Silicon-based photonic integrated circuits have revolutionized the field of photonics, with extensive applications in data communication, computing, sensing, and signal processing.1 Photonics integration technology is compatible with traditional complementary metal-oxide-semiconductor (CMOS) fabrication techniques, and silicon-on-insulator (SOI), as a currently relatively mature platform, has reached the capability for large-scale production and practical applications.2 However, silicon photonics chips manufactured on the SOI platform often feature waveguide mode dimensions at the submicrometer scale. This discrepancy in mode dimensions poses a significant challenge when directly coupling with conventional optical fiber modes, leading to substantial coupling losses.3 Consequently, couplers that serve as efficient input and output interfaces between chips and optical fibers have garnered widespread attention among researchers.

Over the past few decades, researchers have been dedicated to designing couplers that are highly efficient, tolerant of fabrication variations, broad bandwidth, and low polarization dependence.3 Presently, two common solutions for achieving efficient coupling between optical fibers and photonic chips are vertical coupling and edge coupling.3 Vertical coupling, based on diffraction grating principles, offers advantages such as large fabrication tolerances and ease of photolithography.4 However, due to inherent limitations, vertical coupling technologies tend to exhibit lower performance in terms of bandwidth, coupling efficiency, and polarization dependence. On the other hand, edge coupling gradually enlarges the mode by employing tapered waveguides, aligning them with optical fiber modes and consequently providing high coupling efficiency, broad bandwidth, and low polarization dependence.5

The essence of edge couplers lies in the use of inverse taper waveguides, reducing the confinement of modes at the waveguide’s end, increasing the effective cross section, and decreasing the effective refractive index to achieve efficient coupling. To mitigate mode losses during the tapering process, it is often necessary to use tapers longer than $100\mu \mathrm{m}$ to achieve adiabatic transformation.6 Researchers have explored various taper geometries to achieve shorter adiabatic transitions, and multitaper configurations have demonstrated enhanced performance and increased design flexibility compared with single tapers.7^–9 In addition, researchers have leveraged low-index materials as auxiliary cladding layers to gradually transition the mode into the auxiliary cladding during the tapering process, resulting in higher coupling efficiency.10^,11 Common auxiliary cladding materials include polymers10 and silicon nitride.11 Breakthroughs in fabrication technologies have further expanded the design possibilities for edge couplers, including multiple layers and suspended structures,12^,13 thereby enhancing their performance. In recent years, subwavelength grating (SWG) structures have shown exceptional performance in silicon-based device designs.14 Incorporating SWG structures into edge couplers has yielded significant performance improvements,15^,16 Most prior research on edge couplers has been concentrated in the C-band, where they have demonstrated outstanding performance. The growing industrialization of the O-band for high-speed data communication presents new challenges. In the O-band, waveguides exhibit stronger birefringence, resulting in higher polarization-dependent loss (PDL), making it more challenging to achieve low PDL. At present, researchers are endeavoring to diminish PDL in the O-band. Currently, to mitigate polarization-related losses in the O-band, researchers commonly utilize the following approaches: incorporating low-index materials such as silicon nitride17^,18 or integrating SWG structures.19^,20 Nevertheless, these methods often entail additional fabrication steps or impose more stringent process requirements.

In this paper, we present what we believe is a novel approach based on the bilayer and double-tip edge coupler for achieving efficient coupling between optical fibers and chips in the O-band while minimizing PDL. By employing multiple etchings, we reduce the waveguide’s longitudinal mode confinement, achieving high coupling efficiency for both transverse electric (TE) and transverse magnetic (TM) modes. Through experimentation, we validate the performance of the edge coupler across the entire O-band, with coupling losses below 2 dB/facet for both polarization directions when coupled with a lensed optical fiber with a mode field diameter (MFD) of $4\mu \mathrm{m}$.

The schematic of the bilayer and double-tip edge coupler is shown in Fig. 1(a). The edge coupler is designed on the SOI platform with a $2\text{-}\mu \mathrm{m}$-thick buried oxide layer and a 220-nm-thick device silicon layer. The bilayer and double-tip edge coupler is composed of two inverse bilayer taper waveguides and a dual-polarization multimode interference (MMI) coupler, as shown in Fig. 1(b). The inverse bilayer taper waveguide consists of three parts: a 150-nm-high waveguide, a 150- to 220-nm-high waveguide converter, and a 220-nm-high waveguide.

