LONDON – Though ARM may currently have an advantage in low power processing, Intel believes it could eventually take the lead if it maintains its current pace of advancement in process technology.
Speaking to the firm’s executive vice president and head of architecture David (Dadi) Perlmutter recently, EE Times learned that Intel is pushing ahead with its sub 20-nm and 14-nm plans, while the ARM ecosystem struggles to find a business model beyond 22 nm.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
Intel is biding its time. The semi giant appears quite prepared to wait a couple of generations until sub 20-nm nodes are breached before bringing the battle of low power processing to ARM and its mobile partners.
Indeed, Intel claims to have line of sight all the way to a 7-nm process, with all the firm’s CPUs becoming SoCs as of next year. Haswell will be Intel’s 22-nm SoC for tablets and Ultrabooks, while Merrifield will be the company’s SoC for tablets and smartphones.
After that, 14 nm is expected to be Intel's killer node with the most dramatic power improvements and integrated baseband--although the firm has yet to officially announce it.
Meanwhile, the ARM ecosystem has a rather more challenging road ahead of it, with low cost continuing to present a challenge for foundry supply, yield and materials science. ARM is seeing Moore's Law slowing down due to a lack of foundry spending and a deficit in R&D capabilities.
Even at the 28-nm, the ARM ecosystem is feeling the squeeze. Until very recently only one foundry (TSMC) could yield 28-nm chips, with Globalfoundries just starting to produce 28-nm in volume, and the entire industry is under-supplied.
With Intel pushing ahead to introduce FinFETs at the 22-nm and hoping to use EUV lithography at 14 nm and below, ARM faces an even bigger crisis of competition.
Meanwhile, Intel’s design teams work hand in hand with the the company's fabs, in one unified effort. “We always work very closely with the fab,” said Perlmutter. “Each time we put a new big challenge in front of the giant team,” he said, adding that every time the focus centered on new and different aspects of the technology.
“If you develop technology, every day you bounce into a problem which may seem to be a showstopper,” he explained, discussing the various challenges the teams had overcome hitting the targets for Haswell’s design.
“What looked to be hugely complicated 10 years ago looks like a piece of cake now. We have to overcome new things every time around, so we have an 'aha' moment every day, every minute in every single Intel product,” he said.
Dadi must be having a laugh! It's Intel that will hit a wall with its obtuse pursuit of performance and power efficiency with x86 on top, banking solely on advances in the physical layer. These are at best linear in nature, whereas higher level optimisations can deliver much faster improvement rates. Moreover, are we to believe that Intel's investment in advanced technology nodes cannot be matched by all the rest of the semiconductor industry?? That would be arrogant and foolish. The reality is that the world has been heading towards open platforms and networks rather than oligopolies, for sometime now. It is this trend that has delivered unprecedented levels of higher performance and lower power AT LOW COST. Consumers realise that and they are not going back to the Wintel duopoly.
PS. I am surprised some people are still buying Intel's propaganda!
This part seems weird:
"With Intel pushing ahead to introduce FinFET to foundries and EUV at 14nm and under, ARM faces an even bigger crisis of competition.
Meanwhile, Intel’s design teams work hand in hand with the foundries, in one unified effort."
Why would Intel work with foundries? Foundries don't have x86-based customers.
Intel is betting on EUV never happening in the future. It's a reasonable bet. Granted that Intel tends to introduce nodes earlier than foundries, it stands to lose if EUV suddenly becomes available, say between its 14 nm node and some lagging foundries' 14 nm node. But it will suffer the same inevitable costs of two generations of double patterning and two generations of quadruple patterning.
least advanced intel chip being sold is still 2x the estimated cost reported on this site for A6 chip in iPhone. Almost every one of intel chip is more like 10x or greater cost of A6. Current Price list is at intel.com for all to see.
So intel will make some magic architecture at most advanced 14nm node, in brand new D1X and fab42...and will cut chip price in half?? I don't think so. Intel has enjoyed and abused monopoly for years, I do not see how intel can compete in volume against entrenched mobile foundry like TSMC, UMC, Samsung, etc. Anyway, current ARM ecosystem is much more consumer friendly than this one you suggest where intel "crush" ARM
It is neither. Analysts have been predicting difficulties in sub 20-nm for ARM for a long time, and the problems at 28-nm prove that the road ahead will only be more difficult. Yes, there is a significant investment going into foundries like TSMC and GlobalFoundries, but even representatives from those fabs don't like to argue about the business model sub 20-nm. It's simply going to get much more difficult and costly. Hardly a conclusion, but it is a deduction. Everything that Dadi said is either attributed to him or is in quotations.
Is the paragraph below what Dadi said, or is this E.E. Times' conclusion about the non-Intel camp? It does seem like investments are fairly robust at TSMC, GlobalFoundries, and Samsung.
"Meanwhile, the ARM ecosystem has a rather more challenging road ahead of it, with low cost continuing to present a challenge for foundry supply, yield and materials science. ARM is seeing Moore's Law slowing down in its foundries, due to a lack of investment and also to a deficit in R&D capabilities."
Join our online Radio Show on Friday 11th July starting at 2:00pm Eastern, when EETimes editor of all things fun and interesting, Max Maxfield, and embedded systems expert, Jack Ganssle, will debate as to just what is, and is not, and embedded system.