TSMC forecasts near 100% share of 28nm chip market in 2013, may signal massive Apple orders

Hot Damn.. if true, pretty impressive. 

Intel ivy bridge (3rd gen i series) processors are already 22nm

Performance improvement is about the square of the ratio of the transistor sizes. So here (32/28)^2 = 30%

Quote:

Originally Posted by SolipsismX 


And?

If intel could fabricate chips for apple, we could expect dramatic performance and battery life improvements

Quote:

Originally Posted by GadgetCanada 

From the rumors, I'm going to guess an iPhone 5S this year will have a 28nm A6X processor and a fingerprint scanner on the home button. Next year's iPhone 6 could have a 16nm A7 processor, built-in NFC and a new form factor. I'm keeping my 4S until the 6 comes out.

20nm seems more like it for the 2014 iPhone .

Goodbye Samsung, CYA! Wouldn't want to be ya! 

This could be huge, if Apple is finally moving away from Samsung, for at least the CPUs.

That's the news that I take out of this story.

Would Apple go again with just one supplier?

Maybe TSMC for the iPhone on 28nm and Samsung on 32nm for the iPad?

Hopefully they can find a few more reliable suppliers for the chips.
 

Quote:

Originally Posted by arch 

Intel ivy bridge (3rd gen i series) processors are already 22nm

Yes, and they scale like crap wrt the latest from AMD. Global Foundries collaborating with TSMC and to some extent, Samsung have all three stamped out 14nm FinFET for mobile. All three are releasing general consumer products with this tech in a 20nm/14nm FinFET arrangement.

Good article on the 20nm planar/14nm FinFET race: http://semimd.com/blog/tag/14nm/

At any rate, ARM and AMD will be benefiting from this tech to keep pushing the boundaries and Intel will be hitting a brick wall with their poor multi-threading architecture no longer having TDP and single-threading leads that has been beat to death in marketing and fully hiding the fact Intel absolutely stinks at multi-core scalability in fully parallel computing.

With AMD and ARM now partners to create hybrid servers it will be an interesting next 5 years.

I'm not as up to date in knowledge of the chip mfg industry, but here is a list of the top ten mfg equipment companies.  Companies like TSMC, Samscum, Intel, buy the equipment from these equipment mfg and build foundries to actually make the final chips, but they just have to continually invest in new equipment and some are just building foundries to make the latest gen chips.

                                             $bn

ASML                           7.9

Applied Materials           7.4

Tokyo Electron              6.2

KLA-Tencor                 3.1

Lam Research               2.8

Dai Nippon                   2.1

Nikon                           1.6

Advantest                     1.4

ASMI                           1.4

Novellus                        1.3

Quote:

Originally Posted by SolipsismX 


And?

Actually, TSMC also has 20nm technology they have as well.  

Quote:

Originally Posted by SolipsismX 


And?

I read somewhere that TSMC and Apple were working on 20nm processors.  Here's the link.

http://www.digitimes.com/news/a20130116VL201.html

Quote:

Originally Posted by mdriftmeyer 

Yes, and they scale like crap wrt the latest from AMD. 

Intel is working heavily on power consumption and single threaded performance, not ballooning out their cores at the cost of TDP.  The FX-8350 is a 125W TDP, has 8 cores and runs slower than the Ivy Bridge Core i7-3770K with a 77W TDP and sometimes the Core i5 EVEN in some multi-threaded benchmarks. 

IBM has a good process which everyone else (except TSMC) is basing their 20nm/14nm FinFET...but on the ground today they are competing against Intel's 22nm tri-gate process with their current 28nm planar process.  Which for TSMC had some teething pains.  I expect them to get there by 2015 but by then Intel will be at 10nm.

Multi-threaded scalability is also highly dependent on OS and software. It does you no good for your CPU architecture to scale to 32 cores when most OS and apps use at most 4-8 efficiently.  Right now even quite a few enterprise apps don't scale all that great which is why I often see these hugely cored servers used for virtualization rather than performance.

Plus Intel has been working on TSX for improved synchronization efficiency in Haswell so everyone isn't waiting on spinlocks quite as much with better fine and coarse grain locking performance.

