Samsung Electronics will use TSMC's 3-nanometre process to produce the base dies for Nvidia's next-generation custom high-bandwidth memory, the Seoul Economic Daily reported on Sunday. The move would mark the first time Samsung has worked with the Taiwanese contract chipmaker at the HBM manufacturing stage, according to the report, which Reuters summarized on October 18.
An HBM stack combines two kinds of silicon. The core dies hold the memory itself, while the base die underneath acts as a system semiconductor, handling logic and communication with the processor it sits beside. Until now Samsung has made both parts itself for all of its HBM supply, a turnkey approach that let it control the whole stack. For Nvidia's custom product, that approach is ending.
Nvidia has asked all three of its suppliers for the product — Samsung, SK hynix and Micron — to manufacture the base die on TSMC's 3-nanometre process and to standardize the core die on sixth-generation DRAM technology, known as 1c, industry sources told the newspaper. The product, called NVHBM, was unveiled in August and is based on the architecture of HBM4E, the seventh-generation HBM. NVHBM integrates Nvidia's custom memory controller into the base die and is aimed at standardizing implementation across multiple memory suppliers.
The reason is consistency. When each supplier builds the base die differently, Nvidia cannot easily deliver the same performance from a single design, and it must validate each supplier's product separately. A common base-die process and a common core-die process reduce that burden and make the three sources more interchangeable — an important consideration when HBM supply remains a bottleneck for AI accelerators.
Samsung's Device Solutions division, which runs its foundry business, decided after an internal review to accept Nvidia's request, the report said. That is a pragmatic concession, not a wholesale change of strategy. Samsung plans to retain its in-house production model for standard HBM and for custom products for other customers, including AMD.
For standard products, Samsung applies its own 4-nanometre foundry process to the base die of HBM4, which it began mass-producing earlier this year, and of HBM4E, now undergoing mass-production validation. The core dies for both are made on 1c lines at its Hwaseong and Pyeongtaek campuses in South Korea.
Samsung is also keeping an option open. It is assessing whether its own foundry could produce NVHBM base dies if demand exceeds TSMC's capacity, an unnamed source told the newspaper — a hedge, in the Seoul Economic Daily's account, rather than a plan to displace TSMC for this product.
The two companies have cooperated before, but more narrowly. Four years ago Samsung worked with TSMC to ensure compatibility with TSMC's packaging process when connecting HBM to graphics processing units. Partnering from the manufacturing stage goes further, and it underlines how far Nvidia will go to standardize the memory that feeds its AI systems — and how central TSMC's leading-edge process has become, even for a rival that operates its own foundries.