Double-layer metasurface-based edge coupler for low-loss coupling between optical fiber and thin-film lithium niobate photonic chip Authors: Xie Zou, Hongliang Li, Duk-Yong Choi, Jin Tae Kim, Sang-Shin Lee Published: 2025 ABSTRACT Thin-film lithium niobate (TFLN) has recently emerged as a promising platform for next-generation photonic integrated circuits, owing to its exceptional electro-optic, nonlinear, and optical properties. Despite their advantages, efficient light coupling between standard micron-sized single-mode optical fibers and submicron-sized TFLN waveguides remains a critical challenge, primarily because of the mode size mismatch and limited tolerance over broad wavelength ranges. In this paper, we propose and theoretically investigate a novel edge coupler based on an all-dielectric double-layer metasurface (DLMS), designed to overcome these limitations. The DLMS structure, inspired by the optical function of two orthogonal cylindrical lenses, reshapes a Gaussian beam emitted from an optical fiber into an elliptical profile that closely matches the guided mode of rib-type TFLN waveguides. Finite-difference time-domain simulations confirm that the coupler achieves a coupling loss (CL) below 1.9 dB/facet across the full telecommunication band (1460–1630 nm), reaching a minimum CL of 0.6 dB/facet at 1550 nm. Moreover, the device offers relaxed alignment tolerances of ± 1 μm vertically and ± 2.5 μm laterally, facilitating practical implementation. The proposed DLMS-based edge coupler offers a compelling solution for low-loss broadband fiber-to-chip interfaces in TFLN photonic systems.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-32290-6. Introduction In recent years, owing to its great potential for outstanding and versatile electro-optic characteristics, the thin-film lithium niobate (TFLN) photonic platform has attracted increasing attention from industry and academia, and is regarded as a strong competitor for next-generation photonic integrated circuits. A variety of TFLN-based photonic devices have been successfully realized and have demonstrated excellent performance, including low-loss waveguides1, optical frequency combs2,3, high-speed electro-optic modulators4,5, second harmonic generators6–8, and high-Q resonators9,10. However, the fiber coupling efficiency is limited by the substantial disparity between the mode field diameter (MFD) of the waveguide and standard single-mode fiber (SMF). In the vertical direction, the MFD of the TFLN waveguide remains at the nanoscale owing to the TFLN thickness being less than 1 µm. In the horizontal direction, however, the MFD gradually decreases from a microscale size at the input waveguide region to a nanoscale size as light propagates toward the central waveguide region. Conversely, in some spot size converter (SSC) designs, the TFLN waveguide width increases from tens of nanometers to the microscale. In contrast, the mode of the SMF consistently remains at the microscale in the vertical and horizontal directions, respectively. This leads to a substantial mode-field mismatch, which remains a major obstacle to the efficient practical deployment of the device. Various optical coupling structures have been proposed and demonstrated to achieve efficient fiber-to-TFLN optical coupling. For instance, prism couplers exhibit a coupling loss (CL) of approximately 5 dB/facet for waveguides with cross-sectional dimensions as small as 1 μm × 0.73 μm11. Couplers based on gold (Au) gratings demonstrate a CL of 3 dB/facet at a wavelength of 1550 nm, with a 1 dB/facet coupling bandwidth of up to 58 nm12. However, grating couplers exhibit bandwidth limitations and polarization dependence. A prism coupler requires precise mechanical alignment and stable external pressure to maintain its position. With evanescent coupling through the power transfer between two adjacent elements, on-chip optical loss between the high-index capped sharp fiber and tapered waveguide was reported to be less than 1.4 dB13. However, this method usually requires specially made tapered fibers and uncovered parts of the device, which significantly increases the difficulty of packaging. For easier and more compact configurations of fiber-chip coupling, researchers have attempted to configure a horizontal edge coupler using a tapered lens fiber; however, the results are still unsatisfactory because they exhibit a CL of 10 dB/facet6. To further improve the coupling efficiency of horizontal edge couplers, various configurations utilizing mode-size converters have been proposed. Krasnokutska et al. fabricated an inverted taper structure at the end of a TFLN ridge waveguide (RWG), covered with a 3-μm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} cladding layer14. This design achieved a CL of 2.5 dB/facet. However, the structure requires a metal hard mask and lift-off process, which significantly increases sidewall roughness during metal removal. In addition, plasma-enhanced etching engenders intrinsic thin-film defects, resulting in a CL as high as 12.5 dB/facet. Li et al. introduced a SiON waveguide on a single-layer inverted-taper waveguide15. Simulation results showed a loss of 2.68 dB/facet near a wavelength of 1550 nm for the case of coupling to a lensed fiber. The same structure was also applicable at wavelengths of 1310 nm (3.19 dB/facet) and 850 nm (2.37 dB/facet) only for the case of transverse magnetic (TM) polarization mode. He et al. proposed a double-layer inverted-taper SSC1. When paired with a lensed fiber, the structure achieved a CL of 1.7 dB/facet. However, the 30-nm-wide tip extends over 120 μm, making the transmission section extremely sensitive to fabrication variations. Even slight deviations in the taper length or width can significantly alter mode confinement and increase propagation loss. Ying et al. achieved an ultra-low loss based on a double-layer inverted-taper structure, with as low as 0.5 dB/facet, though the design cannot operate effectively at longer wavelengths16. Liu et al. proposed a three-layer inverted-taper SSC suspended within a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} waveguide, which achieved CLs of 0.5 dB/facet at 1550 nm and 1 dB/facet at 775 nm, demonstrating dual-wavelength operation17. Tang et al. further advanced the multilayer SSC concept by incorporating a SiN suspension layer to form a hybrid multilayer coupler18. This configuration achieved a CL of 1.3 dB/facet when coupled to a lensed fiber and maintained the wavelength-dependent loss below 0.4 dB across the C-band (1530–1565 nm). However, it should be remarked that the fabrication of such multilayer tapered structures requires multiple high-precision alignment and etching steps, and the mechanical fragility of the suspended structure leads to complex processing and low fabrication yield. Moreover, various edge couplers have been attempted on other common optical chip platforms (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiN}}_{\mathrm{x}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} platforms), such as quasi-lossless inverse taper SSC19 and high-efficiency, high-tolerance couplers based on tilted periodically segmented waveguides20, which achieve low CLs of ~ 0.2 dB and 0.58 dB, respectively, at 