Samsung Electronics has begun building a Die-to-Wafer, or D2W, hybrid bonding mass-production line at its P5 plant in Pyeongtaek, targeting advanced packaging for next-generation HBM and logic semiconductors.
For equipment, Samsung is said to favor Dutch supplier Besi as its main vendor. The company plans to purchase about 50 hybrid bonding machines. Each machine is priced at about KRW 6 billion, or roughly $4.3 million, which the report said is double the price of competing products.
Negotiations with Besi are moving slowly
Talks between Samsung and Besi have been slow. The report said Samsung has asked for customized changes to the equipment architecture, while Besi also supplies TSMC and Micron and has been cautious about modifying its tools exclusively for Samsung.
Samsung is also reviewing domestic alternatives
Alongside Besi, Samsung is considering South Korean alternatives.
- Its subsidiary SEMES has completed development of a D2W hybrid bonding machine and sent samples for validation earlier this year.
- Hanwha Semitech completed development of its second-generation "SHB2 Nano" in February and sent samples to SK hynix in April.
Hanwha Semitech’s system integrates EFEM, plasma activation, and cleaning modules into a clustered platform, which the report described as a point of differentiation.
Hybrid bonding is aimed at custom HBM packaging
The technology replaces conventional thermocompression bonding with hybrid copper bonding. It can directly bond copper and dielectric materials and sharply reduce interconnect length. Its main use case is custom HBM, where HBM DRAM layers are vertically stacked directly on a logic die containing customer compute circuitry and IP to form a 3D system-level package.
Jukan says Nvidia Feynman is expected to use it
Citrini analyst Jukan said hybrid bonding is expected to be used in Nvidia’s next-generation AI accelerator, Feynman, and that the chip will use custom HBM. Jukan also said the deployment timeline for the technology "has been pushed back from HBM4E."
Samsung internally expects large-scale adoption around 2029 to 2030, a timeframe that overlaps with the anticipated launch window for Nvidia Feynman.