Mannerisms

As they say, talk is cheap. IBM hasn't even been able to get 45nm into production yet while Intel has been shipping it for a year. I say Otellini is right, they're 1 or 2 generations behind in the real world.

And TSMC?

TSMC is a Foundry. They're a one size fits all solution for many customers. They can't match Intel's integrated design/process approach where the process and the design are developed around each other. TSMC has too many customers to give that kind of service too. Same goes for IBM. Neither has HiK/MG working yet either. IBM promised it for 45nm but had to push it out a generation when they couldn't get it to work. IBM is delivering promises, Intel is shipping silicon. Too bad the author couldn't see the difference.

Don't you think TSMC's is a cleverer trick because it has to make its process usable by many customers whereas Intel has only one customer? Moreover TSMC has several, usually three flavours (high performance, low power and general purporse), to each node ,whereas Intel only has one flavour - high performance.

TSMC does a great job for the market they serve and I'm not criticizing them. But back to the article, the author seemed more impressed by far off promises from silicon vendors than real world capabilities. IBM/et al is clearly well behind Intel today. IBM is still unable to ship 45nm silicon in production, they've pushed out HiK/MG at least one process generation and they have a history of suffering yield problems, yet the author is easily seduced by the next press release. That was my point.

Qualcomm announced production volume availability of 45nm chips for cellphones - a larger mass market than PCs - in November 2007. Made by TSMC.

I don't think November 2007 is a fair date to evaluate TSMC's performance relative to Intel. Qualcomm certainly didn't start any form of mass production then even if I guess they theoretically might have in low volumes.

What might be a more reasonable metric of success without insider data is "When do yields start to really matter?" - in Intel's case, this clearly was 4Q07. In TSMC's case, I think it's probably 3Q08 or 4Q08. For GPUs on 40nm, it will be 1Q09... (note that I'm thinking of when the completed wafers come back, not when production starts).

In the end, Intel is still willing to pay a premium to get leading-edge tools before anyone else because their business depends more on raw performance than performance/dollar. Practically none of TSMC's customers are in that situation, so I don't think this is going to change anytime soon.

However I remember Otellini saying a year or two ago that he would even consider selling the previous node in a foundry model. It didn't sound like a great idea back then, but now that TSMC seems like they'll always be either a half-node ahead or a half-node behind, it seems especially ridiculous (can you imagine starting 65nm production at Intel in 1Q09?) - which does indicate they probably underestimated the foundries' apparent capability to reduce the gap further every generation.

That's very interesting. Thanks.
It's a nightmare trying to find out the truth about the state of maturity of processes, but it's interesting that, after all these years, process prowess is still a one-upmanship sport though now, as you point out, one with only two players!