Polarization Maintaining Fiber Array for CPO and Silicon Photonics: Hirundo PM FA Solutions for AI Supercomputing
Why Fiber Arrays Are Becoming the Backbone of AI Optical Interconnects
The explosive growth of AI supercomputing clusters has pushed optical networks to a new limit. Inside a modern AI data center, hundreds of thousands of optical links carry data between GPUs, switches and compute nodes, and every one of those links begins and ends with a precisely engineered interface between light and fiber. That interface is the optical fiber array (FA).
As data rates climb from 400G to 800G and toward 1.6T, and as co-packaged optics (CPO) and silicon photonics move from laboratory concepts into volume deployment, the fiber array is no longer a passive afterthought. It has become a performance-critical component that determines insertion loss, return loss, channel uniformity and ultimately the yield of the entire optical module. Hirundo has focused on this exact challenge, developing polarization-maintaining fiber arrays and MT fiber arrays that deliver micron-level alignment accuracy for the next generation of AI optical infrastructure.
What Is a Fiber Array?
A fiber array is a passive optical device in which multiple optical fibers are precisely positioned and permanently fixed on a common substrate, typically a V-groove silicon or glass chip, and then polished to an optical-grade end face. The array holds each fiber at an exact pitch, usually 127 µm or 250 µm, so that the fiber cores align with the emitter or detector elements of an optical chip.
The precision of this alignment is measured in micrometers and even sub-micrometers. A misalignment of just one micrometer can translate into a substantial coupling loss, which is why the manufacturing of fiber arrays is often described as the art of micrometer-level control.
Polarization-Maintaining Fiber Array (PM FA)
A polarization-maintaining fiber array (PM FA) is built from multiple polarization-maintaining fibers fixed through a specialized packaging process. Unlike a standard single-mode FA, a PM FA must preserve the polarization state of the transmitted light, which means the slow axis of every individual fiber must be oriented with extreme accuracy relative to the V-groove and to each other.
This capability matters wherever polarization-sensitive signals are involved. In coherent optical communication, the integrity of the polarization state directly affects the signal-to-noise ratio. In fiber optic sensing, PM arrays are used to build high-precision interferometric sensors. In laser technology, they act as the critical coupling interface that delivers stable, linearly polarized light into and out of the system.
Hirundo PM fiber arrays are characterized by excellent polarization preservation, high arrangement accuracy and strong long-term stability, making them suitable for integration into complex modules where ordinary arrays would introduce unacceptable polarization crosstalk.
PM MT-FA: Multi-Fiber Termination Fiber Array
MT-FA stands for Multi-Fiber Termination Fiber Array. It combines an array of optical fibers with an MT ferrule, a precision multi-fiber ceramic or plastic insert that performs the terminal connection and integration of many fibers at once. When polarization-maintaining fiber is used, the result is a PM MT-FA, which supports parallel transmission of multiple polarization-stable optical signals in a single compact interface.
This is precisely the format demanded by high-speed, high-density optical transceiver modules. A PM MT-FA allows 4, 8, 12 or more channels to be coupled simultaneously while keeping the polarization state intact, dramatically improving the capacity and efficiency of the optical communication system without expanding its footprint.
Key Performance Metrics of Hirundo Fiber Arrays
| Parameter | Typical Specification | System Benefit |
|---|---|---|
| Insertion loss (per channel) | < 1.5 dB | Low-loss optical coupling |
| Alignment error | < 0.5 µm | High coupling yield |
| Return loss (8° angled, APC) | > 55 dB | Suppressed back-reflection |
| Polarization extinction ratio | ≥ 20 dB | Stable polarization state |
| Channel count | 1CH to 64CH | Flexible port matching |
| Operating temperature | -40 °C to +85 °C | Long-term reliability |
These figures are not marketing abstractions. They are the difference between a module that passes yield on the first attempt and one that requires repeated rework, and they are validated through full-link testing during production.
