backtop


Print 44 comment(s) - last by snarfbot.. on Jun 1 at 3:25 PM


ARM has already conquered the world of smart phones and tablets, thanks to its Microsoft-inspired approach to licensing. Intel has thus far shown little resistance. Can ARM duplicate that success in the notebook market? (NVIDIA "Kal-el" Tegra pictured)  (Source: KitGuru)

Windows 8 is launching next year, complete with ARM support.  (Source: Google Images)

Corresponding with the launch of Windows 8, ARM will next year launch its new Cortex A15 architecture. Early models are expected to pack four cores and run at 2.5 GHz. Later models will pack up to eight cores. NVIDIA and Qualcomm are already designing A15 chips for laptops and tablets. The architecture is expected to increase power efficiency by a factor of 5 from A9, thanks to a die-shrink to 28 nm.  (Source: ARM Holdings)
Intel hopes to cling to eroding market position

Intel Corp. (INTC) is facing the biggest threat to its world dominance of the personal computing market in almost two decades.  Now it has revealed that it may take extreme steps to try to preserve its lead.

I. The Rise of ARM

The threat materialized slowly.  

Leading the way was ARM Holdings plc (ARMH), makers of the ARM architecture.  Over the last couple decades ARM chips came to dominate the embedded electronics market, being found in everything from cars to washing machines.  But despite their low cost and power efficiency, the ARM architecture was dismissed by Intel as being unable to compete in computing power.

Then ARM hit smart phones.

Starting with Apple, Inc.'s (AAPL) iPhone and Research in Motion Ltd.'s (RIM) Blackberry smart phones, ARM slowly established itself as the exclusive architecture of the smart phone market.  

Today it has taken over yet another market -- tablets.  Apple's iPad tablet and subsequent tablets from Google Inc. (GOOG) all use ARM processors.

With computing demands for these mobile devices ever rising, the computing power divide between ARM and x86 designs from Intel has begun to be bridged, with dual-core and quad-core designs hitting the market.  And the platform's market dominance is also drastically reducing the compatibility app, with ARM apps now able to handle most popular file formats like Excel and Word documents.

II. How Mobile Was Won

Key to ARM's dominance has been licensing its architectures rather than insisting on making them itself.  The company has over 200 licensing partners, including seven in China and 15 in Taiwan.  Among the highest profile ARM chipmakers are South Korea's Samsung Electronics (005930) (which co-designs chips with Intrinsity) and U.S.-based Texas Instruments Inc. (TXN), Qualcomm Inc. (QCOM), and NVIDIA Corp. (NVDA).

Much as abandoning a closed-garden model helped Microsoft beat Apple in the 90s PC market, ARM has similarly leveraged its open licensing in mobile to outprice and outclass Intel.

Microsoft has already announced that the next version of Windows -- Windows 8 -- will support ARM notebooks and tablets.  Google's Chrome OS, which is expected to make a spash in the operating system space will also support ARM designs in the near future.  Even Apple is rumored to have ARM netbooks/notebooks in the works.

With the market drifting towards ARM -- in the mobile space, at least -- Intel is desperate to hold off the rival architecture.

To that end Intel, according to Reuters, is considering allowing rival chipmakers like Advanced Micro Devices, Inc. (AMD) to use its fabs.  Such a "coalition of the willing" formed by x86 chipmakers -- primarily Intel and AMD -- could hold perhaps hold off ARM longer than Intel alone.  

Intel has developed significant process upgrades recently, including 3D transistors.  However, ARM's fab partners, such as Taiwan Semiconductor Manufacturing Company, Limited (TSM) are working on similar technologies.

Notably, Intel still seems opposed to licensing its own x86 architecture for modification by third parties and independent fabs, as ARM Holdings has done.  It says it has considered such a move, but characterizes it as "a tough decision".

III. ARM Plots PC Takeover

Even as Intel wavers on the topic of licensing, ARM is hungrily eyeing the notebook or netbook market, where its cheap, power-efficient central processors could make a big splash.

Arm Holdings President Tudor Brown spoke at the Taiwan hardware convention Computex this year, stating, "Today we have about 10 percent market share [in mobile PCs]. By the end of 2011 we believe we will have about 15 percent of that market share as tablets grow. By 2015, we expect that to be over 50 percent of the mobile PC market."

But he says the focus is still on smart phones and the embedded market, commenting, "Traditional PC chips only accounted for 6 percent of our whole PC chip shipments of 6 billion units last year; the contribution will still be less than 10 percent in the future."

The message seems clear -- ARM feels confident that it can take half the tablet, netbook, and notebook market without significant effort.