1550 nm wavelength, demonstrating their superior performance in broadband operation. However, these schemes typically rely on long coupling sections and highly engineered waveguide cross-sections (e.g., high-aspect-ratio tips and specifically tailored segmented geometries), while providing practically limited support for fabrication and system-level integration on TFLN platforms. Consequently, conventional fiber-to-TFLN coupling schemes can be divided into a surface coupling type (such as prism and grating couplers) and a horizontal edge coupling type based on TFLN SSCs. Surface couplers require delicate mechanical alignment and external pressure, suffering from limited bandwidth and polarization dependence. Meanwhile, TFLN SSCs mostly rely on ultra-long, ultra-narrow tapers and complicated multilayered inverse taper structures, thus imposing stringent fabrication and alignment tolerances. In this context, a compact fiber-to-chip coupling solution should be developed to tailor the mode profile at the chip facet to mimic the fiber mode without relying on cost- and fabrication-inefficient ultra-long, ultra-narrow tapers, thereby achieving salient features including high coupling efficiency, relaxed alignment tolerance, wide spectral bandwidth, and low polarization sensitivity. Recently, fiber meta-tip (FMT) metasurfaces integrated onto the facets of fibers have emerged as promising alternatives to conventional edge coupling techniques. By enabling subwavelength control over the phase, amplitude, and polarization of light, metasurfaces offer unprecedented flexibility in beam shaping, thereby positioning FMTs as highly efficient and versatile optical edge couplers. There are two common types of FMT: direct deposition of the meta-atom on the fiber facet and bonding of the metasurface to the fiber via a spacer layer. Kim et al. proposed an all-dielectric metalens by depositing aperiodic silicon (Si) nanopillars on top of a photonic crystal fiber and achieved a focusing efficiency of 88%21. Ren et al. employed a nanoprinted tower structure to interface a metasurface with an optical fiber, enabling near-diffraction-limited focusing over a broad telecom wavelength range of 1250–1650 nm22. At a wavelength of 1550 nm, the full width at half maximum (FWHM) of the focal spot was 3.84 μm. Chen et al. proposed and theoretically studied a metasurface-based fiber-to-chip multiplexing coupler that enables flexible demultiplexing of two or three modes from a few-mode fiber into the corresponding on-chip single-mode waveguides with low crosstalk23. In this study, we propose and theoretically investigate a novel fiber-to-chip edge coupler based on an all-dielectric double-layer metasurface (DLMS), aiming to address the persistent challenges of mode mismatch and alignment sensitivity in TFLN photonic integration. Unlike conventional taper- or grating-assisted couplers, which suffer from high fabrication complexity and limited bandwidth, the DLMS-based FMT efficiently reshapes the Gaussian beam emitted from the fiber into an elliptical profile that closely matches the guided mode of a ridge waveguide. Finite-difference time-domain simulations confirm a high mode overlap efficiency of up to 87% with the mode field profile of a precisely designed RWG, achieving a CL as low as 0.6 dB/facet at 1550 nm for transverse electric (TE) polarization. Moreover, the proposed coupler demonstrates broadband performance with CL < 1.9 dB across the 1460–1630 nm wavelength range. At the same time, it has a large alignment tolerance of ± 1 μm in the vertical direction and ± 2.5 μm in the horizontal direction, ensuring suitability for scalable and robust TFLN photonic packaging. This work represents the first demonstration of edge coupling on TFLN using DLMS-based focal shaping, offering a new pathway toward low-loss, broadband, and polarization-flexible photonic integration. Proposed DLMS-based edge coupler Figure 1a shows a schematic of the edge coupler structure with the DLMS-based FMT. The coupling system comprised two main parts: the fiber and waveguide sides. On the fiber side, the structure consisted of an SMF, an epoxy resin glue, an SU-8 expansion layer, and a pair of cascaded metasurfaces (MS1 and MS2). The metasurface was integrated onto the fiber facet through a 30-μm-thick UV-curable epoxy with a refractive index of n = 1.56, which is similar to that of SU-8. The total thickness of the SU-8 layer and epoxy film combined was 115 μm, carefully designed to allow the Gaussian beam emitted from the SMF to expand and match the aperture size of the metasurface. The DLMS-based FMT reshapes the output beam into an elliptical profile suitable for better waveguide coupling. As light propagates through the cascaded metasurfaces, it undergoes strong vertical compression, while maintaining lateral expansion, thereby forming an elliptical focal spot on the focal plane. On the waveguide side, the structure comprised an oxide (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document}) substrate and a TFLN RWG. A micrometer-scale wider taper was introduced at the input facet, then the upper width of the waveguide was gradually reduced from the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{w}}_{1}$$\end{document} (~ μm) to the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{w}}_{2}$$\end{document} (~ nm). As the width decreases, the guided mode becomes increasingly confined within the narrower waveguide core. At the output facet, the waveguide width is symmetrically expanded back to the micrometer scale. Increasing the waveguide width primarily expands the mode field in the horizontal direction, while the vertical dimension remains nearly unchanged. Based on this, a tapered waveguide structure was adopted in this study to gradually reshape the guided mode and achieve better overlap with the elliptical focal spot generated by the DLMS-based FMT.Fig. 1Schematic diagram of the proposed optical edge coupling between DLMS-based FMT and TFLN RWG. Schematic diagram of the proposed optical edge coupling between DLMS-based FMT and TFLN RWG. The mode overlap efficiency η serves as a key indicator of the field matching between the optical fiber mode and the waveguide mode in edge couplers; it is given by24:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\upeta = \frac{{\left| {\int {{\mathrm{E}}_{1} {\mathrm{E}}_{2} {\mathrm{dA}}} } \right|^{2} }}{{\int {\left| {{\mathrm{E}}_{1} } \right|^{2} } {\mathrm{dA}}\int {\left| {{\mathrm{E}}_{2} } \right|^{2} } {\mathrm{dA}}}}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{E}}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{E}}_{2}$$\end{document} denote the electric fields of the fiber and waveguide modes at the coupling facet, respectively. Integration domain A represents the spatial region that includes both field distributions. According to Eq. (1), the coupler should be designed such that the electric field of the waveguide mode at the input facet closely matches that of the fiber, thereby maximizing the mode overlap efficiency. In edge-coupled systems, the overall coupling efficiency is mainly determined by two factors: mode overlap efficiency and transmission efficiency. This study focuses exclusively on the former, investigating