Seoul Economic Daily reports on October 18, 2026, that Samsung Electronics accepted NVIDIA’s request to use TSMC’s 3nm process for the base die in NVIDIA’s custom NVHBM memory. The report names Samsung, SK hynix and Micron Technology as suppliers asked to use a common base-die process; it describes Samsung’s acceptance and the request to the other two suppliers.Seoul Economic Daily
The reported process applies to NVHBM base dies. The same account says the core die—the memory dies in the stack—would use sixth-generation DRAM, known as 1c.
What NVIDIA says NVHBM changes
NVIDIA’s description of NVHBM places its custom memory controller in the HBM base die rather than on the XPU die. NVIDIA says the design, compared with standard HBM4E, offers up to 30% greater memory bandwidth, 15% lower HBM power consumption and up to 25% more area on the XPU compute die. Those figures are NVIDIA’s stated comparisons.
NVIDIA also says a standard NVHBM implementation across multiple memory suppliers can reduce the engineering work needed to integrate and qualify the memory. That shared design goal is separate from the reported request to use TSMC’s process for NVHBM base dies.
How the reported arrangement fits Samsung’s HBM work
The report describes a limited role for TSMC’s process: the base die for NVIDIA’s custom NVHBM. It says Samsung plans to retain its turnkey approach for standard HBM and custom HBM for other customers, including AMD. For standard Samsung HBM4 and HBM4E, the account describes Samsung Foundry 4nm base dies and 1c core dies. It says HBM4 was in mass production and HBM4E was undergoing mass-production validation when the report appeared.
Seoul Economic Daily also reports that Samsung is evaluating its own foundry as a possible source for NVHBM base dies if TSMC’s capacity proves insufficient.
