🔬 optics

Hybrid-Integrated DFB-Laser-Coupled 1 * 8 Thin-Film Lithium Niobate Modulator Array for High-Speed Parallel Optical Transmitters

This paper presents a hybrid-integrated 1×8 thin-film lithium niobate modulator array passively coupled to a DFB laser, demonstrating a compact, high-speed parallel optical transmitter platform with over 40 GHz bandwidth, uniform power distribution, and low half-wave voltage suitable for advanced optical interconnects.

Original authors: Qiyue Hu, Junxia Zhou, Zhe Wang, Botao Fu, Jinming Chen, Yunpeng Song, Dewei Zhang, Yuheng Chen, Jinxin Huang, Min Wang, Jia Qi, Ya Cheng

Published 2026-05-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Qiyue Hu, Junxia Zhou, Zhe Wang, Botao Fu, Jinming Chen, Yunpeng Song, Dewei Zhang, Yuheng Chen, Jinxin Huang, Min Wang, Jia Qi, Ya Cheng

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 you are trying to send a massive amount of data from one computer to another, like streaming a thousand high-definition movies at once. To do this fast enough, you can't just use a single "pipe" (a single optical fiber); you need a whole fleet of pipes working in perfect sync. This is the challenge of parallel optical interconnects.

This paper describes a new, high-tech "traffic control center" built on a tiny chip that manages eight of these data pipes simultaneously. Here is how they did it, explained simply:

1. The Material: A Super-Skinny Crystal

The team built their device on a material called Thin-Film Lithium Niobate (TFLN). Think of this material as a super-efficient highway for light. Unlike older materials, this one is very thin (like a sheet of paper) but has special properties that allow it to change the speed of light incredibly fast when you apply electricity. This makes it perfect for high-speed data.

2. The Problem: One Light Source, Eight Lanes

The biggest hurdle in building these multi-lane systems is getting light from a single source (a laser) and splitting it evenly into eight separate lanes without losing power or messing up the timing.

The Solution: They built a 1 × 8 splitter right on the chip.
The Analogy: Imagine a single water hose connected to a complex system of pipes. The team designed a "smart junction" (using something called a cascaded MMI splitter) that takes the water from one hose and divides it into eight smaller hoses.
The Result: The water (light) comes out of all eight hoses at almost exactly the same pressure. The paper says the difference in power between the strongest and weakest lane is less than 10%, which is excellent uniformity.

3. The Modulators: The Fast Switches

Once the light is split, each of the eight lanes needs a switch to turn the data on and off (encoding the information).

The Design: They used Mach-Zehnder Modulators. Think of these as ultra-fast traffic lights that can blink on and off billions of times per second.
The Speed: These switches are so fast they can handle data speeds corresponding to a bandwidth of over 40 GHz. To put that in perspective, that's fast enough to handle the data traffic of a major city's internet in a split second.
Efficiency: They needed very little electrical power to flip these switches (about 3.6 to 3.8 volts), which is like using a small battery instead of a car battery to run a heavy machine.

4. The Connection: Plugging in the Laser

A modulator chip is useless without a light source. Usually, connecting a laser to a chip is like trying to plug a thick garden hose into a tiny straw; it's hard to align and you lose a lot of water (light) in the process.

The Hybrid Approach: Instead of trying to grow the laser on the chip (which is very difficult), they took a standard, commercial laser (a DFB laser) and physically pressed it against the chip's input.
The "Spot-Size Converter": To make the connection smoother, they built a special funnel (a Spot-Size Converter) at the entrance of the chip. This funnel gently expands the tiny light beam from the laser so it fits perfectly into the chip's waveguide, minimizing waste.
The Result: When they connected the laser, the system worked. Yes, there was a little bit of light lost at the connection point (about 5 dB), but the eight lanes remained perfectly balanced. The "traffic" didn't get jammed in one lane while another was empty.

5. The Bottom Line

The researchers successfully built a single chip that:

Takes light from one laser.
Splits it evenly into eight lanes.
Modulates (encodes data onto) all eight lanes simultaneously at super-high speeds.
Keeps the performance consistent across all lanes.

Why does this matter?
The paper claims this is a major step toward making compact, high-speed optical transmitters. In simple terms, they proved it is possible to build a tiny, efficient "data factory" on a chip that can handle massive amounts of information in parallel, which is exactly what future AI and cloud computing centers will need to keep up with demand. They didn't just build a single switch; they built a synchronized fleet of eight switches that work together perfectly.

