Samsung Electronics will use rival TSMC’s manufacturing process for Nvidia’s custom memory chips, leaving its foundry division with an awkward explanation.
According to Seoul Economic Daily, Nvidia has asked Samsung, SK hynix and Micron to use TSMC’s 3nm process for the base dies in its NVHBM high-bandwidth memory.
Samsung accepted the requirements after an internal review. The arrangement would mark the first cooperation between Samsung and TSMC at the HBM manufacturing stage. Samsung has previously produced both the memory dies and the underlying logic die itself.
The companies worked together four years ago to ensure Samsung’s HBM was compatible with TSMC’s GPU packaging process. This agreement would take that relationship further into chip production.
NVHBM, unveiled in August, incorporates Nvidia’s custom memory controller into the base die. Nvidia wants consistent implementation across its memory suppliers, reducing differences between their products and the work required to validate each version. Its requirements extend to the memory dies, which would use sixth-generation DRAM technology known as 1c.
Samsung intends to retain its in-house manufacturing approach for standard HBM and custom products supplied to other customers, including AMD. Its HBM4 uses Samsung’s own 4nm process for the base die. HBM4E, undergoing mass-production validation, follows the same approach, with Samsung manufacturing the memory dies at its Hwaseong and Pyeongtaek facilities.
Samsung is assessing whether its foundry could manufacture NVHBM base dies if demand exceeds TSMC’s capacity, the newspaper reported, citing an unnamed industry deep throat.
