The news about Samsung snapping an advanced packaging order for Nvidia’s AI chips paired with high-bandwidth memory (HBM) chips underscores the strategic importance of next-generation packaging solutions. According to a report published in South Korean media outlet The Elec, Samsung’s advanced packaging team will provide interposer and 2.5D packaging technology for Nvidia’s AI processors.
It’s important to note that the GPU and HBM building blocks in these AI processors are supplied by other companies—most likely Nvidia’s GPUs manufactured on a TSMC process node and HBM chips designed and produced by Samsung’s archrival SK hynix.
What’s more important is how industry watchers relate this development to the insufficient capacity of TSMC’s chip-on-wafer-on-substrate (CoWoS) technology, which stacks chips and packages them onto a substrate. However, this supply shortage connected with the recent earthquake in Taiwan doesn’t hold much weight, and it’s most likely related to supply and demand issues.
Samsung calls its 2.5D packaging technology iCube; it places one or more logic dies such as CPUs and GPUs and several HBM dies on top of a silicon interposer, making multiple dies operate as a single chip in one package. It deploys parallel and horizontal chip placement to boost performance and combat heat buildup.
Figure 1 The iCube technology offers warpage control even with large interposers, and its ultra-low signal loss is paired with high memory density. Source: Samsung
Samsung’s advanced packaging pivot
Trade media has been abuzz with reports about Samsung beefing up its advanced packaging division by hiring more engineers and developing its own interposer technology. The company reportedly procured a large amount of 2.5D packaging equipment from Japanese semiconductor equipment supplier Shinkawa.
Another report published in The Elec claims that Applied Materials and Besi Semiconductor are installing hybrid bonding equipment at Samsung’s Cheonan Campus. Hybrid bonding enhances I/O and wiring lengths compared to existing bonding methods. TSMC offers hybrid bonding in its 3D packaging services called System on Integrated Chip (SoIC). Intel has also implemented hybrid bonding technology in its 3D packaging technology called Foveros Direct.
Media reports suggest that Samsung has recently ramped up the production capacity at its key site for advanced production in Cheonan to full utilization preceding Nvidia’s advanced packaging orders. Industry observers also expect that this advanced packaging deal with Nvidia could pave the way for Samsung to win the supply of HBM chips for pairing with GPU’s maker’s AI devices.
SK hynix is currently the major supplier of HBM chips for Nvidia’s AI processors, and Samsung is frantically working to close the gap. In fact, when Samsung established the advanced packaging business team in December 2023, the company’s co-CEO Kye-Hyun Kyung hinted about seeing the results of this investment in the second half of 2024.
Advanced packaging in Samsung’s roadmap
Kyung also pinned his hopes on a competitive advantage with Samsung’s memory chips, chip fabrication, and chip design businesses under one roof. Advanced packaging stands out in this semiconductor technology portfolio due to its intrinsic link to large and powerful AI chips and system-in-package (SiP) devices.
Figure 2 Next-generation packaging technologies are in the limelight due to the massive demand for AI chips. Source: Samsung
Like TSMC and Intel Foundry, Samsung is aggressively investing in advanced packaging technologies like silicon interposers while also steadily expanding its production capacity. Interesting times are ahead for next-generation packaging solutions.
Related Content
- Is 3D IC The Next Big Profit Driver?
- Heterogeneous Integration and the Evolution of IC Packaging
- Understanding the Big Spend on Advanced Packaging Facilities
- Tighter integration between process technologies and packaging
- Advanced IC Packaging: Fundamentals for the ‘More than Moore’ Era
The post Samsung’s advanced packaging pivot with Nvidia production win appeared first on EDN.