South Korea's SK Hynix, in collaboration with institutions including the University of Virginia, published a paper titled 'Co-Packaged Optics (CPO) for High-Performance Computing and Artificial Intelligence' in the leading journal Nature Electronics on Thursday (20th). The paper presents a 'light-centric' architecture, proposing direct connections between memory and processors through photonic interposers.
SK Hynix stated that CPO is key to scaling system-level HBM innovation. As AI systems continue to expand in scale, the importance of optical interconnects will grow increasingly significant. The company plans to expand open collaboration with customers and ecosystem partners to accelerate innovation in AI infrastructure.
According to Zhongzheng Overseas Information, on the same day, Samsung Electronics announced an expansion of its core project 'P5 1&2 Fabs (Fab 5&6 Phases)' at its Pyeongtaek semiconductor production site in Gyeonggi Province—from the originally planned two-story wafer fabrication plant to a three-story facility.
Meanwhile, SK Hynix also intends to adopt a fully three-story structure for its new wafer fabrication plants being built within the Yongin semiconductor cluster. South Korea's memory giants are rapidly upgrading from two-story to three-story production lines.
These moves reflect a strategic capacity grab ahead of the 'super cycle' in AI semiconductors and memory. Market research firm Counterpoint Research predicts the global memory market will surge from approximately 360 trillion KRW last year to 1,500 trillion KRW this year, further expanding to 2,100 trillion KRW next year.
Driven by the boom in AI data center construction, demand for HBM and high-capacity memory has exploded, compelling companies to urgently expand production capacity.
From breaking through the 'memory wall' bottleneck via photonic interconnects to surpassing physical production limits through vertical factory stacking, South Korea's two memory leaders are advancing on both 'technology and capacity' fronts to meet the explosive demands of the AI era.
If CPO transitions successfully from research papers to mass production, it will rewrite the paradigm of data transmission between HBM and processors. If the three-story wafer fabs are completed as scheduled, they will stockpile ample supply for the next memory super cycle. In the race for AI computing power, storage is no longer a supporting role but a strategic linchpin determining the ceiling of system performance.
FACT BOX
- Source: PR Times
- Category: Partnership
- Organizations: Counterpoint Research
- Products / services: HBM