Mr. Brown expressed confidence that it will at least keep up with Intel in process development.  He says that his firm is working closely with hardware giant International Business Machines Corp. (IBM) to put cutting edge transistor technology into the hands of ARM licensees.  He states, "We've been working at 20 nanometer and we've had test chips at 20 nanometer for over a year."

He says that IBM and ARM Holdings are currently testing 14 nm chips.  Intel is currently manufacturing chips at the 32 nm node, though it is working on its own future 22 nm architectures code-named Haswell and Broadwell.  Intel also is reportedly starting to formulate 14 nm architectures dubbed Skywell and Skymont.  Intel has been far more secretive of these designs, though, than ARM has with its own 14 nm plans, so it's uncertain exactly how far Intel has come.

Mr. Brown hinted that he feels Intel will need to make major changes to stay competitive.  He comments, "It's important that Intel makes a change in their business model. Intel has never done this before, so we have to take it very seriously."

IV. Cortex A15 -- Finally a PC-Worthy ARM Core

Key to ARM's netbook/notebook invasion will be the Cortex A15 core. The A15 design introduces core "clusters", allowing the core count to double from four cores with the previous A9 architecture to eight cores (four cores each in two clusters).  The new design also allows chip speeds of up to 2.5 GHz, bumping the A9's limit of 2.0 GHz.

Additionally, the smaller transistor size allows for even more dramatic power reductions.  ARM officials say the new design will offer five times the processing per unit of power versus A9.

NVIDIA, who announced its A9-based Kal-el quad-core system on a chip today, is planning to launch a A15 design dubbed "Wayne" sometime in 2012 (an allusion to Bruce Wayne of DC Comics fame).  Wayne will be built on a 28 nm process.

Qualcomm, currently the leading Android ARM chipmaker, will release a quad-core 2.5 GHz A15 chip, dubbed APQ8064, in "early" 2012.

Many PC enthusiasts scoff at the possibility of ARM gaining significant notebook market share.  This perspective is understandable -- a dual core 1.5 GHz processor doesn't necessarily have the power needed for a notebook.  However, a 2.5 GHz quad-core sure might.

Next year should be an intriguing one in the personal computing market, with the launch of Windows 8 and the Cortex A15 core design.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

ARM partners are a long way from 3D transistors
By lol123 on 5/30/2011 2:45:35 PM , Rating: 2
"Intel has developed significant process upgrades recently, including 3D transistors. However, ARM's fab partners, such as Taiwan Semiconductor Manufacturing Company, Limited (TSM) are working on similar technologies." There is no reason to believe that TSMC or even IBM will catch up on Intel with 3D transistors any time soon. As everyone in the industry knows, Intel's manufacturing advantage is enormous and the gap is widening, not closing.




By JasonMick (blog) on 5/30/2011 3:11:36 PM , Rating: 2
quote:
"Intel has developed significant process upgrades recently, including 3D transistors. However, ARM's fab partners, such as Taiwan Semiconductor Manufacturing Company, Limited (TSM) are working on similar technologies." There is no reason to believe that TSMC or even IBM will catch up on Intel with 3D transistors any time soon. As everyone in the industry knows, Intel's manufacturing advantage is enormous and the gap is widening, not closing.


Somewhat true, somewhat not true.

TMSC plans 3D (tri-gate) transistors at the 14 nm node, which if hopes to reach in 2015 or 2016:
http://www.eetimes.com/electronics-news/4213622/TS...

With sources indicating Ivy Bridge will make an early 2012 launch, that puts TMSC roughly 3-4 years behind.

That said, the picture isn't as simple as you paint it.

Past tri-gate transistors (which have been talked about for nearly a decade now) there's no clear answer to cutting power consumption, other than die shrinks. And Intel's rivals aren't significantly behind in die shrinks -- they're pretty much on par with it.

Further, though Intel looks like it will deploy tri-gate first it is unknown what the cost associated with it will be. Sandy Bridge is already too expensive to compete effectively with ARM/Fusion. Intel's budget offerings (Atom) are currently on the 45 nm node, where as competitive ARM CPUs are on the 28 nm node.

In other words, by the time Intel tri-gate tech trickles on down to ARM's competitor -- Atom -- ARM manufacturers may already have deployed the tech as well.

In short, it's anyone's game at this point, but Intel no longer looks as invincible in the mobile PC market as it once did.


RE: ARM partners are a long way from 3D transistors
By lol123 on 5/30/2011 3:42:38 PM , Rating: 3
Intel has 14nm in the roadmaps for 2013 and will have mature tri-gate technology by then. That's nothing less than a huge advantage over TSMC.

