News AMD to reportedly become TSMC's next major customer in Arizona — HPC AI chips could start US production in 2025

Admin

Administrator
Staff member
TSMC plans to eventually make N3 and N2 chips in Arizona. But even something like N4C could be very useful. As long as the other parts of the supply chain aren't compromised by war or hurricanes.
 
  • Like
Reactions: bit_user
I expect TSMC's US 4nm fabs will be used for AMD's upcoming 4nm cIOD which will have USB4 2.0 and PCI Express gen 6 alongside AMD's upcoming mid-range GPUs which will replace its Radeon RX7600 series.
 
  • Like
Reactions: bit_user
As long as the other parts of the supply chain aren't compromised by war or hurricanes.
Don't forget earthquakes and labor disputes! California is due for another big earthquake, which would shut down access to their ports for a while.

Another issue Arizona faces is the well-documented lack of abundant water supply, which is critical for semiconductor manufacturing. There were even stories about how a recent drought was forcing Taiwanese fabs to bring in water via trucks (although it's an island and presumably has de-desalinization plants on the coast - wouldn't work so well in AZ).
 
  • Like
Reactions: usertests
I expect TSMC's US 4nm fabs will be used for AMD's upcoming 4nm cIOD which will have USB4 2.0 and PCI Express gen 6 alongside AMD's upcoming mid-range GPUs which will replace its Radeon RX7600 series.
Zen 5 uses the N4P node. So, AMD could make the compute dies there which are used in Ryzen 9000 desktops and the Turin Zen 5 server CPUs. I had even wondered whether this might've factored into their reasoning for using N4P, instead of N3E. Also, does the Strix Point laptop SoC also use N4P?
 
Fab 21's Phase 1 is limited to the N4 and N5 technologies, barring the possibility of any consumer chips newer than RDNA 3 and Zen 4.

Zen 5 CCDs are manufactured on N4P, although Zen 5c for EPYC will be manufactured on N3B or N3E, so that will not be compatible.

Some of AMD's current APUs/SoCs are also on N4P, such as Strix Point. Strix Halo will use Zen 5 CCDs with new packaging and a custom IOD. Halo's IOD should be manufactured on N3E to meet lower power targets for that large iGPU, but N4P is also a possibility.

RDNA4 should use N3E or N3P, but depends on availability and cost. Navi 44 will likely be on N4P to keep costs under control. Navi 48 is the larger RDNA4 GPU.
 
  • Like
Reactions: bit_user