how improving the mode overlap at the coupling interface can reduce CL independent of propagation losses or fabrication-induced imperfections. Furthermore, the Fresnel reflection at the fiber–air interface (~ 14%) was not included in the simulation because the input power was normalized to the field entering the metasurface coupler, as recorded by the incident monitor. This ensures that the reported coupling efficiency simply corresponds to the field transmission through the DLMS. The CL, commonly used to evaluate the performance of edge couplers, is directly derived from the mode overlap efficiency as: CL \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$= - 10 \times {\mathrm{log}}_{10}$$\end{document}η. Simulation and analysis Guided mode profile of a TFLN RWG The mode field characteristics of the waveguide were first analyzed. Figure 2a shows a cross-sectional view of a conventional TFLN RWG. As described in the “Proposed DLMS-based Edge Coupler” section on the DLMS-based FMT coupling system, the waveguide features a ridge-type tapered structure with a wide input facet. A 600-nm TFLN is built on a 2 μm-thick oxide substrate. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{w}}_{1}$$\end{document} of the waveguide is 3.5 μm, the etch depth (ED) is about 300 nm, and the sidewall angle (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uptheta$$\end{document}) is 60\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}. This study adopts an x-cut TFLN waveguide with an extraordinary refractive index \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{n}}_{\mathrm{e}}=2.14$$\end{document} and an ordinary refractive index \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{n}}_{\mathrm{o}}=2.21$$\end{document} at a wavelength of 1550 nm. For x-cut TFLN, most studies have adopted TE-polarized light because of its superior nonlinear and electro-optic coefficients. Therefore, for simplicity, the following discussion focuses on the determination of each coupling parameter using the fundamental TE mode (TE0). The coupling performances of both the TE and TM fundamental modes are presented at the end of this section. Figure 2b presents the simulated mode field distribution at the edge cross-section of the waveguide using a finite-difference eigenmode solver based on Mode Solutions (Ansys Lumerical, Canada). An elliptical mode field is obtained, with vertical and lateral dimensions of 0.45 μm and 2.0 μm, respectively.Fig. 2(a) Cross-sectional structure of conventional TFLN RWG. (b) Calculated TE0 mode field at 1550 nm wavelength. (a) Cross-sectional structure of conventional TFLN RWG. (b) Calculated TE0 mode field at 1550 nm wavelength. Design of a DLMS for providing an elliptical beam profile For enhanced coupling efficiency, the field distribution of the focal spot generated by the proposed edge coupler should closely match the elliptical mode profile of the input waveguide, as shown in Figure 2b, with vertical and lateral sizes of 0.45 and 2.0 μm, respectively, at λ = 1550 nm. To meet this requirement, this study selected an orthogonally arranged cylindrical lens system, which was intended to transform noncircular incident light into a symmetric circular output beam25. Considering the reversibility of the optical paths, we adopted the reverse light path of the system for our design, as illustrated in Figure 3a. A pair of orthogonally arranged cylindrical lenses enables independent manipulation of each optical axis. The corresponding metasurface structure was designed to implement this optical function. At an operating wavelength of 1550 nm, achieving a horizontal focal spot of 2.0 μm requires a numerical aperture (NA) of ~ 0.47 based on the diffraction-limited formula d = 0.61λ/NA. However, to reach a vertical spot size of 0.45 μm, the theoretical NA would need to be as high as 2.1. For conventional transmissive metasurfaces, the achievable NA is typically limited to values below 1 owing to constraints in the material properties and structural design, rendering such high-focusing capabilities practically unachievable. To minimize the focal spot, the approximate formula FWHM = 0.51λ/NA can be employed, indicating that increasing the NA of the metalens is beneficial for reducing the focal spot. However, when the aperture exceeds 0.9, evanescent wave components become dominant, leading to significant attenuation and a notable decrease in the theoretical focusing efficiency26,27. Considering the trade-off between focusing performance and transmission efficiency, this work adopts metasurfaces with NA = 0.8 in both the vertical and horizontal directions to balance the evanescent wave suppression and focal spot size reduction.Fig. 3(a) Schematic of orthogonally arranged cylindrical lenses for beam shaping, where the red curves indicate the beam profiles. (b), (c) Phase profiles of MS1 and MS2, which replicate the function of the orthogonal cylindrical lenses in (a) using a DLMS. (d), (e) Simulated phase shift and transmission of the nanopillars used in MS1 and MS2, respectively. Electric-field distributions in the (f) x–z plane and (g) x–y plane. The inset in (g) shows the three-dimensional intensity distribution at the focal plane. (a) Schematic of orthogonally arranged cylindrical lenses for beam shaping, where the red curves indicate the beam profiles. (b), (c) Phase profiles of MS1 and MS2, which replicate the function of the orthogonal cylindrical lenses in (a) using a DLMS. (d), (e) Simulated phase shift and transmission of the nanopillars used in MS1 and MS2, respectively. Electric-field distributions in the (f) x–z plane and (g) x–y plane. The inset in (g) shows the three-dimensional intensity distribution at the focal plane. For the proposed DLMS-based FMT, two cylindrical metasurfaces were orthogonally positioned with a dielectric spacer of 10.8 μm-thick oxide layer between them. The diameter of each metasurface is 30 μm. The thickness of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} is determined by the relation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{1}-{\mathrm{f}}_{2}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{2}$$\end{document} represent the focal lengths of MS1 and MS2, respectively. Based on the relation between the NA and focal length of the metasurface, NA \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$=$$\end{document} n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bullet$$\end{document} sin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left({\mathrm{tan}}^{-1}\left(\mathrm{d}/2\mathrm{f}\right)\right)$$\end{document}28,29, we could determine \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{2}$$\end{document} to allow the beam to be focused on a spot after passing through the two metasurfaces (details are described in the Supporting Information Fig. S1). MS1 exclusively applies phase modulation along the horizontal direction (x-axis) of a spatially isotropic incident Gaussian beam. The MS2 further modulated the vertically oriented component (y-axis) of the spatially anisotropic beam emerging from MS1. This sequential transformation results in the formation of an elliptical focal spot with different beam dimensions in the vertical and horizontal directions. Numerical 3D