密集波分复用技术(DWDM)
Search documents
这种芯片,被数据中心看好
半导体芯闻· 2026-03-17 10:45
Core Viewpoint - The article discusses the launch of the world's first single-chip DWDM optical engine for AI infrastructure by Tower Semiconductor and Scintil Photonics, addressing the critical need for lasers in the co-packaged optics (CPO) ecosystem to meet the increasing bandwidth and power consumption demands of AI data centers [1][2]. Group 1: DWDM Technology and Its Importance - Dense Wavelength Division Multiplexing (DWDM) technology allows multiple optical signals to be transmitted over a single fiber, significantly reducing power consumption and latency when connecting multiple GPUs [1]. - The deployment of DWDM in AI data centers has been limited due to cost and scalability issues, which are now being addressed with new technological advancements [1][2]. Group 2: Integration and Manufacturing Process - Scintil's heterogeneous integrated photonic technology (SHIP) integrates lasers, photodiodes, and modulators onto silicon wafers for mass production, akin to CMOS processes in the photonics field [3]. - The manufacturing process involves using standard 300mm silicon photonic wafers, integrating passive optical components, and employing advanced lithography for precise control of spacing and wavelength stability [3]. Group 3: Next-Generation CPO Development - The next-generation CPO aims to facilitate vertical scaling by integrating lasers directly into the CPO chip, enhancing efficiency and performance in high-density GPU environments [5]. - The LEAF Light chip design allows for single-fiber transmission rates of 1.6Tbps by distributing 50Gbps across multiple channels, significantly improving overall capacity and energy efficiency [5]. Group 4: Future Plans and Market Readiness - Scintil and Tower plan to deliver tens of thousands of devices to customers by the end of 2026 and aim to increase production capacity by an order of magnitude next year [6]. - By 2028, the supply chain will be fully prepared for the deployment of DWDM in vertical scaling networks, opening new possibilities for the industry [6].