🔬 optics

Efficient Mode Conversion at 1064 nm via Bilayer Inverse Taper on Thin-Film Lithium Niobate

This paper presents a bilayer inverse taper structure on thin-film lithium niobate that achieves efficient fiber-to-chip coupling at 1064 nm with a measured loss of 1.9 dB per facet and a theoretical potential of 0.48 dB, offering a high-performance solution for on-chip optical interconnects.

Original authors: Ruidong Xue, Jobayer Hossain, Joshua Arnold, Jiuyi Zhang, Xiaofeng Zhu, Marco Moller de Freitas, Christopher J. Cullen, Shouyuan Shi, Peng Yao, Timothy Creazzo, Dennis Prather

Published 2026-08-04

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Ruidong Xue, Jobayer Hossain, Joshua Arnold, Jiuyi Zhang, Xiaofeng Zhu, Marco Moller de Freitas, Christopher J. Cullen, Shouyuan Shi, Peng Yao, Timothy Creazzo, Dennis Prather

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the internet as a massive, bustling city where information travels as tiny, invisible messengers called photons. For years, these messengers have been running on copper wires, but those wires are like old, narrow streets that get clogged and slow down the traffic. To fix this, scientists are building "photonic highways" on tiny chips, where light zips around at incredible speeds with almost no energy wasted. However, there's a tricky bottleneck: getting the light from the big, round optical fibers (the main roads) onto the tiny, flat waveguides on the chip (the side streets). It's like trying to pour a giant bucket of water into a thimble without spilling a drop. If the shapes don't match perfectly, the light scatters and is lost, making the whole system inefficient. This is especially hard when the light has a specific color, or wavelength, of 1064 nanometers, which is a sweet spot for making super-fast computers and sensors, but where current tools struggle to connect the dots.

This paper tells the story of how a team of researchers built a clever "mold" to solve this pouring problem for that specific color of light. They designed a special structure called a "bilayer inverse taper" on a thin slice of lithium niobate, a material known for being a superstar at handling light. Think of this taper as a funnel that starts wide and gets incredibly narrow, but in a very smart, two-layered way. Instead of just squeezing the light sideways like a standard funnel, their design squeezes it both sideways and up-and-down, gently guiding the light from the fat fiber into the skinny chip waveguide without spilling. They didn't just guess how to build it; they used computer simulations to design a funnel that changes shape gradually (like a slide that gets steeper only at the end) to keep the light happy and on track.

The team then built these funnels in a real lab and tested them. They found that their design works remarkably well. When they shined light through it, they lost only 1.9 decibels of signal strength per connection point. That's a huge improvement over previous attempts at this specific color of light, which often lost 3 decibels or more. While their computer models suggested they could get even lower—down to 0.48 decibels if everything were perfect—their real-world results are still a major step forward. They also checked how picky the device is about being lined up perfectly; they found it's a bit more sensitive to being moved up and down than side-to-side, but it still holds up well within a tiny margin of error. Most importantly, the device works consistently across a range of colors from 1055 to 1085 nanometers, meaning it's robust enough for real-world use. This work proves that we can efficiently connect our high-speed light highways to the tiny chips that will power the next generation of technology, opening the door for faster, more powerful devices that run on the 1064 nm wavelength.

Technical Summary: Efficient Mode Conversion at 1064 nm via Bilayer Inverse Taper on Thin-Film Lithium Niobate

Problem Statement
While thin-film lithium niobate (TFLN) has emerged as a superior platform for high-speed modulators and quantum sensors due to its nonlinear properties and low propagation losses, efficient fiber-to-chip coupling remains a significant bottleneck, particularly at the 1064 nm wavelength. Although low-loss coupling has been achieved in the C-band (1550 nm) and other ranges like 1310 nm and the visible spectrum, the 1064 nm regime presents unique challenges. The reduced wavelength alters modal confinement, fiber-to-chip mode mismatch, and adiabatic conversion conditions, rendering designs optimized for 1550 nm unsuitable without re-optimization. Furthermore, operating at 1064 nm is critical for RF-photonic systems and TFLN modulators, as it allows for a direct reduction in the half-wave voltage ( $V_\pi$ ) via the Pockels effect scaling ( $V_\pi \propto \lambda$ ). However, existing edge couplers at this wavelength have demonstrated relatively high insertion losses (e.g., ~3 dB per facet), necessitating a dedicated design and experimental investigation to enable high-efficiency integration.

