IBM: 2nm: 333.33
TSMC: 10nm: 52.51, 7nm: 91.20, 5nm: 171.30, 3nm: 292.21
Intel: 10nm: 100.76, 7nm(Intel 7): 100.76, 5nm(Intel 4/3): 200
Samsung: 10nm: 51.82, 7nm: 95.08, 5nm: 126.89
(million transistors per mm2)
So going by Intels former scaling, I'd assume their 20A has 400m transistors per mm2.
TSMC started lying about scale way way way before Intel as you can probably tell by that chart:
Their 3nm has 292m transistors per mm2. IBM sends their research to Samsung, so expect IBM's numbers to end up with Samsung.
However, TSMC does still have a slight tech advantage on performance when comparing their latest N3E iteration to Intels 20A. While Intels 20A node with its PowerVIA reportedly beats TSMCs N3E in wattage. With 18A Intel will beat TSMC again on both accounts. Though TSMC claims they will regain the performance and power leadership against 18A with their N2 node (I'm guessing around 450-480m t/mm2?). While the N3X seems to beat 18A in performance, but not in power.
However, TSMC's N2 node seems to have hit delays, and won't be out before sometime in 2026.
We now know that Intel will be at 1.5nm, or 15A (maybe around 700m t/mm2?) more precisely at their Magdeburg fab set for production in 2027 (not sure if this is scale or not). But Pat did not state if this will be Intel's first 1.5 fab. He only said that it will be their most advanced. That won't rule out a potential for a fab in Arizona or Ohio also producing 1.5nm even sooner. The risk of having a new location being the first to produce a new leading edge is probably too high. With 20A/18A, Arizona will produce it before Ohio goes straight to 18A. Because the Ohio site is fresh. Not to mention that Arizona will produce 20/18A a year before Ohio. I suspect it will be the same story with Magdeburg. Another fab will potentially be at say 16A a year prior. And then Magdeburg will go straight to the 14A version (unless they call it 15A and 13A).
To sum it up, Intel could be at 16A or 15A in 2026, or possibly early 2027. The same year as TSMC moves to N2, or later in the year.
Tsmc reportedly also wants to keep using normal EUV until the end of the decade and their N1 node in 2030. This could potentially put tsmc in the same spot as Intel was in during their 14 nm delays. Multi-patterning is more difficult, and tsmc will need to use that.
The story regarding node delays is nothing more than a story regarding TSMC's machines. And the adaptation of them.
To counter the accusations of Pats so called anti-chinese rethoric, I would say this. Pat only responds to the current global political climate. It has nothing to do with being anti-chinese, but everything to do with wanting production to move back home. The chips act is nothing more than what Taiwan and SK did two decades ago, and what China has done almost a decade ago now as well. It is nothing more than a question of being self sustained with food. You don't want to rely on having all your foods shipped to you from half a world away. That makes no sense to me, and all the asians claiming self sustainment is the same as being anti-chinese, needs to look in the mirror of reality a bit more closely.