The light coupling process begins with a double-tip waveguide, each spaced at a distance of $g$, with a height of 150 nm, a width of $W_{\text{tip}}$, and a length of $L_1$. These waveguides gradually transform the optical fiber mode into a waveguide mode composed of a double-tip waveguide, efficiently transferring most of the energy into the waveguide. Subsequently, a 150- to 220-nm-high waveguide converter extending over a length of $L_2$ facilitates the mode conversion between two waveguides of different heights. A cross-sectional view during this height transition process is illustrated in Fig. 1(c). Following the transition section, a waveguide with a height of 220 nm and a length of $L_3$ gradually adjusts the waveguide spacing and width in a linear manner to $d$ and $W_0$, respectively, ultimately feeding into a single-mode waveguide output through an MMI coupler. Using an MMI coupler for combining provides a compact solution compared with traditional adiabatic tapering couplers. In this design, a wedge-shaped multimode waveguide region is employed instead of the conventional rectangular waveguide. The wedge-shaped multimode waveguide structure is designed to balance the effective mode overlap and propagation characteristics for both TE and TM modes. By introducing a taper, the width of the multimode waveguide gradually transitions, allowing smoother mode transformation and reducing abrupt changes in mode confinement. This gradual transition minimizes modal mismatches and reflections that could lead to polarization-dependent losses. In addition, the taper provides greater flexibility in the design of mode evolution paths for TE and TM modes, enabling near-equal coupling efficiencies for both polarizations. This approach not only improves polarization insensitivity but also ensures a compact device footprint, which is critical for integrated photonic applications.

The coupling loss in the edge coupler primarily arises from the mode field mismatch between the optical fiber mode and the chip’s input mode. The mode overlap can be defined as follows:$$\text{Overlap}=\frac{{|\int E_1{E}_2\mathrm{d}A|}^2}{\int |E_1{|}^2\mathrm{d}A\int |E_2{|}^2\mathrm{d}A},$$(1)where $E_1$ and $E_2$ are the complex electric field amplitudes of the optical fiber mode and the on-chip waveguide mode, respectively. To enhance coupling efficiency, we need to design waveguide modes that closely match the input optical fiber mode. As shown in Figs. 2(a) and 2(b), we computed the TE and TM mode field distributions at 1310 nm for two different height waveguide structures through simulation analysis. From Fig. 2(a), it is evident that the effective mode area of the TE mode is larger than that of the TM mode. This is due to the strong birefringence characteristics of the waveguide at 1310 nm, where the 220-nm-high waveguide exhibits stronger confinement for the TM mode, resulting in a smaller mode size. Therefore, we set the waveguide height to 150 nm to reduce its confinement for the TM mode. As shown in the simulation results in Fig. 2(b), it can be observed that the effective mode areas of both TE and TM modes are roughly matched at this configuration.

To quantitatively analyze the differences between the two waveguide structures and achieve higher coupling efficiency, we optimized the design with respect to the tip waveguide width and spacing, and calculated the mode overlap with a $4\text{-}\mu \mathrm{m}$ MFD lensed optical fiber mode. First, we conducted optimization simulations for the tip waveguide width, as shown in Fig. 3(a). In the figure, solid lines and dashed lines represent waveguide structures with heights of 150 and 220 nm, respectively, whereas red and blue lines represent TE and TM modes, respectively. From the simulation results, it is evident that the solid red line has an overall overlap higher than 20% compared with the dashed red line, indicating the effectiveness of our design approach. When the waveguide structure height is 220 nm, there is a significant gap in the overlap between TE and TM modes, making it challenging to achieve both high-efficiency coupling and PDL simultaneously. However, when the waveguide structure height is 150 nm, the overlap of both TE and TM modes is close, exceeding 80%, within the tip width range of 100 to 150 nm. Moreover, this design exhibits a gradual change in mode overlap with respect to tip width, improving manufacturing process tolerance. Considering manufacturing capabilities, we selected a tip width of 130 nm.

Next, we conducted optimization simulations for the tip waveguide spacing, as shown in Fig. 3(b). The curve meanings are consistent with Fig. 3(a), and we simulated and calculated the mode overlap for TE and TM modes of waveguide structures with heights of 150 and 220 nm as a function of waveguide spacing. Simulation results indicate that, for the 220 nm waveguide structure, there is still a gap of over 25% in the overlap between TE and TM modes. In contrast, the overlap integral gap between TE and TM modes for the 150-nm waveguide structure remains within 5%. It is evident from the figure that as the waveguide spacing changes, the mode overlap gradually increases and then decreases. We chose a spacing of $1.6\mu \mathrm{m}$ as our design parameter, as it falls within a range that offers substantial manufacturing tolerance.

