Samsung Electronics is reportedly set to use Taiwan Semiconductor Manufacturing Company’s (TSMC) 3-nanometre manufacturing process to produce base dies for Nvidia’s next-generation custom high-bandwidth memory (HBM) chips. The reported arrangement highlights how the race to build more powerful AI processors is reshaping the semiconductor supply chain.
According to South Korea’s Seoul Economic Daily, Nvidia has asked Samsung, SK Hynix and Micron Technology to standardise manufacturing requirements for its custom memory technology. Reuters reported the development on October 18, citing the South Korean newspaper.
Why Nvidia wants standardised memory
Nvidia’s custom memory technology, known as NVHBM, integrates a customised memory controller into the HBM base die. This design aims to standardise how the memory is implemented across different suppliers.
High-bandwidth memory is essential to modern AI accelerators because it helps move large volumes of data between memory and processors. Improving memory performance and supply is increasingly important as AI workloads expand.
According to the report, Nvidia wants suppliers to use TSMC’s 3nm process for the base dies and sixth-generation DRAM technology for the core dies. Standardising these components could make it easier to maintain consistent performance across memory supplied by different manufacturers.
A notable development for Samsung and TSMC
The reported arrangement would bring the two semiconductor companies together at the HBM manufacturing stage, despite their competition in chip fabrication.
Samsung has traditionally produced both the memory core and base die for its HBM products using its own manufacturing capabilities. The reported plan would allow it to use TSMC’s advanced process for the base dies in Nvidia’s custom memory programme.
However, Samsung is reportedly keeping open the possibility of manufacturing these components in its own foundries if demand exceeds TSMC’s available capacity. The arrangement should therefore not be interpreted as Samsung abandoning its in-house HBM manufacturing strategy.
What this means for the AI chip industry
The development underscores the growing importance of memory technology in the AI hardware market. AI performance depends not only on powerful processors but also on how quickly data can move between the processor and memory.
For Nvidia, standardising the manufacturing approach across memory suppliers could help simplify product integration and support a more consistent supply chain. For Samsung and other memory manufacturers, meeting Nvidia’s technical requirements could influence how they develop their next-generation HBM products.
The reported Samsung–TSMC arrangement remains subject to confirmation from the companies involved. Nevertheless, it highlights the increasingly interconnected roles of chip designers, memory manufacturers and contract foundries in the global AI industry.
