Artilux, a leading provider of germanium-silicon photonics technology, and Worldchip, a top ASIC and chiplet design service provider in the AI computing field, announced a strategic collaboration today (25th) to focus on developing an ultra-low-power photonic interconnect platform for large-scale AI accelerator vertical and horizontal scaling applications. This collaboration will create a scalable AI transmission technology platform, driving future innovation to accelerate the deployment of next-generation computing infrastructure and achieve superior efficiency and growth momentum.
From Copper Wire Transmission to Photonic Interconnects: Embracing Infrastructure Revolution in the AI Era <br /> As modern AI models have grown from billions of parameters to trillions, data transmission has become a major bottleneck in model training and inference. When large language models (LLMs) and other advanced neural network architectures must collaborate across hundreds or thousands of processors, the bandwidth requirements between chips and nodes will grow exponentially. According to Ciena's 2025 global data center network infrastructure market research report, the industry predicts that data center interconnect (DCI) bandwidth requirements will grow sixfold over the next five years (2025-2030), primarily driven by AI applications requiring high bandwidth and low latency.
Current copper-based electrical interconnect technologies, from Active Electrical Copper (AEC) and Peripheral Component Interconnect Express (PCIe) to High Bandwidth Memory (HBM) interfaces, are insufficient to meet future transmission demands. When transmission rates exceed 50 Gbps per channel, signal integrity challenges increase dramatically, requiring reliance on encoding and error correction mechanisms and significantly increasing power consumption. Furthermore, the line density of packaging substrates and circuit boards further limits scalability, leading to high costs and heat dissipation problems. In contrast, photonic interconnects can achieve megabit-per-second (Tbps) performance while significantly reducing power consumption, providing a new and highly scalable solution for the growing AI computing infrastructure.
Ultra-low power photonic interconnects accelerate AI computing performance and data center deployment and applications . Chen Shu-lu, CEO of Opto-Tech, stated, "To simultaneously achieve both vertical and horizontal scaling of AI, heterogeneous integration, including chiplets, CoWoS advanced packaging, and photonic I/O, is indispensable. By combining Opto-Tech's proprietary germanium-silicon photonics technology with Worldchip's expertise in ASIC design, we will jointly create an ultra-low power photonic interconnect platform for AI processors, which is expected to bring breakthrough performance optimizations."
Shih-Lin Shen, CEO of Shih-Lin, stated, "Traditional copper wires have become a bottleneck for the massive data throughput of AI. Our feedback from the supply chain consistently indicates that bandwidth, power consumption, and latency are the biggest obstacles to system performance. Photonic interconnects not only overcome these limitations but also simultaneously ensure efficiency and reliability, making them a key technology for driving AI computing into the next stage."
This strategic collaboration will combine Opto-Tech's technological expertise in photonics with Worldchip's strengths in ASIC and chiplet design to launch modular, ultra-low-power photonic interconnect solutions. Leveraging the comprehensive ecosystem of both companies, encompassing wafer foundry, advanced packaging, and leading-tier customers, this partnership is expected to accelerate the deployment and application of photonic interconnect technology in global AI data centers, ushering in a new era of AI computing performance.