simulations were conducted to validate the functionality of the proposed FMT using a commercial finite-difference time-domain software, FDTD Solutions (Ansys Lumerical, Canada). Figure 3b and c show the individual phase profiles of each metasurface. The phase distributions of cylindrical MS1 and MS2 can be expressed as30,31:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upvarphi }_{{{\mathrm{MS}}1}} \left( {\mathrm{x}} \right) = \frac{{2\uppi {\mathrm{n}}_{{{\mathrm{SiO}}_{2} }} \times \left( {{\mathrm{f}}_{1} - \sqrt {{\mathrm{x}}^{2} + {\mathrm{f}}_{1}^{2} } } \right)}}{\uplambda }$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upvarphi }_{{{\mathrm{MS}}2}} \left( {\mathrm{y}} \right) = \frac{{2\uppi {\mathrm{n}}_{{{\mathrm{air}}}} \times \left( {{\mathrm{f}}_{2} - \sqrt {{\mathrm{y}}^{2} + {\mathrm{f}}_{2}^{2} } } \right)}}{\uplambda }$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\uplambda$$\end{document} is the wavelength of the incident light, and x and y represent the coordinates corresponding to the center of each unit cell in the middle of the metasurface. By setting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{n}}_{{\mathrm{SiO}}_{2}}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{n}}_{\mathrm{air}}$$\end{document} to 1.45 (for MS1) and 1.0 (for MS2), respectively, the focal lengths of the cylindrical metasurfaces, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{2}$$\end{document}, were designed to be 21.6 μm and 10.8 μm. The unit cells selected for the metasurfaces are shown in Figure 3d and e. A set of hydrogenated amorphous silicon (a-Si:H, n = 3.43) nanopillars with varying radii r was arranged in a square lattice with equal periods in both the x- and y-directions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{p}}_{1}$$\end{document} = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{p}}_{2}$$\end{document} = 700 nm. The heights of the a-Si:H nanopillars are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{h}}_{1}$$\end{document} = 900 nm and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{h}}_{2}$$\end{document} = 800 nm for the two metasurfaces. The selected unit cell configurations for MS1 and MS2 achieve full 2π phase coverage while maintaining transmission efficiencies above 85%. Figure 3f shows the roles of DLMS in creating an elliptical mode profile at the focusing position at a wavelength of 1550 nm in the x–z plane. Figure 3g shows the simulated electric-field distribution at the focus position in the x–y plane. The inset shows the three-dimensional intensity distribution in the focal plane. The DLMS exhibits a focal length of 11.2 μm, with the focal spot dimensions measured to be 1.1 μm in the vertical direction and 2.1 μm in the horizontal direction. CL between the DLMS and the TFLN RWG When the waveguide thickness is fixed at 600 nm, the vertical mode size at the edge coupling interface remains constant at 0.45 μm, resulting in a consistent 60% mismatch with the vertical size of the incident elliptical beam. By contrast, the size of the horizontal mode field was primarily influenced by the upper width of the waveguide. Therefore, the relationship between the upper width of the waveguide and CL was investigated. Under quasi-TE mode excitation, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{w}}_{1}$$\end{document} of the waveguide was varied, and the calculated CL at the edge-coupling interface as a function of the width is shown in Figure 4a. In the simulations, the incident beam was precisely aligned with the input facet of the waveguide to ensure no significant lateral or angular misalignments. Consequently, the main factor affecting the coupling efficiency was identified as the physical size mismatch between the incident beam and waveguide modes. As the upper width increases from 1.0 to 3.5 μm, the horizontal mode size deviation drops from 64 to 5%, leading to a significant improvement in mode overlap and a sharp reduction in CL. However, further increasing the width to 6.0 μm causes the waveguide mode to exceed the beam size, reducing overlap efficiency and increasing CL again, as shown in Figure 4b. The deviation calculation method is as follows:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Mode}}\;{\mathrm{size}}\;{\mathrm{deviation}} = \frac{{\left| {{\mathrm{RWG}}\_{\mathrm{FWHM}}_{{{\mathrm{x}}\left( {\mathrm{y}} \right)}} - {\mathrm{DLMS}}\_{\mathrm{FWHM}}_{{{\mathrm{x}}\left( {\mathrm{y}} \right)}} } \right|}}{{{\mathrm{DLMS}}\_{\mathrm{FWHM}}_{{{\mathrm{x}}\left( {\mathrm{y}} \right)}} }} \times 100{\mathrm{\%}}$$\end{document}Fig. 4(a) Calculated CL with respect to the upper width of the input waveguide. (b) The deviation between the mode size of waveguides with varying upper widths and that of the incident beam. (a) Calculated CL with respect to the upper width of the input waveguide. (b) The deviation between the mode size of waveguides with varying upper widths and that of the incident beam. Furthermore, the dependence of CL on the wavelength was investigated. Based on the theoretical analysis in Figure 4, a waveguide upper width of 3.5 μm was selected for optimal coupling. As shown in Figure 5a, the simulation results demonstrate that, for a 600 nm-thick waveguide, CL remains between 2.0 and 3.1 dB/facet across the wavelength range of 1460–1630 nm. Figure 5b shows the variation in the MFD between the incident beam and the waveguide at different wavelengths. WG_v and WG_h denote the vertical and horizontal dimensions of the waveguide mode fields, respectively; DLMS_v and DLMS_h are the vertical and horizontal dimensions of the incident elliptical spot, respectively. As the wavelength increased from 1460 to 1630 nm, the vertical dimension mismatch between the elliptical beam and the waveguide mode remained at approximately 61%. In the horizontal direction, the deviation varies from 0 to 41%.Fig. 5(a) Calculated CL as a function of operation wavelength for 3.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upmu m}$$\end{document} waveguide width. (b) Comparison of the MFD between the elliptical spot and the 3.5 μm-wide waveguide at different wavelengths. (a) Calculated CL as a function of operation wavelength for 3.