Technical Summary: Hybrid-Integrated DFB-Laser-Coupled 1 × 8 Thin-Film Lithium Niobate Modulator Array

Problem Statement
The rapid expansion of artificial intelligence, cloud computing, and high-performance computing has intensified the demand for optical interconnects with higher aggregate capacity and improved energy efficiency. While thin-film lithium niobate (TFLN) has emerged as a leading platform for single-channel high-speed modulators due to its large Pockels coefficient and low loss, scaling these devices into practical transmitter arrays presents significant challenges. Key hurdles include achieving uniform optical power distribution across multiple channels, ensuring matched modulation characteristics between channels, and integrating commercial laser sources (such as DFB lasers) into the TFLN chip via manufacturable packaging processes. Direct heterogeneous integration of III-V gain media on lithium niobate remains technically complex; therefore, a hybrid integration approach using passive edge coupling is a more immediate practical route.

Methodology and Device Design
The authors designed and fabricated a monolithic 1 × 8 TFLN electro-optic modulator array on an x-cut lithium-niobate-on-insulator (LNOI) wafer. The device architecture integrates several critical components:

Power Distribution: A three-stage cascaded 1 × 2 multimode-interference (MMI) splitter was employed to distribute input light uniformly to eight arms. This topology was chosen over a single-stage 1 × 8 splitter to reduce sensitivity to fabrication dimensional errors and wavelength drift.
Modulation Units: Each of the eight arms features a traveling-wave Mach-Zehnder modulator (MZM) with a 7 mm modulation length.
Electrode Design: Ground-signal-ground coplanar traveling-wave electrodes were optimized to balance modulation efficiency and optical absorption. An electrode gap of 6.5 µm was selected to achieve a favorable trade-off between the half-wave voltage-length product ( $V_\pi L$ ) and optical loss. The design incorporates microwave-optical group-velocity matching (via a 2.5 µm SiO₂ cladding), impedance matching (targeting 50 Ω), and on-chip 50 Ω terminations to suppress reflections.
Coupling Interface: Spot-size converters (SSCs) utilizing SiON-based waveguides were designed to match the mode field of the TFLN rib waveguide to that of UHNA7 fibers, facilitating both fiber-to-chip coupling and passive butt-coupling to DFB lasers.
Fabrication: The waveguides were fabricated using femtosecond-laser-assisted chemical mechanical polishing (PLACE). The electrodes were patterned via femtosecond laser lithography combined with wet chemical etching, followed by the deposition of Au/Ti electrodes and NiCr termination loads.

Key Results
Experimental characterization of the fabricated device yielded the following performance metrics:

Power Uniformity: The cascaded MMI splitter demonstrated high channel-to-channel uniformity, with a maximum normalized power deviation of 9.7% across the eight output ports.
Bandwidth and Efficiency: All eight channels exhibited electro-optic 3 dB bandwidths exceeding 40 GHz. The measured half-wave voltages ( $V_\pi$ ) ranged from 3.60 V to 3.83 V, corresponding to a $V_\pi L$ product of 2.52–2.68 V·cm.
Extinction Ratio: The modulators achieved an extinction ratio of approximately 25 dB.
Insertion Loss: The bare-chip insertion loss for the eight channels ranged from 15.19 dB to 16.55 dB.
Hybrid Integration: A commercial 1550 nm distributed-feedback (DFB) laser was passively butt-coupled to the chip. This integration introduced an additional coupling loss of approximately 5 dB, resulting in a total hybrid-integrated insertion loss of 20.19–21.55 dB. Crucially, the inter-channel insertion-loss difference remained at 1.36 dB after bonding, indicating that the laser coupling did not degrade the array's uniformity.

Significance and Claims
The paper establishes a practical TFLN-based multi-channel modulator platform that addresses the array-level integration challenges necessary for high-speed parallel optical interconnects. By successfully demonstrating a hybrid-integrated 1 × 8 transmitter chain—from a DFB laser source through on-chip power splitting to parallel electro-optic modulation—the work validates a scalable route toward compact optical transmitters. The authors claim that this device represents a significant step toward practical optical engines, offering a manufacturable alternative to complex heterogeneous integration while maintaining high bandwidth and channel uniformity. Future improvements are identified as focusing on reducing the DFB-to-chip coupling loss through mode-matched SSCs and improved alignment, as well as eventual co-packaging with driver electronics.