Based on the analysis and optimization of the tip waveguide structure as described above, we determined the width and spacing of the tip waveguide. Then, we systematically optimized each component of the device to minimize PDL and coupling loss. For the tapered MMI coupler, the most critical parameters were the length of the tapered multimode waveguide ($L_m$) and the two widths ($W_{m1}$ and $W_{m2}$). These were first optimized to minimize PDL and losses, followed by refinement of other related parameters. For the 150- to 220-nm-high waveguide converter, we optimized the length ($L_2$) to ensure low-loss transitions for both TE and TM modes during the waveguide height conversion.

After defining the parameters of the individual components, we connected them and optimized the lengths of the transition regions among sections using three-dimensional finite-difference time-domain (3D-FDTD) simulations. This ensured seamless integration and minimized additional PDL and coupling losses. The final device parameters, summarized in Table 1, resulted in an overall device length of just $86.5\mu \mathrm{m}$, demonstrating the effectiveness of this systematic optimization process.

Figures 4(a) and 4(b) display simulated mode field propagation at 1310 nm for different polarizations of the input light source. The insets in the figures, from right to left, represent the input to the chip, the waveguide transition from 150 to 220 nm in height, and the cross-sectional mode field distribution before and after the MMI coupler. The mode field propagation shown in Figs. 4(a) and 4(b) illustrates the expected mode conversion and coupling, where most of the input light is converted into waveguide modes propagating within the chip. Figure 5 illustrates the losses of our designed structures with a tip height of 150 nm and conventional structures with a height of 220 nm in the O-band range. In Fig. 6, solid and dashed lines represent waveguide structures with heights of 150 and 220 nm, respectively, and red and blue lines correspond to TE and TM modes. As indicated by the dashed lines, the traditional structure exhibits TM mode coupling losses over 1.5 dB higher than TE mode losses across the entire O-band range, with significant variation with wavelength. In our designed edge coupler, TM mode coupling losses are $<1.24\mathrm{dB}$, and TE mode losses are $<1.09\mathrm{dB}$. The higher TM coupling losses are attributed to significant transmission losses for TM modes during the waveguide height transition, where the optical field dissipates into the cladding, as shown in Fig. 4(b). Overall, our designed edge coupler outperforms traditional structures in terms of losses and PDL in the O-band range.

The device we designed was fabricated on an SOI wafer with a top silicon layer thickness of 220 nm and a buried ${\mathrm{SiO}}_2$ layer thickness of $2\mu \mathrm{m}$. Our design was divided into two layers, created through multiple etching steps. Using electron beam lithography (EBL), we initially spun ZEP520 electron beam resist onto the wafer’s surface. Then, the alignment pattern was transferred to the EBL resist using EBL. Subsequently, a $10/50\mathrm{nm}$ Cr/Au metal layer was deposited by electron beam evaporation, followed by lift-off to remove excess metal, leaving the patterned structure. In a second EBL step, the first-layer structure was transferred to an electron-beam resist layer. To transfer the structure to the silicon layer, inductively coupled plasma (ICP) etching was employed to etch the top silicon layer by 220 nm. In another round of EBL and ICP processes, the second-layer structure was transferred to the silicon layer and etched by 70 nm. Finally, a $2\text{-}\mu \mathrm{m}$-thick ${\mathrm{SiO}}_2$ cladding layer was deposited on top using plasma-enhanced chemical vapor deposition. Figure 6(a) displays the optical microscope images of the fabricated chip, with the inset showing an enlarged view of the edge coupler. Figures 6(b)–6(d) show scanning electron microscope (SEM) images of the regions marked in the inset of Fig. 6(a).