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upmu m}$$\end{document} waveguide width. (b) Comparison of the MFD between the elliptical spot and the 3.5 μm-wide waveguide at different wavelengths. Impact of variations in the thickness of the TFLN RWG on the CL To improve the coupling efficiency between the DLMS-based coupler and the input facet of the tapered waveguide, the mode size of the waveguide should be preferably matched to the quasi-elliptical beam generated by the coupler. When it comes to foundry-compatible TFLN platforms (commercially available TFLN on insulator), the TFLN thickness typically lies in the range of 500–700 nm, which represents a practical trade-off between tight confinement, propagation loss, and device footprint for electro-optic modulators32,33 and nonlinear devices34,35. In light of the widely used platforms, the waveguide thickness was set to 600 nm in the simulations. However, the DLMS-based edge coupler provides an incident beam with a vertical FWHM of ~ 1.1 μm at the chip facet. To improve the mode overlap, several approaches can be adopted. These approaches include optimizing the metasurface design, introducing a low-refractive-index cladding layer above the waveguide, and modifying the waveguide geometry. Among these strategies, adjusting the dimensions of the waveguide is considered simple and effective. It should be noted that the waveguide proposed in this study serves solely as a passive transmission channel without any additional functional sections. The optimization of the waveguide thickness is intended to illustrate the impact of mode distribution on the coupling efficiency. We performed a parameter sweep across the TFLN waveguide thickness, and the result is shown in Supporting Information Fig. S2. The simulation results indicated that increasing the waveguide thickness gradually weakens the confinement in the vertical direction and enlarges the corresponding mode. When the thickness reaches 1.95 μm, the waveguide supports a mode profile that deviates from the ideal elliptical shape observed at 600 nm thickness, instead exhibiting a quasi-rectangular elliptical distribution with a vertical dimension of approximately 1.1 μm. In the optimized design, the θ is set to 60°, the upper width of the waveguide is defined as 3.1 μm, and the etch depth is 1 μm, leading to a horizontal mode size of 2.1 μm. Figure 6a and b show the cross-sectional structure at the input facet and the corresponding simulated mode field profile, respectively. Notably, the simulation shows that the minimum CL does not occur exactly at the metasurface geometric focus but is slightly upstream of it. This shift is attributed to the fact that the metasurface is designed to produce an ideal elliptical focal spot at a specific working distance, whereas the 1.95 μm-thick waveguide supports a quasi-rectangular elliptical mode. At the optimal coupling position, the incident beam becomes slightly compressed in one direction, yielding improved mode matching with the waveguide mode. Consequently, a maximum mode-overlap efficiency of approximately 87% was achieved over the S + C + L band. Therefore, although this position was offset from the nominal geometric focus, the degree of mode matching exceeded that of an ideal elliptical spot at the metasurface focus. Figure 6c shows the coupling performance of the proposed design. For TE-polarized light, a CL below 1.9 dB/facet was achieved across the S + C + L telecommunication band, with a minimum value of 0.6 dB/facet at 1550 nm wavelength. To further evaluate the mode behavior at the waveguide interface, we quantified the mode distribution at the input facet of the TFLN waveguide through eigenmode expansion. For TE-polarized input light, the results show that over a wavelength span exceeding 100 nm, more than 80% of the coupled power was carried by the TE0 mode, while the total power belonging to the higher-order TE modes remains below 20%. The slightly larger fraction of the higher-order modes mainly arises from the relatively large TFLN thickness of 1.95 µm used in this work. It inevitably drives the structure into a weakly multimodal regime and allows a small portion of the power to be captured by the higher-order modes. In contrast, for TM-polarized input, more than 80% of the coupled power was concentrated in the TM0 mode within a narrower spectral window of about 30 nm. This behavior is consistent with the fact that the DLMS parameters are primarily optimized for the TE-mode operation. These issues could potentially be addressed through further optimization of the metasurface structure. Mode distributions and their wavelength dependence are given in Supporting Information Fig. S3 in detail. To further quantify the polarization-dependent performance of the proposed coupler, we evaluated the polarization-dependent loss (PDL), defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{PDL}=\left|{\mathrm{CL}}_{\mathrm{TE}}-{\mathrm{CL}}_{\mathrm{TM}}\right|$$\end{document}. As plotted in Figure 6d, its wavelength dependence shows that within an 80-nm wavelength window (1520–1600 nm), the PDL remains below 0.6 dB. This indicates that the DLMS-based edge coupler exhibits a low polarization sensitivity over the operating bandwidth, comparable to that of previously reported low-PDL edge couplers on the TFLN platforms36,37. Moreover, the wavelength range over which the PDL stays below 1 dB exceeds 100 nm, and the PDL is confined within 1.6 dB across an entire 170-nm span of interest. It is also necessary to practically evaluate the robustness of the proposed coupler design considering a standard SMF could inevitably exhibit unpredictable polarization variations. To that end, we rigorously analyzed the impact of variations in the input state of polarization by rotating the linear polarization (LP) of the LP01 mode of the SMF. The additional CL relative to the TE input was quantified as presented in Fig. S4 of the Supporting Information. The results at three representative wavelengths spanning 1460 nm, 1550 nm, and 1630 nm show that the DLMS-based edge coupler exhibits a low sensitivity to variations in the state of polarization over the entire 0–180° range of LP. In practical applications, the impact of thermal effect on coupling efficiency also needs to be considered. The extremely small thermo-optic coefficients of TFLN (2.1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times 10^{ - 6}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{K}}^{ - 1}$$\end{document} for the ordinary index and 8 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times 10^{ - 5}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{K}}^{ - 1}$$\end{document} for the extraordinary index) allow the TFLN waveguide to maintain a stable mode size over the dynamic operating temperature range of 20–120 °C38. In addition, materials constituting the DLMS (a-Si:H39, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document}40, epoxy41, and SU-842) exhibit low thermo-optic coefficients (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 5}$$\end{document}–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10^{ - 4}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{K}}^{ - 1}$$\end{document}) in the telecom wavelength range. As a result, the refractive index changes in all the materials used in our design remain in the vicinity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {\Delta {\mathrm{n}}} \right| < 10^{ - 3}$$\end{document}. For a subwavelength-thick nanostructure, this only induces a phase perturbation of a few \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times 10^{ - 3}$$\end{document} radians according to the phase shift \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \emptyset = \frac{{2\uppi {\mathrm{h}}}}{\uplambda }\Delta {\mathrm{n}}$$\end{document}43. Thus, the DLMS is invariant to thermal disturbances, which impose no significant impact on the polarization-dependent performance of the proposed edge coupling scheme. In this study, we found that the designed DLMS achieves an optimal optical CL in a TFLN waveguide with a width of 3.1 μm, a height of 1.95 μm, and an etch depth of 1 μm. This may be due to the fact that, while a lensed fiber produces a circular focal spot that mismatches the typically elliptical