Methodology
The authors propose and realize an on-chip optical mode converter based on a bilayer inverse taper structure specifically optimized for 1064 nm.

Design Strategy: Unlike simple scaling of 1550 nm designs, the team performed wavelength-specific simulations using a 2.5 $\mu$ m mode field diameter (MFD) Gaussian fiber mode. They utilized a 300 nm thick TFLN waveguide with a 150 nm etch depth. The design features a bilayer taper where the optical mode transitions from a fundamental fiber mode to a ridge-etched waveguide.
Nonlinear Taper Profile: To minimize propagation loss and ensure adiabatic transition, the authors employed a nonlinear taper profile for the second layer, defined by a power-law equation $y(x) = w_r + (w_l - w_r)(1 - x/L)^m$ . This approach slows the geometric variation in the initial section of the taper where the mode is weakly confined, suppressing non-adiabatic coupling.
Parameter Optimization: Using Lumerical FDTD simulations, the team systematically optimized four key parameters: input width ( $w$ ), taper length, the length ratio of the second layer ( $\gamma$ ), and the exponent ( $m$ ). The optimal configuration was identified as an input width of ~160 nm, a total taper length of 300 $\mu$ m, a second-layer length ratio of 0.3, and a nonlinear exponent of $m=0.5$ .
Fabrication: Devices were fabricated on 300 nm TFLN wafers. The process involved electron-beam lithography (EBL) with HSQ masks to define features as small as 30 nm at the taper tip. A two-step EBL process with critical alignment (<40 nm) was used for the bilayer structure. The devices were clad with 1.5 $\mu$ m of SiO2 and polished to expose the taper input facet.

Key Results

Coupling Loss: The fabricated devices demonstrated a coupling loss of 1.9 dB per facet for the TE mode. Theoretical simulations indicated that with ideal fabrication and a 2.5 $\mu$ m MFD lensed fiber, the loss could be reduced to 0.48 dB per facet.
Bandwidth: The device exhibited a 1 dB bandwidth spanning 1055–1085 nm, confirming stable performance across a broad spectral range around the target wavelength.
Alignment Tolerance: Polarization-dependent alignment tolerance was measured. For TE polarization, the lateral alignment tolerance was less sensitive than the vertical. The vertical alignment tolerance for a 1 dB loss was approximately 1.0 $\mu$ m, while the TM polarization state showed a narrower vertical tolerance of 0.74 $\mu$ m.
Loss Analysis: The discrepancy between the simulated (0.48 dB) and measured (1.9 dB) loss was attributed to realistic fabrication variations, including a 15 nm reduction in input width/height, the presence of the substrate stack (SiO2/BOX/Si), and a polished-facet angular deviation of approximately 4.5°, which collectively contributed to the excess loss.

Significance and Claims
The paper presents a high-efficiency bilayer inverse taper coupler specifically designed and optimized for 1064 nm on TFLN, addressing the limitations of previous designs such as the 3 dB loss reported by Zhang et al. [27] for a 1064 nm coupler. The authors position this work as a necessary step to enable the development of RF-photonic systems and high-performance TFLN modulators that leverage the lower $V_\pi$ achievable at shorter wavelengths.

The measured loss of 1.9 dB per facet is highlighted as the lowest experimental demonstration of mode coupling at 1064 nm using a bilayer taper structure to date. The results align with the observed literature trend where coupling loss improves with increasing wavelength (e.g., 3 dB at 775 nm, 3 dB at 1064 nm in prior work, 1 dB at 1310 nm, and 0.5 dB at 1550 nm). The authors conclude that while further improvements in fabrication tolerances and polishing angles are required to close the gap with simulation, the demonstrated device provides a feasible, high-efficiency solution for on-chip optical coupling at 1064 nm, paving the way for future integration of devices operating in this wavelength band.

More like this