I've been getting some inclinations that Zen3 has been pushed to 5nm. Anyone on 5nm this year(2020) is actually paying less for final wafers than on 7nm/7nm+.
There however appears to be a trick maybe? TSMC might have slipped in a 5nm higher-perf/low-density/high-yield RTO lib for RTO'ing 7nm+. [Access to HM-Fins/BEOL-mitigations for higher Fmax/Fsustained]
7nm+ EUV fab doesn't have access to the pellicle replacement: particle blower goes ptooey! and no pellice is needed for Mobile or HPC.
Much how Excavator was done on two nodes 28nm/20nm. So, was the 19h Zen3 on 7nm/5nm. Development of both implementations occurred at the same time. With the 5nm parts getting better yields sooner than later.
So keep mobile Zen3 on 7nm or move to 7nm+? Honestly, I'd normally say this was a super crazy theory but I've also been pondering this. We know that when Huawei got the ban, there was a huge allotment of wafers that AMD and Apple snatched up, then came the others. If you also consider in the number of consoles Microsoft and Sony expect to ship in N time period, then you're left wondering how AMD wouldn't face shortages once again even if they bought all the allotment. There was some rumor a few weeks ago about TSMC developing a special node but it was mis-attributed to being for AMD only when I doubt that's the case. If it was, then it says something about their relationship. PC Gamer reported it, but as usual Reddit and HOCP tore it apart. The general consensus was that TSMC designed a variant of the 5nm node for AMD first, but to be used by whatever partner wants it and can pay for it. Again, I'm not 100% sure on the claims' veracity but it's very interesting. You're maybe the fifth person I know who's said something like this and it's something I'd been keeping to myself because it didn't make much sense initially, and I'm pretty sure I was several beers in during a heatwave here a while back when it popped into my head.
And potentially for its RDNA 3 graphics cards too.
www.pcgamer.com
The idea of a 5/5 launch is out the window, and moving to 5nm now would allow AMD to get rid of some growing pains than if they went on a new node for AM5 for Zen4. I think AMD may go with 5nm+ for Zen 4 or 5, and then slowly move to 3nm? One thing I hadn't read until just now was I was under the impression that Apple was only moving their SoC for their mobile devices to 5nm, but it seems like the press is saying they'd be going with TSMC's 5nm node. And since Apple has a relatively quick ramp to full ARM, it makes a lot of sense to me, Nosta. Apple would effectively pay off 5nm in the long run and make it cheaper for AMD and others to use. TSMC did state back in December their yields for 5nm test runs were >80% which was impressive. It makes a lot of sense. Another reason for AMD to be so silent for so long. If AMD do pull it off, they'll be 2 "generations" ahead of Intel.
Apple going to Apple Silicon is probably the best thing for AMD. It'll put pressure on TSMC to ensure future nodes are flawless and provide decent ROI for architects to work off of. I wonder if AMD will save more money/increase margins per monolithic and per chiplet now if all this is even remotely correct.
There was that pesky rumor of AMD focusing on shorter times between releases, around a 12 month mark unlike their current 12-17 month one.
Edit: It makes sense when you consider that AMD prefers to keep mobile on monolithic for performance reasons, and Zen3 mobile is supposed to be "drastically different" that it may very well kick back TigerLake into the neolithic era of computing.