GIven that Intel will likely still enjoy a process node and TDP advantage in 2015 and show that they are working improving the x86 ISA to support writing improved multi-threaded software in terms of locks I'm thinking that it is hard to assume that ARM and AMD will be kicking Intel's butt in two years.

Possible but not likely and I've completely glossed over Marvin's favorite:  Xeon Phi.  With 50 cores anything that is embarrassingly parallelizable will do well on Xeon Phi and the tech used on Phi and TSX will cross pollinate between HPC CPUs and mainstream CPUs.  The fine grained control needed in Xeon Phi to manage 50 cores efficiently can be applied to general server chips if Intel needs to massively scale those up to a lot more cores.  

More importantly, Intel's compilers and tools will likely be much better than ARMs.  I've always found them best of breed back in the day when I coded closer to the metal.  That may have changed but the only ones I liked better for parallel/multi-threaded work was Sun's toolchain.  We saw a good deal of performance gains just switching to Intel's compilers over gcc and on par with green hills with IMHO a better tool chain.

Quote:

Originally Posted by drblank 

I read somewhere that TSMC and Apple were working on 20nm processors.  Here's the link.

 

http://www.digitimes.com/news/a20130116VL201.html

http://www.digitimes....... I stopped being interested once I saw that, sorry.

Quote:

Originally Posted by arch 

If intel could fabricate chips for apple, we could expect dramatic performance and battery life improvements

Ha-ha-ha-ha! 

good one. 

Quote:

Originally Posted by SolipsismX 

Intel could but there is no indication that Intel would fab ARM chips for Apple or anyone else and there is plenty of reasons why Intel would avoid doing so. I believe AnandTech ran through the possibilities recently.

Heh, I can't remember what year I predicted Intel chips in the iPhone/iPad but 2014/2015 14nm Intel SoCs should be pretty damned impressive.  

Shame that the 22nm ValleyView (Bay Trail-T tablets) got pushed to Q1 2014.  We'll see the 22nm Merrifield smartphone SoC at MWC and that should be pretty interesting considering the Medfields weren't all that bad.

With the volumes that Apple needs moving the X86 and I think that Intel would be amenable to the same kinds of enhancements that Apple has made in the A series chips for a custom run of Apple SoCs.

Cost per unit is the kicker though.  The A6 is cheap in comparison.

Quote:

Originally Posted by SolipsismX 


Intel could but there is no indication that Intel would fab ARM chips for Apple or anyone else and there is plenty of reasons why Intel would avoid doing so. I believe AnandTech ran through the possibilities recently.

I think people are being blinded by intel's past success here.  They've shown categorically that they don't really have any design chops in this area.  At the very least they've shown that they aren't any better at designing mobile chips that many other shops.  Their advantage has always been scale, and huge production capability, aided by the fact that they had an almost complete monopoly on what they are selling.  

The monopoly doesn't matter now because it's on desktop and laptop processors, not the new mobile ARM stuff.  

Apple's processor design has already shown itself to be the top performer in this new field.  Now they are helping TSMC increase their production capability and scale.  If they can pull that off, intel is history in this market, or at the very least five years away from re-tooling everything for ARM chips and catching up.

If I had intel stock I would have already sold it last year.  

Quote:

Originally Posted by Gazoobee 

Ha-ha-ha-ha! 

 

good one. 

Quote:

Originally Posted by haar 

excuse ME... intels OWN design's, and "power saving mobile" chips are not better than Apples designs... thus how is your statement correct? (where's my coffee...)

perhaps you have confused the power/performance of intel's CPU designs with AMD's ... (AMD's processor designs while working (I have one are disappointing) compared to Intel's Cpu designs... and yes, AMD's GPU's are better than intel's...

and also the game of "what if's" is pointless when talking about intel...

He means because of the 22nm process node advantage over Samsung 32nm.  If Intel were to fab the A6 and A6X for Apple you'd see a nice little power and performance bump just from that.

Plus I think that folks will stop laughing at Intel after MWC.  The Merrifields look pretty decent if the unofficial benchmarks are to be believed and the Bay Trails look like contenders in the tablet market...especially for Surface Pro tablets.

Quote:

Originally Posted by Gazoobee 

I think people are being blinded by intel's past success here.  They've shown categorically that they don't really have any design chops in this area.  At the very least they've shown that they aren't any better at designing mobile chips that many other shops.  Their advantage has always been scale, and huge production capability, aided by the fact that they had an almost complete monopoly on what they are selling.  

