Intel Has No Process Advantage In Mobile, says ARM CEO

Intel has no advantage in IC manufacturing when it comes to manufacturing processes used for mobile ICs, Warren East, CEO of ARM, tells EW.

“This time last year there was a lot of noise from the Intel camp about their manufacturing superiority,” says East, “we’re sceptical about this because, while the ARM ecosystem was shipping on 28nm, Intel was shipping on 32nm. So I don’t see where they’re ahead.”

Furthermore, with the foundries accelerating their process development timescales, it looks increasingly unlikely that Intel will be able to find any advantage on mobile process technology in the future.

“We’re supporting all the independent foundries,” says East. That includes 20nm planar bulk CMOS and 16nm finfet at TSMC; 20nm planar bulk CMOS and 14nm finfet at Samsung and 20nm planar bulk CMOS, 20nm FD-SOI and 14nm finfet at Globalfoundries.

It gives the ARM ecosystem a formidable array of processes to choose from. “I’m no better equipped to judge which of these processes will be more successful than anyone else,” says East, “our approach is to be process agnostic.”

The important thing is that the foundries’ process roadmap is on track to intersect Intel’s at 14nm.

14nm will be the first process at which Intel intends to put mobile SOCs to the front of the node i.e. putting them among the first ICs to be made on a new process.

Asked if the foundries were prepping their next generation processes with the intention of putting mobile SOC at the front of the node, East replies: That’s the information we’re seeing from our foundry partners.”

Globalfoundries intends to have 14nm finfet in volume manufacturing in 2014, the same timescale as Intel has for introducing 14nm finfet manufacturing.

In fact, GF’s 14nm process may have smaller features than Intel’s 14nm process because, says Mojy Chian senior vp at Globalfoundries, because “Intel’s terminology doesn’t typically correlate with the terminology used by the foundry industry. For instance Intel’s 22nm in terms of the back-end metallisation is similar to the foundry industry’s 28nm. The design rules and pitch for Intel’s 22nm are very similar to those for foundries’ 28nm processes.”

Jean-Marc Chery, CTO of STMicroelectronics points out that the drawn gate length on Intel’s ˜22nm” process is actually 26nm.

Furthermore Intel’s triangular fins, which degrade the advantages of finfet processing could underperform GF’s rectangular fins which optimise the finfet advantage.

At the front of the GF 14nm finfet node will be mobile SOCs says Chian. GF has been working with ARM since 2009 to optimise its processes for ARM-based SOCs.

At TSMC the first tape-out on its 16nm finfet process is expected at the end of next year. That test chip will be based on ARM’s 64-bit V8 processor.

Using an ARM processor to validate its 16-nm finfet process should give TSMC’s ARM-based SOC customers great confidence.

Asked about the effects of finfets on ARM-based SOCs, East replies: “There’ss no rocket science in what you get out of it. The question is does it deliver the benefits at an acceptable cost? You don’t get something for nothing. How much does it cost to manufacture? How good is the yield? And that, of course, affects cost.”

And so on goes Intel beating its head against the wall to get into the low-margin mobile business.

Recently Intel  said it expected its Q4 gross margin to drop 6% from Q3’s 63% to 57%. Shock, horror said the analysts

But if Intel succeeds in the mobile business, its gross margin will drop a lot more than that.

It’s a funny old world.


Comments

11 comments

  1. Drawn gate length was always the figure put on the node, Anonymous, so, if Intel’s drawn gate length is 26nm would it not have helped transparency to call it a 26nm process? I know that may not fit in with marketing’s needs but it would be more in keeping with scientific integrity.

  2. Strange article !!
    Intel 22nm process has a contact pitch of 80nm versus 117nm of TSMC 28nm process so the process is much more denser and advanced. Intel has chosed a 26 nm gate lenght (instead of 20nm for example) because it’s goal is to reduce the leakage without rising too much the clock speeds versus 32nm.
    But i fully understand the desperation of some foundries 🙂

  3. Ok [Anonymous], I haven’t been inside a fab for years — even our own 40nm one — because even if you can get permission to go in you can see precisely nothing and it tells you the same.
    Since nobody has the “next-generation” processes I was talking about yet — including Intel — of course it’s speculation that “it looks like they’ll all be similar performance and available at the same time”.
    “Similar performance” is based on knowledge of the processes and benchmarks, “at the same time” is based on announcements from all three companies, any of which may be optimistic or not — the proof will be when they actually deliver.
    Intel does indeed have the only experience of finFET in mass production, and this was undoubtedly done as a “pipe-cleaner” for their 14nm finFET process because there’s not really such a big advantage at 20nm. It also seems than their “22nm” finFET process is sub-optimal, they had to freeze it to get into production and save some of the tweaks for 14nm. And it’s optimised for CPUs not mobile SoC, partly because Intel don’t care so much about cost for high-margin CPU business.
    The real issue with 14nm finFET is getting it to work with good yield in mass production for mobile SoC, where you can’t sell fast leaky chips as high-speed CPUs and slow low-power chips as mobile CPUs, both at high prices, where restrictive design rules have a much bigger design impact than for CPUs, and where yield and cost are crucial.
    This challenge is exactly the same for everybody including Intel, and many people in the industry are very worried about it, especially because lack of cost saving per chip (or even a cost increase) means that one big driver to move to the next process generation is gone — so many designs will probably stay at 28nm.

