly lower than the polarization conversion efficiency of the meta-atoms and maintains a high level of over 80% within the frequency range from 0.7 to 0.76 THz. Such a high performance for SP excitation in the THz regime is rarely reported in previous literature.

The key factors responsible for the low SP excitation efficiencies of previous THz couplers are the inherent loss of materials, the normal-mode reflection at the surface, and the decoupling of SPs back into propagating waves. A thorough analysis of our scheme indicates that it has successfully resolved most of the previously identified issues. The material of our metacoupler is assumed to be lossless in the simulation, without considering the ohmic loss of the metal and the non-radiative loss of the dielectrics. In the experiment, we also selected COC film with minimal loss to act as a spacer layer. In addition, using a bright–dark mode coupling-based bilayer metasurface instead of a single-layer metallic resonance can easily break through the theoretical limitation of 25% in polarization conversion because this design significantly suppresses the channels other than cross-polarization transmission. The high transmission of the meta-atoms forms the basis for efficient SP conversion. Additionally, through the optimization of the geometric parameters, a reasonable phase gradient is provided to compensate for the impedance and momentum mismatches between the SPs and the free-space wave. Consequently, the excited SPs can penetrate the materials to the maximum extent, minimizing specular reflection and backscattering of the electromagnetic waves and thereby achieving efficient excitation. Moreover, by properly designing the thickness of the coated PI layers, the wave vector of the driven surface waves generated in the structural region matches that of the eigen SPs propagating in the guide-out region, preventing the SPs from leaving the interface.

We further compared the SP excitation efficiencies as a function of the number of supercells $n$ for our metacoupler and a reflectarray coupler at 0.73 THz. The diagram of the reflectarray coupler is shown in Fig. 5(c), where the optimized structural parameters under ideal conditions are fixed as follows: $l_1=94\mathrm{\mu m}$, $l_2=47\mathrm{\mu m}$, $l_3=114\mathrm{\mu m}$, $l_4=60\mathrm{\mu m}$, and the supercell period $\mathrm{\Lambda }=400\mathrm{\mu m}$. The reflectarray coupler comprises two distinct meta-atoms with varying parameters. Both meta-atoms exhibit an amplitude close to unity at a frequency of 0.73 THz, accompanied by a phase difference of $\pi$. Consequently, the design of the reflectarray coupler also embodies the features of bidirectional symmetric excitation. The SP excitation performance of our metacoupler is relatively low initially when $n$ is small [Fig. 5(d)]. However, it increases with $n$ and reaches the highest level when $n$ is around 10, and then slightly decreases with $n$. Nevertheless, no matter how $n$ changes, the excitation efficiency is always maintained at a high value of above 75% at the working frequency. In sharp contrast, upon achieving the peak efficiency, the reflectarray coupler experiences a swift and continuous decrease in efficiency with the continued augmentation of the excitation region.

Increasing the number of the excitation region structures yields a concurrent rise in the cross-coupling effects among them, which results in partial dissipation of SP energy, thereby diminishing the excitation efficiency. Moreover, scattering loss includes two types: one due to the inhomogeneity of the microstructure, and the other due to the decoupling caused by reciprocity. Both our design and the reflectarray metasurface exhibit the second type of scattering loss, which is why the SP coupling efficiency of the two couplers decreases after saturation as the excitation area further increases. However, regarding the first type, the non-uniformity of the metallic structure in the reflectarray metasurface is significantly greater than that of our metacoupler, resulting in a more severe scattering loss. This also explains why the SP coupling efficiency of the reflectarray metasurface declines more rapidly and noticeably when the number of supercells increases.

Finally, we compared a traditional transmissive slit coupler with our metacoupler by considering material losses. The diagram of the optimized slit coupler is shown in Fig. 5(e). The width of the slit is $h=80\mathrm{\mu m}$ and the supercell period is $\mathrm{\Lambda }=400\mathrm{\mu m}$. Both the slit coupler and our metacoupler are designed based on the same conditions in terms of operation frequency, structural area, and bidirectional excitation characteristics, all pertinent to a fair comparison of SP coupling efficiencies. The efficiencies of the two couplers at different frequencies are shown as the green and red lines in Fig. 5(f), respectively. The metacoupler has a maximum efficiency of 53.02% at 0.73 THz, while the slit coupler has only 16.6%, a difference of 3.2 times.

This difference between the two couplers can also be attributed to the scattering losses as discussed above. However, the excitation efficiency of our metacoupler with material losses considered [53.02% in Fig. 5(f)] is lower than that of lossless materials [95.01% in Fig. 5(b)]. This is because the operation mechanism of our design utilizes localized SP resonances and strong near-field coupling between the two modes. When material losses are considered, the interaction process will amplify the impact of metal and dielectric medium absorption on the SP coupling efficiency, preventing it from exceeding 60% at the working frequency. Admittedly, the control principle of the slit coupler is similar to a single-layer hole array [46], exhibiting extraordinary optical transmission. By comparison, our design leverages the near-field coupling between the bilayer bright and dark resonators to achieve orthogonal polarization conversion and generate an additional $\pi$ phase difference through mirror symmetry. This capability, which cannot be achieved with single-layer metal array couplers, enables efficient and wideband SP excitation. Certainly, there exist numerous metrics to assess the performance of an SP coupler. Here, we highlight the advantages of our metacoupler in its efficiency in exciting SPs. However, the bilayer metallic structures come with apparent drawbacks in fabrication complexity and cost.

In summary, our study introduced and experimentally demonstrated a bilayer metacoupler with a phase gradient enabled by bright–dark mode coupling. This design effectively connects SPs and free-space propagating waves in the THz frequency range, yielding exceptionally high efficiency. The high efficiency of this SP metacoupler stems from its ability to circumvent such issues as initial reflection and the decoupling effect, which have plagued previous THz SP couplers and diminished their performance. Model simulations in our study predict an optimal excitation efficiency of 95% under ideal conditions. This marks a substantial improvement over existing designs. To validate the effectiveness of our approach, we fabricated a sample and conducted tests on bidirectional symmetric SP excitation. Our metacoupler maintains consistently high efficiency levels, surpassing 80% over the frequency range from 0.7 to 0.76 THz in the simulation. Importantly, the performance of our metacoupler exhibits minimal susceptibility to variations in the number of supercells and a degree of robustness against vertical and lateral coupling changes. Surface wave launchers with bidirectional excitation are convenient and compact for applications such as real-time online spectral analysis that require reserved reference light. Our approach paves the way toward high-specificity chip-scale sensing and high-speed communication applications in the field of THz science.