Samsung may not be able to mass-produce HBM chips using hybrid bonding technology until 2029, when it will be matched with Nvidia's next-generation Feynman AI GPU. Hybrid bonding is the next generation HBM manufacturing technology. It removes micro-bumps between DRAM layers, reduces the inter-layer space, and provides better performance. It is considered the future direction of HBM. Compared with traditional micro-bump connections, hybrid bonding can make the chip layers more closely adhered to each other, resulting in faster data transmission and lower power consumption.
Samsung is purchasing 50 hybrid bonding machines from BE Semiconductor in the Netherlands, and plans to start installation at the end of this year and mass production in 2029-2030.
At the same time, Samsung is also developing customized HBM chips, integrating logic chips directly into HBM, and using 3D system-in-package (SiP) architecture to further improve performance.
Nvidia's Feynman AI GPU is expected to launch after Rubin Ultra and will rely on 3D stacking and custom HBM, coinciding with Samsung's hybrid bonding mass production timing.
Samsung's hybrid bonding mass production is later than expected, which may affect its progress in catching up with SK Hynix in the HBM market, but the timing of matching with Nvidia's new GPU in 2029 shows that Samsung is targeting the next-generation market.
Do you think Samsung’s hybrid bonding technology can catch up with SK Hynix’s HBM advantages?
