NVIDIA’s Development Of Custom HBM Drops Samsung’s And TSMC’s Foundry Rivalry, As Both Manufacturers Join Forces To Develop “Base Die” On The 3nm Node

Samsung has typically adopted a “turnkey” approach when developing HBM’s core and base dies, but according to the latest report, NVIDIA’s custom NVHBM requires a more streamlined approach in manufacturing involving Samsung and TSMC. Both firms have seemingly joined forces to develop the base die on the 3nm node, with the end goal being a standardized product requiring fewer quality verification steps.

Micron and SK hynix have also joined Samsung to achieve the same purpose, as NVIDIA aims to obtain consistent performance by adopting one manufacturing process

As Feyman GPUs have been mentioned to be treated to custom HBM, NVIDIA has also promised that the in-house solution can deliver up to 30 percent more memory bandwidth than standard HBM4E while utilizing up to 15 percent lower power draw. However, to achieve these advantages, NVIDIA must ensure standardization across memory manufacturers, which is why the GPU maker has reportedly asked Samsung, Micron, and SK hynix to manufacture the NVHBM’s base die on TSMC’s 3nm node.

Additionally, the core die needs to rely on the sixth-generation DRAM process, also known as 1c. The motivation behind bringing all the massive memory giants toward utilizing a single manufacturing process is consistent performance from the finished product. With only TSMC’s 3nm process being used for the base die, NVIDIA won’t have to verify each vendor’s sample separately, saving valuable time…