I don't see what you mean when you say that "And Intel's rivals aren't significantly behind in die shrinks -- they're pretty much on par with it." GlobalFoundries are struggling to get their 32nm process out, and TSMC to get 28nm out Q4 this year, while Intel is preparing their move to 22nm. Intel's processes also tends to have better performance at any give node compared to the processes of most other companies. (Look for "Process Technology at IEDM 2008" at Realworldtech and the tables at page 11, apparently I can't post links)

The cost increase of tri-gate technology has been estimated by Intel to be 2-3% compared to the equivalent planar process.

"In other words, by the time Intel tri-gate tech trickles on down to ARM's competitor -- Atom -- ARM manufacturers may already have deployed the tech as well."

That's simply not true, by any estimates. Atom will be at 22nm (with tri-gate) at the latest in 2013, and your own source stated that TSMC hopes to implement tri-gate technology in their 14nm process in 2015 or 2016.


By JasonMick (blog) on 5/30/2011 4:45:49 PM , Rating: 2
quote:
"In other words, by the time Intel tri-gate tech trickles on down to ARM's competitor -- Atom -- ARM manufacturers may already have deployed the tech as well."

That's simply not true, by any estimates. Atom will be at 22nm (with tri-gate) at the latest in 2013, and your own source stated that TSMC hopes to implement tri-gate technology in their 14nm process in 2015 or 2016.


Let's consider...
Silverthorne (45 nm) -- Q3 2008 (launched)
Cedar Trail (32 nm) -- Q4 2011 (tentative)
----------------------------
Gap: 3 1/4 years.

Intel has promised a dramatically accelerated Atom schedule with die shrinks every year (2013 launch of 22 nm, 2014 launch of 14 nm). But at this point it's purely speculative whether Intel will be able to stick with that.

If it does great, but the company's current track record shows its Atom die shrinks taking 3 years. Is it possible that this will dramatically accelerate? Perhaps, but don't be surprised to see 22 nm get bumped to 2014, 14 nm to 2016, etc.

quote:
The cost increase of tri-gate technology has been estimated by Intel to be 2-3% compared to the equivalent planar process.


Doubtful, when the full cost of fab upgrades is factored in.

Regardless, that's a 2 to 3 percent premium over Sandy Bridge, not Atom. ARM competes with Atom in price.

So to be cost competitive Intel will need to price Ivy Bridge/Atom 22 nm at a comparable price point to ARM chips. That's a much bigger hurdle to jump than pricing it close to Sandy Bridge (which is already quite expensive).

quote:
I don't see what you mean when you say that "And Intel's rivals aren't significantly behind in die shrinks -- they're pretty much on par with it." GlobalFoundries are struggling to get their 32nm process out, and TSMC to get 28nm out Q4 this year, while Intel is preparing their move to 22nm. Intel's processes also tends to have better performance at any give node compared to the processes of most other companies. (Look for "Process Technology at IEDM 2008" at Realworldtech and the tables at page 11, apparently I can't post links)


Again, I disagree.

Samsung is supposed to be starting 20 nm test production by the end of this year.
http://www.xbitlabs.com/news/other/display/2011022...

Intel has shown working 22 nm silicon, but I don't believe they've started full fledged test production of 14 nm silicon, so I believe Samsung will have the smallest test production chips, for the moment, if it can stick to this schedule.


By lol123 on 5/30/2011 6:11:38 PM , Rating: 2
Intel's tick-tock model over the last few years has been very accurate and they have not been late by more than a few months (mostly due to the Sandy Bridge chipset recall) with a new process shrink or a new architecture. Whether to put Atom on 22nm as soon as that node is available in 2011/2012 or to do that in 2013 is a business decision, not one limited by their technology, as was probably the decision to delay the first 22nm chip, Ivy Bridge. It's also likely that they want the process to mature a bit before they move the Atoms to it. They would not have put 22nm in the roadmaps for late 2011/early 2012 if they were not confident that they could deliver chips at that node by then, and you can be very sure that they have worked at their 22nm technology including the tri-gates since a long while back.

I don't know if you actually read the article about Samsung, but it says right in it that "Actual commercial chips made using 20nm process technologies are years away" and that the manufacture of 20nm wafers later this year will be a test run (just like it says in the title). I also don't know what makes you believe that Samsung would test its 20nm process (which likely will not be as good as Intel's 22nm, even if it was planar; have a look at the tables I referenced in my previous post) several years before production, while Intel would not or could not do the same with its upcoming nodes in the same timeframe. Intel's manufacturing advantage is unquestioned and that's unlikely to change in the foreseeable future. Their research and roadmaps are for what will be relased in maybe 10 years from now, because their stock price depends on node shrinks being continual and punctual.

