Three giants of the sector are aware of what TSMC does. Nothing less than AMD, NVIDIA and Apple are with the bow drawn waiting to launch the arrow on the production of the Taiwanese titan and curiously this may be one of the problems that the company has found to meet. It was said at the time that the 3nm or commercially speaking its N3 would be in mass production for the 2H of 2023, but…
TSMC’s 3nm is causing too many problems
Things don’t seem to be going well for TSMC 3nm, I think Zen5 and RDNA4 are likely to switch to 4nm🤔
— Greymon55 (@greymon55) February 21, 2022
One of the key points of this node is that it will be the last to integrate FinFET transistors. Intel and Samsung are going to implement GAA in their respective high design nodes so as not to deal with the conductivity and control issues of shrinking a FinFET transistor to the bare minimum: too expensive, too complex.
TSMC accepted the challenge not to be left behind and continue to lead in this generation of chips, but it can pay dearly. In fact, there are already rumors that Zen 5 and RDNA 4 they could end up in a higher and much less dense node like N4.
The problem reaches such a depth that there will not be the typical LP or HP variants, or simple planned node improvements such as the current N7+, also called N7P. Not here, TSMC has placed a variant N3B Besides of N3E which he announced as a base. Why do this now when it’s your last node before making the jump to GAA? Well, it seems to be that it has to do with the needs of the clients, or it could also be that they simply cannot meet the demand and the performance that was set in the first place.
AMD, NVIDIA and Apple looking to Intel?
> TSMC’s N3 process still not looking good. Now there’s a “N3B” variant in addition to previously announced N3E. And multiple tweaks to suit different customers’ needs…
— RetiredEngineer® (@chiakokhua) February 21, 2022
Curiously, Intel had its pluses and minuses with a node that today is a reality such as 10nm (Intel 7 now), where the blue giant lost a lot of development time due to multiple problems. It seems that the situation will repeat itself to a lesser extent with TSMC, just when Pat Gelsinger’s have stepped on the accelerator.
Therefore, there are several theories about what the three companies that face the blues will do. The first is to stick with TSMC’s 5nm or 5nm+, the second is to port the masks and focus the designs on N4, N4P or N4X depending on the type of product, while the third is focused on taking a step towards Samsung, where the Koreans have gone ahead of everyone and announced GAA at 3nm for the next quarter.
The problem is that curiously Intel will have the densest node again this year, since its Intel 4 (old 7nm) arrive, they will achieve it with FinFET and also in a higher volume than the rest of the pursuers, which is crucial. This added to your strategy IDM 2.0 with your services IFS totally open for any company has unleashed rumors that some competitor could make use of its node to make its chips, specifically it could be AMD, since it cannot afford the scalability of TSMC’s N4 while Intel would have by then Intel 3 (old 4nm) next year.
Will it be the paradox that AMD, NVIDIA or Apple end up buying chips from Intel? TSMC is struggling, Samsung isn’t loud enough, and Intel is hitting the gas at a hellish pace.