guided mode of a TFLN waveguide, resulting in additional mode mismatch and higher CL, the proposed DLMS generates an elliptically focused beam that precisely matches the waveguide mode.Fig. 6(a) Cross-sectional structure of 1.95 μm-thick waveguide. (b) Calculated TE0 mode field at 1550 nm wavelength. (c) Calculated CL at different wavelengths for a 3.1 μm waveguide width. (d) Calculated PDL in terms of the wavelength for the TE0 and TM0 modes. (a) Cross-sectional structure of 1.95 μm-thick waveguide. (b) Calculated TE0 mode field at 1550 nm wavelength. (c) Calculated CL at different wavelengths for a 3.1 μm waveguide width. (d) Calculated PDL in terms of the wavelength for the TE0 and TM0 modes. Fabrication tolerance and misalignment tolerance The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} dielectric layer serves as an adhesive layer between the two metasurfaces. Its thickness of 10.8 μm ensures that the spacing between the metasurfaces matches the focal length difference \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{f}}_{1}-{\mathrm{f}}_{2}$$\end{document}. The deviation from this designed spacing can induce an elliptical focal spot distortion, ultimately degrading the coupling efficiency. Therefore, the effect of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} thickness variation due to fabrication defects on the coupling performance between the metasurface coupler and the waveguide is worth exploring. Figure 7a illustrates the edge CL of the waveguide for various \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} thicknesses. It can be observed that when the thickness of the dielectric layer deviates within ± 30% from its nominal value, the proposed cascade metasurfaces remain largely insensitive to the variation. The CL between the coupler and the waveguide exhibited minimal fluctuation, remaining within the range of 0.6 to 0.7 dB/facet. Moreover, the variations in the nanopillar structures during the fabrication process and the alignment between the two metasurfaces must be taken into account. Detailed analyses of the nanopillar radius and height variations, as well as the alignment between the two metasurfaces, can be found in the Supporting Information Fig. S5 and Fig. S6. Simulation results show that when the radius variations of ± 15 nm and height variations of ± 90 nm from the design values, and the relative rotation angle between MS1 and MS2 is within 20°, the corresponding additional CL remains below 1 dB/facet.Fig. 7(a) CL of DLMS-based edge coupler (based on different \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} thicknesses) and waveguide at 1550 nm wavelength. (b) Dependence of additional CL on the misalignment of the TE0 mode at 1550 nm. The top images in each graph show the concepts of fabrication tolerance and misalignment. Δx is an out-of-plane misalignment and is not visible in the yz cross-section. (a) CL of DLMS-based edge coupler (based on different \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} thicknesses) and waveguide at 1550 nm wavelength. (b) Dependence of additional CL on the misalignment of the TE0 mode at 1550 nm. The top images in each graph show the concepts of fabrication tolerance and misalignment. Δx is an out-of-plane misalignment and is not visible in the yz cross-section. We also verified the misalignment tolerance between the coupler and waveguide. The alignment tolerance of the DLMS-based edge coupler in the x and y directions is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}x and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}y, respectively. The z-axis represents the propagation axis of the incident light. According to Sect. 3.4, the minimum CL could be achieved when the coupler was coupled with the waveguide at a position in front of the ideal focal point, which was calculated to be 9.2 μm. This position is defined as the origin of the z-axis, and its alignment tolerance is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}z. In the S + C + L band, the mode size of the waveguide remains approximately 1.1 × 2.1 μm2, indicating that, compared to the horizontal direction (x-axis), the waveguide displacement tolerance is smaller in the vertical direction (y-axis) due to the smaller mode size. Figure 7b shows that when the additional CL between the coupler and the waveguide is 1 dB at 1550 nm wavelength, the alignment tolerances in the x, y, and z directions are approximately ± 2.5 μm, ± 1 μm, and ± 2 μm, respectively. These results indicate that the DLMS-based edge coupler exhibits high tolerance to misalignment, which reduces the cost of optical packaging and improves the reliability of TFLN devices in practical applications. Discussion and perspectives Table 1 presents a comparison of the proposed device with other previously reported fiber-to-TFLN chip edge couplers. Those studies have been investigated theoretically and/or experimentally. Although the TFLN devices employing SSC with different number of layers improve the coupling efficiency for the TE mode, they exhibit significantly increased CL for the TM mode, indicating strong polarization dependence17. Subsequently, several works have introduced Si-based dielectric claddings and multi-stage tapered structures on TFLN waveguides to mitigate the polarization sensitivity; however, achieving both low CL and small PDL over a broad wavelength range remains challenging15,44. Recently, Chen et al. demonstrated a low-loss and broadband polarization-diversity edge-coupling scheme on an x-cut TFLN platform by cascading a cantilever edge coupler with a two-stage polarization splitter–rotator (PSR), achieving a minimum coupling loss of approximately 1 dB/facet and a PDL below 0.5 dB over the 1.5–1.58 μm band36. Guo et al. and Fang et al. further proposed polarization-insensitive and broadband TFLN edge couplers, in which three-section fully etched tapers or three \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Si}}_{3}{\mathrm{N}}_{4}$$\end{document} taper stages were exploited to realize coupling losses below about 1 dB/facet for both TE and TM modes, while maintaining PDL < 0.1 dB over a bandwidth exceeding 100 nm37,47. When it comes to the polarization management front on the fiber side, Nezamdoost et al. capitalized on a YOLO-based machine-vision alignment algorithm to achieve high precision alignment between the slow axis of a Panda polarization-maintaining fiber and chip side grooves, providing a new route towards stable polarization-maintaining fiber to chip interfaces48. Ma et al. introduced a polarization-dependent metasurface on the fiber facet to independently collimate and deflect x- and y-polarized incident beams, further highlighting the importance of polarization-related fiber coupling structures in practical systems49.Table 1Comparison of the optical characteristics of the fiber-to-TFLN chip edge couplers.ReferencesEdge coupler combinationInvestigation methodCL for TE [dB/facet]CL for TM [dB/facet]Coupling bandwidth for TE [1 dB]Alignment tolerance [1 dB]14Lensed fiber / Single layer SSCExperiment2.