Applications in High-Speed Optical Modules
QSFP+ PSM4
PSM4 uses four-channel single-mode parallel transmission, operating at 4 × 25G NRZ or 4 × 28G PAM4. The FA-MT aligns the VCSEL or photodiode array with the four-core fiber with an alignment error below 0.5 µm and a single-channel insertion loss below 1.5 dB. The integrated MT ferrule enables compact deployment of 4, 8 or 12 fibers in a confined space, perfectly suited to the miniaturized design of QSFP+ modules. Temperature-induced optical path shift is minimized through metallized packaging or low-CTE materials such as glass substrates.
40G and 100G Active Optical Modules
Lasers such as VCSEL, DFB and EML, and photodetectors, must be precisely coupled to single-mode or multimode fibers through a fiber alignment mechanism to guarantee low-loss transmission. Typical insertion loss requirements are below 1.5 dB for 40G SR4 and PSM4 multimode links, and below 2.0 dB for 100G LR4 and ER4 single-mode links. Multi-channel parallel transmission is supported across configurations including 4 × 10G MMF with a 4-core FA, 4 × 25G SMF with a 4-core MT ferrule, and 4 × 25G wavelength multiplexing with a single-core input FA.
400G, 800G, 1.6T and CPO Architectures
At 400G, 800G and 1.6T, the fiber array must support the parallel transmission of dozens or even hundreds of optical signals, accurately guiding high-speed light generated by a silicon photonic chip into the external optical link, and efficiently delivering incoming light back into the chip. In effect, the array becomes the signal relay station between the silicon photonic die and the CPO package.
Fiber Arrays as the Optical Bridge for CPO and Silicon Photonics
CPO technology derives its core advantage from near-package transmission, which shortens the physical distance between the optical module and the computing chip and thereby reduces latency. Fiber arrays, with their compact structural design, can be integrated directly into the CPO package, creating a short-distance direct connection between the silicon photonic chip and the optical module. Compared with traditional fiber routing, this significantly reduces optical signal transmission loss, further amplifying the low-latency advantage of CPO and providing stutter-free transmission for high-frequency data interaction in AI supercomputing.
Equally important is scalability. AI supercomputing centers place stringent demands on large-scale deployment and long-term stability of optical links. Hirundo fiber arrays are engineered for high reliability and high consistency, enabling mass production with minimal performance deviation, while their flexible array specification design can be adapted to the port count of CPO and silicon photonics chips, reserving expansion space for future computing power upgrades.
Customization: V-Groove Precision and End-Face Angles
Hirundo specializes in supporting 400G, 800G and 1.6T optical modules for data centers and offers full-spectrum customizable fiber array solutions. Micron-level precision V-groove manufacturing allows flexible customization of 1CH to 64CH multi-channel designs with tightly controlled pitch tolerances, which dramatically improves coupling yield.
End-face geometry is equally customizable. Hirundo offers arrays with 8°, 0°, 45° and other custom angles to meet diverse customer requirements, matching SR and DR parallel modules as well as high-speed silicon photonic modules that require vertical coupling. Self-developed detachable CPO-specific FAs and polarization-maintaining special FAs are compatible with innovative co-packaged optics architectures, advancing new-generation optical interconnection for AI intelligent computing centers.
Why Hirundo
Hirundo approaches fiber arrays as a precision manufacturing discipline rather than a commodity. Fully automatic alignment processes keep polarization axis alignment error within one degree, while a full-link test regime covering optical time-domain reflectometry, polarization analysis and end-face inspection ensures that every array leaving the line meets its specification. Custom fiber types and connector options are supported, and the platform is designed for the volume, consistency and reliability that hyperscale AI deployments require.
Conclusion
As optical interconnect moves closer to the chip, the fiber array has become the optical bridge linking silicon photonics, CPO and AI supercomputing. Polarization-maintaining FA and PM MT-FA products from Hirundo deliver the combination that matters most in this environment: sub-micron alignment, low insertion loss, high return loss, stable polarization performance and the customization flexibility to match any port configuration. In a world where computing power is increasingly limited by the optical link, that precision is what keeps AI infrastructure moving.
