It's fairly amazing to think that, if you want a fully functional 28nm chip, you'll be able to get one by the end of this year. TSMC is promising to have a 28nm CyberShuttle prototyping service available by the end of the year.
What's more TSMC says that 28nm will be a full node process rather than a half node process which means that it will be fully supported by IP. "We will field a full IP portfolio", TSMC's K.T.Sung told me, "and the process includes an RF option."
Sung said that TSMC had customers for its 28nm CyberShuttle, but wouldn't say which process flavour they wanted.
The process comes in a low power flavour and a high performance flavour. The low power process uses silicon oxynitride, and the high performance uses high-k metal gate.
The low-power process will be the first to kick off with initial production in Q1 2010.
As with all these announcement about advanced process, you have to ask: Where's Intel? Well Intel is saying it will have a 32nm process in full production in the second half of next year. It's not saying anything about 28nm.
While TSMC is saying 'initial' production on 28nm in Q1 2010.
So, the Intel-TSMC process prowess rivalry still looks pretty much even-stevens.
Subscribe to this blog's feed
Comments (1)
In a way, I think it's fair to say TSMC has systematically dropped the ball regarding High-K, and what I've been told with regards to their active power claims is that they're stretching the truth; yes, High-K is much less of an advantage in that context, but if implemented properly it does allow for lower leakage at a given active power level.
At the same time, it is true that it's unlikely to be a big deal for handheld SoCs. And TSMC keeps being very aggressive litography-wise. They definitely have a edge with 40nm, and it is my understanding that their previous roadmap called for 32LP in 1Q10, while now this has changed to 28LPT (with 32G still in 4Q09 and 32HP also becoming 28HP in 2Q10).
Practically speaking, I think the major difference between TSMC and Intel nowadays is that the former doesn't buy enough tools to start any kind of mass production until much later than Intel. That way they get them cheaper than Intel while still allowing for early prototyping/sampling, which is ideal for handheld vendors but probably not so much for the likes of NVIDIA/AMD.
BTW, do you have any clarification on this part of your article: "However Marced did say that the 28nm process would be 'less than 2X' more expensive to customers than TSMC's 45nm process." - did you basically ask whether it would be twice as expensive and he replied that, or did you just ask about pricing generally? And I presume that is for the high-k version? This sure is a big increase either way.
Posted by Arun Demeure | October 1, 2008 5:05 PM
Posted on October 1, 2008 17:05