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}–100 nm –1Lensed fiber / Bilayer SSCExperiment1.7 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–– ± 1 μm (Horizontal)16Lensed fiber / Bilayer SSCExperiment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${0}{\text{.5 (}}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–35 nm–17Lensed fiber / Tri-layer SSCSimulation1.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {775}\;{\mathrm{nm)}}$$\end{document}0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}7.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {775}\;{\mathrm{nm)}}$$\end{document}3.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550 }\;{\mathrm{nm)}}$$\end{document}––44UHNAF / Multi-layer SSCSimulation1.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–120 nm–18UHNAF / SiN + Tri-layer SSCExperiment1.3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–35 nm ± 2 μm (Vertical) ± 2 μm (Horizontal)15Laser diode / SiON + Single layer SSCSimulation–2.4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {850}\;{\mathrm{nm)}}$$\end{document}3.2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1310}\;{\mathrm{nm)}}$$\end{document}2.7 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}––45UHNAF / SiON + Bilayer SSCExperiment0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}0.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}100 nm ± 1.3 μm (Vertical) ± 1 μm (Horizontal)46UHNAF / Bident + Bilayer SSCExperiment1.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}–35 nm ± 1 μm (Vertical) ± 1 μm (Horizontal)36 Lensed fiber / Cantilevered single layer SSCExperiment1.4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}1.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}100 nm–37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{SMF / SiO}}_{2}$$\end{document} + Single layer SSCExperiment0.9 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1300}\;{\mathrm{nm)}}$$\end{document}1.1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1300}\;{\mathrm{nm)}}$$\end{document}80 nm ± 2 μm (Vertical) ± 2.5 μm (Horizontal)47SMF / Three \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Si}}_{3} {\mathrm{N}}_{4}$$\end{document} + three-stage SSCSimulation0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}100 nm ± 1.7 μm (Vertical) ± 2 μm (Horizontal)This workDLMS-based FMT / TFLN RWGSimulation0.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}1.2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}130 nm ± 1 μm (Vertical) ± 2.5 μm (Horizontal) ± 2 μm (Propagation)“–”: not mentioned in the article. UHNAF: ultra-high numerical aperture fiber. Table 1Comparison of the optical characteristics of the fiber-to-TFLN chip edge couplers.ReferencesEdge coupler combinationInvestigation methodCL for TE [dB/facet]CL for TM [dB/facet]Coupling bandwidth for TE [1 dB]Alignment tolerance [1 dB]14Lensed fiber / Single layer SSCExperiment2.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}–100 nm –1Lensed fiber / Bilayer SSCExperiment1.7 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–– ± 1 μm (Horizontal)16Lensed fiber / Bilayer SSCExperiment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${0}{\text{.5 (}}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–35 nm–17Lensed fiber / Tri-layer SSCSimulation1.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {775}\;{\mathrm{nm)}}$$\end{document}0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}7.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {775}\;{\mathrm{nm)}}$$\end{document}3.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550 }\;{\mathrm{nm)}}$$\end{document}––44UHNAF / Multi-layer SSCSimulation1.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–120 nm–18UHNAF / SiN + Tri-layer SSCExperiment1.3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document}–35 nm ± 2 μm (Vertical) ± 2 μm (Horizontal)15Laser diode / SiON + Single layer SSCSimulation–2.4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {850}\;{\mathrm{nm)}}$$\end{document}3.2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1310}\;{\mathrm{nm)}}$$\end{document}2.7 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}––45UHNAF / SiON + Bilayer SSCExperiment0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}0.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}100 nm ± 1.3 μm (Vertical) ± 1 μm (Horizontal)46UHNAF / Bident + Bilayer SSCExperiment1.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}–35 nm ± 1 μm (Vertical) ± 1 μm (Horizontal)36 Lensed fiber / Cantilevered single layer SSCExperiment1.4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}1.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}100 nm–37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{SMF / SiO}}_{2}$$\end{document} + Single layer SSCExperiment0.9 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1300}\;{\mathrm{nm)}}$$\end{document}1.1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1300}\;{\mathrm{nm)}}$$\end{document}80 nm ± 2 μm (Vertical) ± 2.5 μm (Horizontal)47SMF / Three \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Si}}_{3} {\mathrm{N}}_{4}$$\end{document} + three-stage SSCSimulation0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}100 nm ± 1.7 μm (Vertical) ± 2 μm (Horizontal)This workDLMS-based FMT / TFLN RWGSimulation0.6 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}1.2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document}130 nm ± 1 μm (Vertical) ± 2.5 μm (Horizontal) ± 2 μm (Propagation)“–”: not mentioned in the article. UHNAF: ultra-high numerical aperture fiber. Comparison of the optical characteristics of the fiber-to-TFLN chip edge couplers. 1.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {775}\;{\mathrm{nm)}}$$\end{document} 0.5 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550}\;{\mathrm{nm)}}$$\end{document} 7.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {775}\;{\mathrm{nm)}}$$\end{document} 3.0 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1550 }\;{\mathrm{nm)}}$$\end{document} ± 2 μm (Vertical) ± 2 μm (Horizontal) 2.4 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {850}\;{\mathrm{nm)}}$$\end{document} 3.2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(}\uplambda = {1310}\;{\mathrm{nm)}}$$\end{document} 2.7 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\uplambda = 1550\;{\mathrm{nm}})$$\end{document} ± 1.3 μm (Vertical) ± 1 μm (Horizontal) ± 1 μm (Vertical) ± 1 μm (Horizontal) ± 2 μm (Vertical) ± 2.5 μm (Horizontal) ± 1.7 μm (Vertical) ± 2 μm (Horizontal) ± 1 μm (Vertical) ± 2.5 μm (Horizontal) ± 2 μm (Propagation) “–”: not mentioned in the article. UHNAF: ultra-high numerical aperture fiber. Compared with the aforementioned works that primarily rely on multi-stage tapered waveguides or complex SSC architectures, the DLMS-based FMT proposed in this work leads to polarization-insensitive performance at a single wavelength of 1550 nm, with CLs of 0.6 dB/facet for the TE mode and 1.2 dB/facet for the TM mode, while preserving a decent performance (CL < 1.9 dB/facet) over a wavelength span of 170 nm, demonstrating its potential for broadband and dual-polarization applications. More importantly, the DLMS-based FMT performs wavefront and mode-field engineering directly at the fiber–waveguide interface via a transmission-phase metasurface, without requiring complex and difficult-to-realize (or even impractical under certain process