Scale and huge production capability is key to being able to afford to go to 10nm and less.  The expectation is only Samsung and maybe TSMC will follow to 10, 7 and 5 nms.

Intel, like Apple, enjoys a huge profitability advantage over ARM fabs given the Atom chips sell for more.

In any case, Medfield phones are middle of the pack.

"The x86 power myth is finally busted. While the X900 doesn't lead in battery life, it's competitive with the Galaxy S 2 and Galaxy Nexus. In terms of power efficiency, the phone is distinctly middle of the road - competitive with many of the OMAP 4 based devices on the market today. If you've been expecting the first x86 smartphone to end up at the bottom of every battery life chart, you'll be sorely disappointed.

...

The performance side is obviously even more competitive. Atom isn't always industry leading in our tests, but the X900 is rarely more than a couple places away from the top (with the exception of GPU performance of course, but that's a matter of licensing a different IP block in future versions). For a reference design that an Intel partner can just buy, barely customize, and ship - that's not bad at all. Smartphone vendors spend a considerable amount of time building phones that perform well - Intel's offer to internalize much of that can be either scary or amazing depending on who you're talking to."

http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone

It should be an interesting couple years in the SoC wars.  Apple jumping to custom Intel SoCs for the iPad is a lot more likely to me than an ARM based MBA at this point.

If Apple is going to help TSMC get to 5nm then having 100B in the bank is a good thing.

Intel generally doesn't make System-on-a-Chip (SoC) modules, they make pure microprocessors.  There are different techniques involved, and Intel doesn't have a lot of experience making tiny SoCs. Samsung is the world leader at it, but Apple seems to be co-investing heavily with TSMC to bring them up to higher volumes and undermine Samsung (thus the billions in capex Apple has spent in the last year).

Besides the fact that Intel hasn't made ARM processors in over a decade and doesn't generally make SoCs, I think working with Intel on this would be slower than TSMC and other companies Apple's iOS team has a working relationship with.  The desktop folks don't even seem to particularly like Intel lately, who seems to try to make a generic version of every hit Apple design.

Quote:

Originally Posted by shompa 

Intel could produce A series chips.
But please understand that Apple would need almost ALL of Intels 22nm manufacturing capacity. Intel don't have the capacity.

People seems to forget that Apple needs 50K 22nm wafers starts / 70K 28/32nm wafer starts each month.

Apple should buy a fab from Global Foundries.

22nm yields are reported to be really good and on par with 32nm yields.   Given that Intel hedged it's bet with 4 22nm fabs.

"This brings us to Intel’s unexpected revelation during their earnings call and the primary reason we're covering this: idle fabs and inventory buildups. Due to a buildup in inventory and a desire to prevent further buildup in what Intel is projecting to be a weak quarter, Intel will be taking the unusual step of letting quite a bit of fab capacity go idle. For Q4 Intel’s fab utilization will be sub-50%, with that being a combination of capacity idled to keep supply down and another fraction idled for the upgrade to 14nm.

...

On that note, perhaps the cruelest part of this is that because Intel’s overcapacity appears to be at 22nm – Intel’s standing inventory is nearly 70% Ivy Bridge – Intel’s options are to produce Ivy Bridge or go idle."

http://www.anandtech.com/show/6378/intel-q312-earnings-3-billion-profit-on-weakening-market-intel-to-idle-some-fab-capacity

I believe that if Tim Cook asked Intel to make 22nm A6 and A7s with a promise for an 14nm Atom iPad design win Intel would take it even if 22nm Atom SoC production took a hit. 

Those 22nm fabs probably aren't idle any more but an Apple design win would be huge for Intel.  Apple could ask for a lot of concessions and it wouldn't hurt Intel if TSMC didn't have the revenue to go 10nm leaving it just Samsung and Intel at the sub-14nm process nodes.

An A6 is 97 mm^2 at 32nm.  On 22nm it would be around 55mm^2 and get around 75%-80% more chips per wafer giving Intel 30% gross margins even at current Samsung A6 prices.

Apart from the tests being set up by Intel, another reason that the Intel part is in any way slightly competitive on power is because the graphics capabilities are abysmal.

