TSMC plans to introduce next-gen CoW-SoW (System-on-Wafer) advanced packaging technology by 2027, enabling the development of gigantic chips.
TSMC Plans To Lead The Future of Packaging Technology Through CoW-SoW, Allowing Development Of Huge AI & Data Center Chips
With the demands of the AI industry rapidly evolving, the need for innovation in the supply chain is now more desperate than ever, which is why companies like TSMC are in pursuit of integrating new technologies into existing products. Ctee is now reporting that the Taiwan giant plans to develop a new "CoW-SoW" advanced packaging technology, which is said to stack memory and logic chips on a single interface, providing faster performance and increasing precision in bridging across the onboard dies.
There's a bit of a backstory behind this report, particularly highlighting a reason why NVIDIA's Blackwell architecture witnessed a delay in the first place. It is alleged that NVIDIA's Blackwell is the largest AI chip in the market, and due to the hefty components onboard, it has created manufacturing complexities, with the interconnect technique creating huge issues for suppliers like TSMC.
With disparities present between the GPU die and the main substrate, NVIDIA decided to revisit the taping out of Blackwell, which is why the associated AI products saw a recent design change but will still be heading for volume production in Q4. Apart from this, it is revealed that NVIDIA's upcoming GeForce RTX 50 series consumer GPUs will also be affected by the same problem, which led Team Green to redesign the top metal layer and bumps.
TSMC has already given us a slight glimpse of CoW-SoW packaging, claiming that it will have 40 times the current reticle limit and allow the integration of 60 times the HBM capacity through its massive substrate. The Taiwan giant revealed that SoW is explicitly targeted for future data center clusters and verified that the packaging technique is expected to enter mass production by 2027.
Interestingly, TSMC's CoW-SoW is employed by Cerebras, which is said to be the largest wafer-scale chip for the AI markets; hence, the Taiwan giant has experience with gigantic AI chips. Integration of CoW-SoW isn't expected to be done by 2027, which means that NVIDIA can utilize the technology in its next-gen Rubin architecture.
It will be interesting to see what sort of performance NVIDIA's Blackwell brings to the table, given that the firm's initial claims look quite impressive. The upcoming AI architecture will likely prove to be NVIDIA's "most successful product" in its history.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