The 2-3% cost increase figure is per wafer and is the same regardless of the architecture of the chips on it. I don't know if Intel have included the cost of upgrading the fabs, but those apply to all process node shrinks and the same price increases, if any, from the capital cost of fab upgrades would have been seen on every new chip generation up until now.


By omnicronx on 5/30/2011 6:59:00 PM , Rating: 2
quote:
Again, I disagree. Samsung is supposed to be starting 20 nm test production by the end of this year.


Test production being key word here, from what I've seen there will be limited test quantities, but the production release date is pretty far away.

"Risk production is slated for the second half of 2012, with early production due in the first quarter of 2013."

http://www.eetimes.com/electronics-news/4212280/Sa...

Other articles on the matter even mention it will be a while off before we see commercial quantities.

http://www.thinq.co.uk/2011/2/24/samsung-promises-...


RE: ARM partners are a long way from 3D transistors
By Samus on 5/30/2011 10:46:35 PM , Rating: 2
No matter how superior Intel's manufacturing process is, there is no denying the efficiency of RISC. It's entirely possible an A15 on 32nm could be competitive with a 18nm Intel chip.

But there is nothing stopping Intel from licensing ARM (they have in the past with Itanium) and manufacturing a competing chip. x86 has been showing its age for over a decade with the introduction of MMX and other various extensions to improve its efficiency. Modern chips support nearly a dozen additional instruction set extensions just to help it along with various rendering, encoding and [de]compression tasks. x64 extensions just to map larger memory space. etc... From a software perspective, this is ridiculous.

RISC architecture has the advantage of not needing extensions as whatever is implemented in hardware is utilized, and if you run modern software on old hardware, the missing hardware support is self-emulated at the cost of performance. There is no need to modify software, ever.

This is a forward thinking approach to pushing technology forward. Intel is blindly holding onto x86, instead of embracing Microsoft's decision to (finally) support an alternative architecture.


By encia on 5/30/2011 10:51:18 PM , Rating: 2
Read http://www.dailytech.com/AMD+Fusion+Emerges+as+Ser...

"AMD Fusion Emerges as Serious Threat to Intel in the Notebook Mid-Market"


By encia on 5/30/2011 11:07:44 PM , Rating: 2
MS Windows NT PPC/MIPS/Alpha/Itanium says Hi. Somebody didn't read up on MS Windows NT's history.

Read http://i328.photobucket.com/albums/l327/encia/AMD_...

Notice how small is the X86 decoders with AMD Bobcat i.e. 1 to 2 percent of the die size.


By encia on 5/30/2011 11:17:03 PM , Rating: 2
ACER Iconia W500 tablet reportedly has 5 watts version AMD C-50. Cutdown to 40 stream processors from 80 stream processors, 3 watts would be in range.

From http://www.fudzilla.com/processors/item/22893-amd-...


By omnicronx on 5/30/2011 6:33:36 PM , Rating: 2
quote:
In other words, by the time Intel tri-gate tech trickles on down to ARM's competitor -- Atom -- ARM manufacturers may already have deployed the tech as well.
You may want to check the supposed release date for tri gate Atom chips. That date is slated for 2013 (even Anand states this), 2 full years after their first trigate tech is deployed (so the technology itself is not in its infancy either).

That implies 2 full year advantage over when TMSC before they will START using tri gate like technology, not when it will actually be scaled up for mass production use.

Atom being one die size behind (and usually using older facilities) very well may not have any kind of true ramp up period as their first attempt. (to occur later this year)

So at the very least we are talking 2 years, probably more when you consider any ramp up period..


By omnicronx on 5/30/2011 6:43:36 PM , Rating: 2
The reality of the matter is its not tri gate alone that will keep perhaps give ARM a run for its money. Its if they can stay a step competitive with die shrinks.

Unless Intel can match ARM step for step, the tri gate advantage will be vastly mitigated.


By SPOOFE on 5/30/2011 9:07:33 PM , Rating: 2
quote:
That implies 2 full year advantage over when TMSC before they will START using tri gate like technology, not when it will actually be scaled up for mass production use.

Yup, and I'm excited to see how the tech does when it's released. However, you're looking at it from the wrong angle. Right now, with regards to the computing space under discussion, the momentum is in ARM's favor. Intel dominates high-performance computing, but they've been talking up smartphones for years now with no success. It's their game to lose, and they NEED the tri-gate transistors to get their product into smartphones and tablets. ARM? The worst they have to "lose" is the desktop market, which they've never had and which has become slightly less relevant in terms of the consumer market.

Of course, we all know that Intel is capable of really knocking one out of the park when the heat is on, so like I said, it'll be interesting to see how it pans out.


"Young lady, in this house we obey the laws of thermodynamics!" -- Homer Simpson














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki