Korean chip giant Samsung is likely to start mass producing chips with hybrid bonding by the end of the decade, according to Korean industry sources. Hybrid bonding, which relies on removing the micro bumps in individual DRAM layers when manufacturing high bandwidth memory (HBM), is widely believed to be the next-generation HBM manufacturing technology due to its ability to reduce the space between the layers and offer superior performance.
Samsung Likely To Introduce Hybrid Bonding With NVIDIA's Feynman AI Chips, Believe Sources
NVIDIA's Feynman AI GPUs are slated to follow the Rubin Ultra chips, and NVIDIA shared quite a bit of details about them earlier this year at the GTC Conference. Among these, the firm revealed that the GPUs would rely on 3D stacking, feature Intel as a packaging partner and rely on custom high bandwidth memory (HBM). With Feynman's predecessor, the Rubin Ultra chips, expected to rely on HBM4E memory chips, the latest chips could either use HBM5 or a custom variant that is differentiated from competitors' offerings.
In line with these specifications, a recent report also claimed that hybrid bonding might start to be used with HBM4E memory chips instead of earlier generations. Now, a report from the Korean press makes similar claims and outlines that Samsung believes that large-scale production with hybrid bonding would be feasible in 2029 and 2030.
According to the details, Samsung is in the process of acquiring 50 hybrid bonding machines from the Dutch firm BE Semiconductor. As the firm wants medications for the machines and BE sells off-the-shelf equipment, sources suggest that Samsung is also interested in procuring the machinery from local companies in Korea.
More interestingly, the report also discusses custom HBM chips. It outlines that custom HBM replaces the base die on the chip with a customer's logic die as part of a 3D system-in-package (SiP) architecture.
As for the hybrid bonding machines, the sources believe that Samsung will begin installing them at its Pyeongtaek facilities by the end of this year. Full-scale and large-scale production is expected to commence in 2029 and 2030, according to the sources, or around the time NVIDIA's Feynman chips will enter the fray.
About the author: Ramish is a seasoned technology writer and editor with more than a decade of experience. He specializes in semiconductor fabrication and market analysis. With a background in finance and supply chain management - via his bachelors in Finance and a micromasters in supply chain management from MIT - Ramish combines financial rigor with deep industry insight to deliver accurate and authoritative coverage.
Follow Wccftech on Google to get more of our news coverage in your feeds.
