https://www.anandtech.com/show/17385/applied-materials-new-t…
A word to the wise, if you are interested in the topic definitely don’t skip the comments–they seem to have more information than the article.
For those who want the “drive-by” story, Samsung’s announcement indicates that 3 nm will show up first in low-power and DRAM parts. AMD almost certainly won’t use TSMC’s N3 process for Zen 5. They may use the same process as for Zen 4 or a “4 nm” process which is derived from N5. Why stay off the bleeding edge? That it is the bleeding edge might be a good reason. But Zen 4 is starting (volume) production, and AMD won’t want to wait while 3 nm design rules are determined.
Intel on the other hand is trying to push ahead at one generation per process node. I think they will be (or are) satisfied with the migration to Intel 7 (Raptor Lake, Granite Rapids) but they have pushed Intel 3 out to 1H2024, and Intel 4 is apparently out of the server picture entirely. Intel has announced an additional server generation Emerald Rapids in 2023, but it will be on Intel 7. That’s almost a detail. Intel 4 will be used for Meteor Lake in 2023. The only remaining product on Intel 20A is Arrow Lake currently scheduled for 2024. Intel has weasel-worded Arrow Lake–or they may just be trying to highlight finally moving to a tiled architecture. In any case, some chiplets in Arrow Lake are expected to be manufactured using Intel 20A. Could they be thinking like AMD that the I/O parts of the chip might as well be manufactured on an earlier process.