Samsung’s hybrid bonding technology is expected to be used in NVIDIA’s next-generation Feynman AI accelerator, with the chip featuring custom HBM.

Samsung Electronics has commenced construction of a Die-to-Wafer (D2W) hybrid bonding production line at its P5 plant in Pyeongtaek, targeting advanced packaging for next-generation High Bandwidth Memory (HBM) and logic semiconductors. Samsung’s top pick for core equipment suppliers is Dutch firm Besi; the company plans to procure around 50 hybrid bonding machines, each priced at roughly 6 billion won (about $4.3 million) — twice the cost of competing products. However, negotiations have moved slowly due to Samsung’s demand for custom modifications to the machine architecture, and Besi, which also supplies TSMC and Micron, is cautious about modifying equipment exclusively for Samsung. Samsung is simultaneously evaluating local alternatives: its subsidiary SEMES has completed development of D2W hybrid bonding machines and delivered samples for validation earlier this year. Hanwha Semitech finished developing its second-generation "SHB2 Nano" in February, and sent samples to SK Hynix in April. Its cluster system integrating Equipment Front End Module (EFEM), plasma activation, and cleaning modules offers distinct competitive advantages. This technology replaces traditional thermocompression bonding with hybrid copper bonding, enabling direct bonding of copper and dielectric materials and drastically shortening interconnect lengths. Its core application is custom HBM: vertically stacking HBM DRAM layers directly on logic dies containing customers’ computing circuits and intellectual property (IP) to form 3D system-in-package (SiP) solutions. Citrini analyst Jukan stated that hybrid bonding is expected to be used in NVIDIA’s next-generation AI accelerator Feynman, which will adopt custom HBM, with the technology’s deployment timeline "pushed back from HBM4E". Samsung’s internal projections indicate large-scale commercialization will not be achieved until around 2029–2030, a timeline that aligns with NVIDIA’s expected window for Feynman.