  4. Yes [anonymous] I have to admit I’ve been in fabs owned by all those companies. but that doesn’t mean tiddly squat. It’s what comes out the other end that matters. Are the ICs Which come out of the fabs ICs which the mobile OEMs want to buy? So far, at least, the mobile OEMs have voted with their wallets and bought ARM-based chips and not Intel ones. As to gate length it is well known that the gate length on Intel’s ’22nm’ process measures 26nm and 13% of TSMC’s Q3 revenues were on 28nm. So don’t fall for the Intel marketing BS here.

  5. O RLY? Can I ask then have you been inside Intel’s FABs? GF? TSMC? You know what their (Intel) advantage is? They can actually make the chips they are talking about and they have a proven track record.
    “looks like they’ll all be similar performance and cost and available about the same time”
    ^^^This IS speculation and nothing more.^^^
    Guys you are getting way ahead of yourselves here. IF (and its a big one) GF and TSMC can close the gate length GAP they still won’t have anywhere near the manufacturing experience to match Intel (maybe TSMC… MAYBE)

  6. Intel’s current “22nm finFET process” is 28nm metal (shrunk 32nm) with a slightly shorter gate length (26nm?) finFET.
    I’d expect that their “14nm finFET process” (due late next year) will be 20nm metal with a shorter gate length finFET.
    TSMC’s 16nm (CLN16FF) and GF’s 14nm (14XM) processes are also 20nm metal with shorter gate finFETs, and are both due late next year.
    Metal pitch is what sets chip size, so die sizes will all be similar to 20nm bulk planar but cost will be a little bit higher because of the extra process complexity for the FETs.
    All these processes will give lower power and/or higher speed than 20nm bulk planar processes because of lower capacitance/higher drive current for the FETs and also less variation due to random dopant fluctuation.
    So they will all be able to operate at lower (and similar) supply voltages which is where the speed/power improvement comes from — as shown by Intel in their “32nm planar vs. 22nm finFET” comparison.
    This means their main targets *will* certainly include mobile.
    And it looks like they’ll all be similar performance and cost and available about the same time.
    So just where is Intel’s advantage then? (or disadvantage…)

  7. Interesting that TSMC would choose to validate 16nm finfet on the 64bit version of the ARM processor. This is a real indication that the 16nm process is aimed at the high horsepower end of the market rather than the high mileage end.
    To me this indicates that the first devices commercialised from the TSMC 16nm finfet process will be aimed at low-end server farms, high end tablets and even the return of the ARM PC. They won’t be going in to smart- or feature-phones.
    No wonder Mr Buffett sold out of Intel.

  8. Well I remember a Qualcomm guy publicly saying Intel’s Medfield chip-set was “a product of desperation rather than inspiration” [Anonymous]

  9. Mobile is different than the traditional PC market I agree, where I disagree is that Intel is somehow unable to compete for any other reason than they don’t want to (in my opinion they are simply trying to get as much profit from PC sales as possible (if they start making mobile chips as good as their PC chips…)). I mean they do have an ARM license don’t they? They DO have the process advantage (Period!) and they DO have the manufacturing advantage (Not the same thing!).
    As far as Qualcomm is concerned I agree but, you don’t hear them talking smack about Intel do you?

  10. Mobile, mobile mobile [Anonymous] don’t forget wer’re talking moble and the proof of the process pudding is in the power/ performance of the ICs made on the process. And how do Intel’s mobile SOCs stack up against the ARM ecosystem ICs.? Just ask Qualcomm’s customers the answer to that one. And, by the way, Qualcomm can’t make an IC end-to-end but they appear to be doing OK in the IC business.

  11. ARM smack talking Intel what a joke! ARM ecosystem? Remind me again how many sub 28nm ARM chips have shipped? If you can’t make a chip start to finish you shouldn’t even be talking about shipping this or that. All YOU do is draw pretty pictures of what a chip should maybe theoretically possibly kinda sort of look like. OH and TSMC is the only real foundry out there unless Samsung decides its getting in after apple leaves. This idea that 14nm is just around the corner for Foundries is laughable. You know when Intel announces its new products? When they already have the next generation of products working. Think about that when you see GF talking about 14nm (remember 20nm TSV?).

Leave a Reply

Your email address will not be published. Required fields are marked *

*