Very abysmal. http://semiaccurate.com/2013/01/18/gaming-on-intels-atom-z2670/

A large amount of the TDP on the ARM SoCs is in the GPUs, because the SoCs are far more balanced than the CPU-heavy Intel solutions.

Quote:

Originally Posted by iSteelers 

Actually the fx8350 out performs the i7 in certain tasks. It does suffer in single threaded performance and of course power consumption though only if the chips in question are running at full load. Also that chip is marketed to compete against the i5 in performance not a chip that costs $100+ more. With the resources at its disposal, Intel should be leading the pack in development and yet is still behind Arm in terms of efficiency. The use of cheap thermal paste in the Ivy bridge CPUs heat spreader implies a "it's good enough" attitude not one that is trying to be the innovation leader.

/shrug

Yes, it's cheaper and priced against the Core i5 market but its still considered their flagship CPU.  The contention was that Intel can't scale vs AMD which is a weird assertion given the TDP and IPC advantage Intel enjoys.

Bulldozer was a disappointment and Piledriver isn't that much better.  Piledriver is what Bulldozer should have been.  The FX line USED to be top tier in terms of price and performance...not a budget part that competes against Intel's second string.

In terms of Intel vs ARM, Intel appears to me to be driving power usage down faster than ARM is driving performance up.  10W Haswell is going to have very impressive performance-per-watt numbers.   Enough so that Tom's reports:

"Intel is picking the easiest competitor to beat (Tegra 3 running under Windows RT on a Microsoft Surface), but our own preliminary estimates suggest the 32 nm Atom is going to be roughly equivalent to Qualcomm's 28 nm APQ8060A in the ATIV Tab, and more efficient than the 32 nm Exynos 5 Dual in the Chromebook Series 3 XE303C12.

...

Before we had actual Atom-based hardware running in our labs, it was easy to accept that ARM was more efficient than x86. The Cortex-A9 core is slower than Atom, so it should be more efficient. Then, when we started seeing the -A15's performance numbers and how much faster that architecture is, x86 appeared to be doomed. But nobody really stopped to ask whether -A15's power consumption was still as good, nor had they sat down to run the numbers.  Once we started looking at battery life for a given capacity, we began to think about power use in terms of actual watts consumed during a particular task."

http://www.tomshardware.com/reviews/atom-z2760-power-consumption-arm,3387-5.html

Anand writes something similar:

"We already know that Atom is faster than Krait, but from a power standpoint the two SoCs are extremely competitive.

...

If the previous article was about busting the x86 power myth, one key takeaway here is that Intel's low power SoC designs are headed in the right direction. Atom's power curve looks a lot like Qualcomm's, and I suspect a lot like Apple's. There are performance/power tradeoffs that all three make, but they're all being designed the way they should.

...

I'd always heard about Haswell as the solution to the ARM problem, particularly in reference to the Cortex A15. The data here, particularly on the previous page, helped me understand exactly what that meant. Under a CPU or GPU heavy workload, the Exynos 5 Dual will draw around 4W. Peak TDP however is closer to 8W. If you remember back to IDF, Intel specifically called out 8W as a potential design target for Haswell.

...

The really crazy part is that it's not too absurd to think about being able to get a Core based SoC into a large smartphone as early as 14nm, and definitely by 10nm (~2017) should the need arise"

http://www.anandtech.com/show/6536/arm-vs-x86-the-real-showdown/14

If Intel isn't leading today, it's come a helluva long way and looks like it could lead in the future. Anand seems bullish on the idea.

If I can get Core i7 performance in an iPad Mini by 2016 I will welcome our new Intel mobile overlords.

Keep in mind TSMC also has Qualcomm and Nvidia as customers.  Qualcomm ships alot of chips, and they've been 28 nm for quite some time (Krait).  Even most of the US bound Samsung S3s have Qualcomm krait chips, as do most of HTC's lineup.  

Quote:

Originally Posted by SolipsismX 


Intel could but there is no indication that Intel would fab ARM chips for Apple or anyone else and there is plenty of reasons why Intel would avoid doing so. I believe AnandTech ran through the possibilities recently.

I think one of the reason adnandtech gave was because arm is intels atom series chip's direct competition.  I dont think they want to help the demise of there own product more than has already happened.