ARM on TSMC’s 28nm beats Atom on Intel’s finfet 22nm process, says TSMC
ARM on TSMC’s 28nm delivers a better processor than Atom on Intel’s 22nm finfet process, says TSMC.
“Today, ARM on TSMC’s 28nm gives better performance and power than Atom on Intel’s 22nm finfet process,” TSMC’s President for Europe, Maria Marced, tells EW, “it’s not just technology it’s also the architecture.”
Nowadays, the term ‘28nm’ or ‘20nm’ does not refer to a gate-length. It is merely a descriptor of a node. So the way to judge the quality of a node is no longer the number attached to it, but the characteristics of the devices made on it.
In that respect, TSMC feels it has the edge. “Last year, in an industry which had no growth, we grew 9.2%,” says Marced, “this is driven by the requirements for power in mobile internet devices. It’s all about power consumption. Once again our technology is providing exactly the right products for these devices. More than 50% of our revenues in Q4 came from communications.”
500m smartphones were made last year and TSMC expects a billion to be sold in 2015/6. 60m tablets were sold last year and TSMC expects this market will grow at 30% CAGR over the next five years. Cisco reports that mobile internet traffic will hit 2,000 TBytes per month during 2012.
Asked about reports that customers were having yield problems with 28nm, Marced responds: “Defect density on 28nm is better than on 40nm at the same stage of product life. We have 36 products on 28nm. Compared with 40nm at the same stage, we have three times the number of tape-outs and three times the speed of the deployment.”
28nm production represented 2% of TSMC’s revenues in Q4. “We expect, in Q1, to reach 5% of revenues coming from 28nm,” says Marced, “and in the whole year it will be larger than 10%.”
TSMC says it will be in risk production on 20nm at the end of this year. 20nm will deliver 1.9x density improvement over 28nm, with 25% less power consumption and 15-20% more performance.
The first technology node using finfets will be 14nm entering production in 2014.
With Intel introducing finfets at 22nm, why is TSMC waiting a generation?
“The main reason for not introducing finfets earlier was we have to make sure our ecosystem is ready to design the products,” replies Marced, “we believed the ecosystem was not ready for finfets at 20nm, and we didn’t want to delay 20nm.
What about the reports in the US press that Intel is two and a half generations ahead of the world on process?
“There’s about one generation difference with Intel. We want to reduce that,” replies Marced, pointing out: “Intel does it on its own. We do it with our partners. Is Intel going to foundry? If they are, they will have to produce IP and develop an ecosystem.”
If Intel does go to foundry it will have other challenges beside those two. “It’s difficult to move from one internal customer to many customers,” says Marced, “TSMC makes more than 4,000 different products every year. This is not easy to do.”
And Intel, as a foundry player, would not have a unique advantage which TSMC possesses. “In foundry, customer trust is vital,” says Marced, “we are the only foundry today not competing with their customers. It gives customers the assurance that we take care of their products, their IP, and their business models. We take customers very very seriously. I don’t believe the others can say so.”
TSMC does not think that having a global fab chain is a good idea. “We believe our strength is in having the fabs concentrated,” says Marced, “that’s the best way of improving the way we manage – the way we operate. Having fabs everywhere is inefficient”
TSMC is expecting strong growth from Europe this year which currently delivers about 9% of TSMC’s total revenues. “We expect this year to increase this substantially,” says Marced, “because of the strength of Europe in mixed signal in smartcards, in mobile internet devices a nd in automotive. NFC is going to take off.”
Although TSMC is particularly aggressive in its approach to 450mm, stating that it will be producing on 450mm in 2014/15, it says it does not want hand-outs to develop the technology.
“We don’t expect to get money from the US or the EU to develop 450mm,” says Marced, “innovation should be on top of the agenda. Most of the innovation in Europe is outside the tool vendors. I think the tool industry is small compared to the overall semiconductor industry. It would be better spent boosting the innovative capabilities of fabless companies. Why should we help the tool-makers rather than ARM?”