constraints) multilayer SSCs or long multi-stage tapered waveguides on the TFLN platform. Taken together with recent advances in on-chip polarization-diversity/polarization-insensitive edge couplers and fiber-side polarization management, such a DLMS front-end interface can be combined with PSR networks or other mode/polarization converters and possibly co-integrated with fiber-end metasurfaces, facilitating the construction of polarization-aware TFLN photonic integrated subsystems. Although the proposed edge coupler based on a DLMS has been constructed on a TFLN RWG, the underlying design methodology is rarely limited to specific optical waveguide materials. The proposed coupling scheme can be readily applied to other material platforms widely adopted in planar integrated optics, such as the commercially prevalent Si or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiN}}_{\mathrm{x}}$$\end{document} photonic platforms. Instead of using an inverse-taper-assisted mode size converter, Si or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiN}}_{\mathrm{x}}$$\end{document} waveguide facets with a width of 450 nm can be directly edge-coupled to a DLMS-installed fiber. In this case, the metasurface phase distribution of the DLMS is re-engineered in a way that it exactly matches the target guided modes of the two different photonic platforms. This will be another research topic. As a phase-engineered element, the DLMS inevitably gives rise to a wavelength-dependent phase distribution, which leads to slight chromatic behavior at the TFLN waveguide facet, manifesting as a gradual change of the focal position and spatial mode profile with wavelength. These chromatic spatial effects engender a progressive variations in the mode overlap between the fiber mode and the guided mode, which is quantitatively captured by the simulated coupling efficiency spectrum as shown in Figure 6c. Over the S + C + L bands of interest, the wavelength-induced variation in insertion loss for the TE polarization is limited to approximately 1.3 dB, indicating that the variations in coupling efficiency caused by DLMS spatial aberrations are practically permissible to facilitate broadband operation. In contrast, the additional temporal distortions (such as group delay and group-delay dispersion) caused by the subwavelength-thick, non-resonant dielectric metasurface are expected to be negligible relative to the intrinsic material and waveguide dispersion of the fiber–TFLN case. The DLMS coupler in this work has not been fabricated yet, but the process flow proposed for embodying it is predicated on fabrication techniques which were previously validated. The DLMS-installed functionalized fiber facet can be realized in two main steps. First, a 10.8-µm-thick \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} layer is grown on a Si wafer and used as the starting substrate for MS1. Using standard lithography and etching, the oxide film is divided into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} tethers and the MS1 pattern is defined. An SU-8 layer with a thickness of ~ 85 µm is then spin-coated on top. This layer protects MS1 during subsequent fiber bonding and allows the Gaussian beam emitted from the fiber to expand to a size comparable to that of MS1. A “dip and vibration” process is adopted to coat the cleaved facet of a SMF with viscous UV-epoxy, forming a uniform adhesive film with a thickness of ~ 30 µm50. To secure an optimal epoxy thickness, the fiber end is dipped into the epoxy, gently withdrawn, and immediately fixed in a vertical orientation using a fiber clamp. The vibration frequency and amplitude are controlled by the free length of the fiber outside the clamp and by the initial displacement applied to the fiber tip. The SMF is then motorized into precise contact with MS1 and the epoxy is cured under UV illumination. At the same time, the underlying Si substrate is selectively removed by wet etching. This process yields a freestanding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document}/MS1 stack and effectively bonds the 10.8-µm-thick \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document}/MS1 structure onto the fiber facet. For MS2, another Si wafer with a 20-µm-thick \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} layer is prepared, and MS2 is fabricated atop the oxide layer. A polymethyl methacrylate (PMMA) layer is spin-coated, followed by deposition of a polydimethylsiloxane (PDMS) handling layer. The sample is immersed in buffered oxide etchant (BOE) to remove the 20-µm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} layer and release a PDMS/PMMA/MS2 membrane from Si substrate. A thin adhesive layer is generated using a “dip and vibration” step. MS2 is “transfer-printed” onto the pre-fabricated MS1-on-fiber assembly51. A high-precision translation stage, rotation stage, and high-resolution imaging system are relied on to facilitate precise alignment between MS1 and MS2. By removing the PDMS and PMMA layers via simple peel-off and aceton-assisted dissolving, respectively, finally a stacked MS1/MS2 double-layer metasurface that is supported by a 10.8-µm-thick \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{SiO}}_{2}$$\end{document} spacer layer is obtained. A detailed step-by-step description of the fabrication procedure is provided in Fig. S7 of the Supporting Information. In view of the possible fabrication schemes as aforementioned, it should be admitted that the experimental realization and precise alignment of transfer-printed metasurface may be a future technical challenge. To date, no experimental demonstration of a-Si:H nanostructures transfer has been reported. Therefore, in this study, we provide only a detailed qualitative description of the potential transfer-based fabrication route and its associated challenges. This concept has been explicitly discussed as a prospective research direction, emphasizing the need for integrating high-resolution micromanipulators or automated alignment systems to enable sub-micron positioning accuracy in future implementations. Conclusion To improve the edge coupling efficiency between the SMF and TFLN chips, this study introduces a novel edge coupler based on a DLMS. Inspired by the beam-shaping function of orthogonally arranged cylindrical lenses, the DLMS structure generates a highly elliptical focal spot tailored to match the mode profile of a 1.95 μm-thick TFLN RWG, achieving a mode overlap efficiency of up to 87%. This design enables efficient and low-loss coupling at 1550 nm for both TE (0.6 dB/facet) and TM (1.2 dB/facet) modes. It surpasses the performance of previous approaches (Table 1) and maintains a CL below 1.9 dB/facet across a 170 nm bandwidth, highlighting its potential for broadband, dual-polarization applications. In addition, the coupler exhibits relaxed alignment tolerances (± 1 μm vertically, ± 2.5 μm laterally), significantly reducing packaging complexity and enhancing system robustness. Overall, this work pioneers the use of DLMS-based focal shaping for fiber-to-chip edge coupling on TFLN RWG, offering a low-loss, polarization-flexible, and broadband solution for next-generation photonic integration. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1