To characterize the performance of the edge coupler we fabricated, we utilized a tunable laser source with a wavelength range from 1260 to 1360 nm as the input light source. The optical power meter was used to measure the output chip-coupled power. The polarization of the input light was controlled for TE and TM modes using a polarization controller. The coupling setup consisted of two precision six-axis alignment stages and a chip holder, as illustrated in the inset of Fig. 7. We fixed a $4\text{-}\mu \mathrm{m}$ MFD lensed optical fiber on the fiber holders on both sides and adjusted the six-axis alignment stages to maximize the received power at the power meter. By tuning the input light wavelength and the polarization controller, we characterized the device’s TE and TM coupling losses across the entire O-band. Figure 8 presents the testing and simulation performance of the device, with solid and dashed lines representing experimental and simulation results, and red and blue denoting TE and TM mode results, respectively. It is observed that the experimental results are slightly worse than the simulation results but exhibit a consistent trend, showcasing impressive performance regarding polarization-related losses. At 1310 nm, TE and TM mode coupling losses were measured as 1.18 and $1.46\mathrm{dB}/\text{facet}$, respectively. Some noise in the spectrum is due to oscillations in the coupling setup during testing and reflections between the optical fiber and the chip. Across the entire O-band range, TE and TM mode coupling losses were $<1.52$ and $2\mathrm{dB}/\text{facet}$, respectively.

In Table 2, we present the performance of O-band edge coupler devices in recent years and compare them with our results. Our proposed structure achieves competitive coupling loss and PDL without the need for additional fabrication steps or more stringent process requirements, demonstrating very promising performance.

We have designed and fabricated a double-layer, double-taper edge coupler that exhibits low losses and polarization-related losses within the O-band range. In this work, we conducted theoretical simulations to analyze the significant polarization-related losses in traditional edge couplers caused by the strong birefringence properties of the O-band. To mitigate these losses, we introduced a double-layer structure. We manufactured this edge coupler using an EBL process and characterized it using a lensed optical fiber ($4\text{-}\mu \mathrm{m}$ MFD). The experimental results showed that the device achieved TE and TM mode coupling losses below 1.52 and $2\mathrm{dB}/\text{facet}$, respectively, across the entire O-band range. Our designed device offers significant process tolerance and is compatible with the standard SOI platform, enabling large-scale production. It provides a versatile solution for fiber-edge coupling in the O-band, making it applicable to various applications.

Building on these results, the designed coupler demonstrates significant potential for practical implementation in advanced optical communication systems.24 It offers substantial advantages in high-speed optical modules, such as 400G and 800G systems, where advanced modulation formats such as PM-QPSK and PM-16QAM require low insertion loss and balanced polarization performance. In addition, its low PDL and high coupling efficiency make it ideal for polarization-division multiplexing systems,25 enhancing data capacity by transmitting signals on orthogonal polarizations. These features enable efficient fiber-to-chip coupling and broaden the coupler’s applicability in data center interconnects and metropolitan area networks, paving the way for future advancements in silicon photonics.

Yuanjian Wan received his BS degree from Nanjing Tech University, Nanjing, China, in 2020. Currently, he is pursuing his PhD at Wuhan National Laboratory for Optoelectronics and School of Optical and Electronic Information at Huazhong University of Science and Technology, Wuhan, China. His research interests include silicon photonics and optical computing.

Yu Zhang received his BS degree in optoelectronics from Huazhong University of Science and Technology, Wuhan, China, in 2010, and his PhD in electronic and computer engineering from The Hong Kong University of Science and Technology, Hong Kong, in 2016. He worked as a postdoc and an assistant project scientist in electrical and computer engineering at the University of California, Davis, from 2016 to 2020. Currently, he is a professor at Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, China. His current research interest focuses on 3D integrated silicon photonics, large-scale optical interposers, heterogeneous integration of lasers, and amplifier and intra-chip optical interconnect devices.

Jian Wang received his PhD in physical electronics from Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, Wuhan, China, in 2008. He worked as a postdoctoral research associate in the Optical Communications Laboratory, University of Southern California, Los Angeles, USA, from 2009 to 2011. Currently, he is a professor at Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology. He is the vice director of Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology. He leads the Multi-Dimensional Photonics Laboratory. His research interests include optical communications, optical signal processing, silicon photonics, photonic integration, orbital angular momentum, and structured light. He has authored or co-authored more than 300 refereed international journal papers in such journals as Science, Science Advances, Nature Photonics, Nature Nanotechnology, Nature Communications, Light: Science and Applications, Physical Review Letters, etc. He has authored and co-authored more than 150 international conference papers for OFC, ECOC, CLEO, etc. He has also given more than 130 tutorial/keynote/invited talks at international conferences, including a plenary talk at Photonics Asia 2023, an invited talk at OFC 2014, a tutorial talk at OFC 2016, and an invited talk at OFC 2024. He is currently an IEEE fellow, Optica fellow, SPIE fellow, and COS fellow.