TSMC teams up with SK Hynix to accelerate development of HBM4 memory. The goal: to bring HBM4 to market faster than Samsung.

The South Korean news media Pulse News writes about the alliance between TSMC and SK Hynix. One can imagine a merger of the Taiwanese and South Koreans as a “supergroup” within the chip industry. One is the largest semiconductor producer, the other is the world market leader in memory chips. Both companies want to work together on HBM4 memory for AI chips under the appropriate project name “One Team”.

The collaboration appears to be mainly a move against Samsung. Samsung and SK Hynix are each other’s biggest competitors in the memory market. After a dramatic year for Samsung’s chip division, the battle has swung completely in SK Hynix’s favor over the past year. By collaborating with TSMC, SK Hynix aims to further strengthen its leadership position and be the first to commercialize HBM4, catching up with Samsung while the company is on the ground.

HBM4 will not immediately become the standard for AI chips. These are currently HBM3 and HBM3e. We first saw the faster version of HBM3 in the Nvidia H200, the “super chip” from the GPU specialist. Thanks to the smaller 10-nanometer process, this new version is around 50 percent faster than HBM3. Overall, HBM3e delivers a combined bandwidth of 10 TB/s.

Nvidia first?

With HBM4, even higher performance can be achieved on an AI chip. Nvidia is expected to once again be the first to incorporate the new memory into its GPUs. The upcoming Blackwell generation will likely still contain HBM3(e), but the following generation (named after American astronomer Vera Rubin, according to Wccftech) would switch to HBM4. SK Hynix and TSMC already want to make sure they’re at the front of the queue by then.