Samsung Foundry Taking Extreme Steps To Win 3nm Orders From NVIDIA For Next-Gen GPUs

wccftech.com:

Samsung has reportedly taken extreme steps to ensure that it can win orders for its 3nm node technology from NVIDIA.

Samsung Foundry Eyes NVIDIA, Doubling Down on 3nm To Ensure That It's Ready & Meets the Requirements of the Chipmaker

The Korean news outlet, FN News, reports that Samsung has made its top priority for this year to receive chip orders from NVIDIA. NVIDIA has long been using TSMC's advanced process nodes to develop its GPUs, whether they be for the consumer or the data center segment. The most recent families which include Ada Lovelace, Hopper, and Blackwell, are all made using TSMC's 5nm (4N) processes.

Related Story AMD To Detail Next-Gen “Zen 5” Core Architecture At Hot Chips 2024, Intel To Talk “16th Gen” Lunar Lake CPUs

Samsung Foundry is aware of TSMC's leading position and wants to win back NVIDIA as a major customer. For this purpose, they are doing all they can to ensure that their 3nm process node that makes use of GAA (Gate-All-Around) architecture is up to the task and meets the requirements of the green giant.

For this purpose, Samsung has imposed an internal strategy codenamed "Nemo" which refers to NVIDIA and each department is making an effort to ensure that they can receive orders from the chipmaker. It should be remembered that NVIDIA last leveraged Samsung's Foundry (8nm) for its GeForce RTX 30 "Ampere" GPUs which were designed for the gaming (client) segment. The follow-up to Ampere, Ada Lovelace "GeForce RTX 40" switched over to TSMC (5nm).

So far, Samsung Foundry eyes the mass production of the 3nm GAA process in the first half of 2024. The GAA technology will eliminate some of the major bottlenecks associated with the old FinFET processes but it remains to be seen if the process will be enough to win NVIDIA over.

The company was recently reported to have failed the HBM3E memory qualifaction procedure from NVIDIA which could cause a major setback however Samsung is making sure that it can refine its HBM memory and is already