Location： Home page

AI Artificial Intelligence Computing Center

With the explosive growth in demand for AI computing power, AI intelligent computing centers are gradually becoming the cornerstone of the next-generation data infrastructure. In high-density optical interconnect architectures, traditional electrical interconnections can no longer meet the requirements for high speed, low power consumption, and high bandwidth. High-speed optical interconnect solutions based on passive optical components are emerging as a key pathway to enhance the computing density and energy efficiency ratio of AI computing centers. Among these, passive optical products such as high-channel fiber FAUs, lensed FAUs (fiber array units with lenses), polarization optical components, and small-mode-field-radius fiber assemblies play an important role in providing critical technological support.

The lensed FAU, by integrating micro-lens structures onto the fiber end face, can significantly enhance optical coupling efficiency, reduce insertion loss, and achieve greater packaging compatibility. In the connection between silicon photonics chips and fiber arrays at AI computing centers, it can effectively optimize optical alignment tolerances, reduce packaging complexity, and improve the yield and reliability of modular optical interconnects.

Polarization optical components play a critical role in controlling signal quality in AI-enabled optical communication links. By employing polarization-maintaining and polarization-division multiplexing technologies, it is possible to achieve efficient transmission of multi-channel parallel optical signals, thereby enhancing data bandwidth utilization, reducing system noise and crosstalk, and ensuring high-speed, stable operation for AI training and inference tasks.

In addition, small-mode-field-radius fiber (SMF) products enable higher optical power density and coupling precision, making them well-suited to meet the high-bandwidth interconnection requirements between high-speed silicon photonic chips and miniature photonic devices. These fibers demonstrate outstanding performance in short-distance interconnections and high-power coupling, providing a low-loss, highly stable transmission foundation for optical interconnect networks in AI computing centers.

Meanwhile, integrated passive connection solutions based on thin-film optical filtering (TFF) technology are becoming key to solving the problems of insertion loss, integration, and cost in high-speed optical modules. This solution uses the Z-BLOCK series of passive devices as its core, achieving a performance leap through optical path redesign. It employs integrated coupling and packaging of components such as receptacles, collimators, and lens arrays, completing full optical path calibration before shipment. Optical module manufacturers can directly package the chips, achieving "plug and play."

Its core advantage lies in breaking through the bottlenecks of traditional silicon-based optical waveguide (AWG) technology: TFF theoretical coupling efficiency reaches 96%, insertion loss is less than half that of AWG, channel consistency is superior, and wide temperature range characteristics meet industrial-grade operation requirements. Coupled with a precision manufacturing process achieving repeatability accuracy of 2σ≤0.05dB, it can accurately control the beam coordinates and parallelism of the optical path, adapting to rapid switching between CWDM/LWDM and other multi-wavelength schemes.

In summary, passive optical components based on Lensed FAU, polarization optics, small mode field radius, and integrated Z-BLOCK technology are accelerating the transformation of AI intelligent computing centers from electrical interconnects to optical interconnects, providing crucial support for future high-efficiency, high-bandwidth, and low-latency intelligent computing infrastructure.