Blog
Comment
PV-Geek
For Intel the foundry business is a small side project. It can afford to ...
ibm221
Intel is a technology explorer, not a foundry
Peter Clarke
5/1/2012 12:21 PM EDT
Intel will not become the next big foundry. As advertised, it will remain a technology explorer content to disrupt the fabless/foundry ecosystem that supports such competitors as Qualcomm, Broadcom and others.
Intel CEO Paul Otellini has said Intel fabricates wafers for others as a way of checking out technologies that may be relevant to Intel in the future. Otellini specifically mentioned fast processor variation. The best way to learn about these technologies, he added, is to find partners who want to make them. As a result, Intel gets paid to learn if these technologies will fly or whether they required modifications.
Intel's work with Lilliputian Systems on fuel cells has implications for ultrabooks and smartphones. FPGA fabric for processors fits this bill as does novel memory technologies, 2.5-D and 3-D IC assembly.
At the same time, supplying just a few wafers to the likes of Achronix, Tabula and Netronome can be very disruptive to the established fabless/foundry ecosystem. This will be the first time a rival FPGA vendor can make Xilinx, Altera, Lattice feel the pain of being one or two process nodes behind in terms of power and performance. Similarly, Intel partners Netronome may be able to give other fabless network processor vendors like Broadcom and Cavium a run for their money.
ARM CEO Warren East acknowledges that Intel retains a manufacturing advantage but argues that the battle for ARM partners is about the relative merits of manufacturing and design. East argues that its Cortex-on-28-nm technology is superior to Atom-on-22 nm. But in other sectors, such as networking and FPGAs, it's unclear that there is sufficient design superiority to make up for the manufacturing benefit Intel provides. That may hurt pure-play foundries, but that does not mean Intel wants to be one.
That said, Otellini has openly discussed the possibility of taking on foundry business for Apple or Qualcomm. Under present circumstances, that would promote the ARM architecture in mobile at the expense of Intel's aspirations for Atom. That's because Intel wants its Atom-based processors to gain traction in the mobile market to replicate its past success in PCs, but most smartphones and tablets are based on ARM designs.
If either Qualcomm or Apple is prepared to design around the Atom processor core, that would be disruptive. On the other hand, such a move would put Intel in competition with its new foundry customers.
There is one other point to bear in mind: It's no longer your father's foundry business.
The days when foundries just churned out CMOS wafers with little knowledge about the underlying designs are long gone. IDMs and foundries both must be collaborative technology explorers in order to prosper.
Indeed, pure-play foundries likes of TSMC, Globalfoundries and United Microelectronics Corp. may instead be forced to become more like Intel and Samsung, finding new ways to accelerate their process R&D so that they can catch up with Intel. Failure to adapt may mean being relegated to work behind the leading-edge on "more-than-Moore" designs.
Intel CEO Paul Otellini has said Intel fabricates wafers for others as a way of checking out technologies that may be relevant to Intel in the future. Otellini specifically mentioned fast processor variation. The best way to learn about these technologies, he added, is to find partners who want to make them. As a result, Intel gets paid to learn if these technologies will fly or whether they required modifications.
Intel's work with Lilliputian Systems on fuel cells has implications for ultrabooks and smartphones. FPGA fabric for processors fits this bill as does novel memory technologies, 2.5-D and 3-D IC assembly.
At the same time, supplying just a few wafers to the likes of Achronix, Tabula and Netronome can be very disruptive to the established fabless/foundry ecosystem. This will be the first time a rival FPGA vendor can make Xilinx, Altera, Lattice feel the pain of being one or two process nodes behind in terms of power and performance. Similarly, Intel partners Netronome may be able to give other fabless network processor vendors like Broadcom and Cavium a run for their money.
ARM CEO Warren East acknowledges that Intel retains a manufacturing advantage but argues that the battle for ARM partners is about the relative merits of manufacturing and design. East argues that its Cortex-on-28-nm technology is superior to Atom-on-22 nm. But in other sectors, such as networking and FPGAs, it's unclear that there is sufficient design superiority to make up for the manufacturing benefit Intel provides. That may hurt pure-play foundries, but that does not mean Intel wants to be one.
That said, Otellini has openly discussed the possibility of taking on foundry business for Apple or Qualcomm. Under present circumstances, that would promote the ARM architecture in mobile at the expense of Intel's aspirations for Atom. That's because Intel wants its Atom-based processors to gain traction in the mobile market to replicate its past success in PCs, but most smartphones and tablets are based on ARM designs.
If either Qualcomm or Apple is prepared to design around the Atom processor core, that would be disruptive. On the other hand, such a move would put Intel in competition with its new foundry customers.
There is one other point to bear in mind: It's no longer your father's foundry business.
The days when foundries just churned out CMOS wafers with little knowledge about the underlying designs are long gone. IDMs and foundries both must be collaborative technology explorers in order to prosper.
Indeed, pure-play foundries likes of TSMC, Globalfoundries and United Microelectronics Corp. may instead be forced to become more like Intel and Samsung, finding new ways to accelerate their process R&D so that they can catch up with Intel. Failure to adapt may mean being relegated to work behind the leading-edge on "more-than-Moore" designs.
|
|
|
|
|
Navigate to related information
chipmonk
5/1/2012 1:28 PM EDT
Discrediting the ARM-Fabless-Foundry model by demonstrating superior technology like 22 nm finFETs and panicking them is clearly a part of the strategy. At last week's Ivy Bridge launch it was claimed that performance will be 20 % higher at a 20 % lower power consumption. But die price too could be cheaper as there also seems to be a 60 % shrink in die size ( 160 sq mm vs 212 sq mm ) when going from the 32 nm Sandy Bridge to Ivy Bridge. This in itself is a larger margin than the profit margin of the ARM based fabless wonders who now dominate the smart phone / tablet space and could be used to upset the "Apple" cart in the next 2-3 years while the Foundries keep struggling to stabilize their 28 nm yield, not to mention rise to the challenges of 22 nm and finFET.
Sign in to Reply
dylan.mcgrath
5/1/2012 2:55 PM EDT
Another way of looking at this question is: Is TSMC afraid? TSMC certainly doesn't appear to be quaking in its boots. But as you yourself pointed out, Peter, TSMC recently increased its capex plan for the year to about $8.5 billion, much more than most people though they would spend. Is that a signal it's girding for battle with Intel? Also, at the recent TSMC Technology Symposium, Morris Chang made some comments about how TSMC would never be in competition with its own customers. I thought (and still do, basically) that those comments were a shot at TSMC, but maybe they were also meant to put doubt in potential customers' minds about Intel.
Sign in to Reply
dylan.mcgrath
5/1/2012 7:08 PM EDT
Sorry- in above comment, meant to say "I thought those comments were a shot at Samsung."
Sign in to Reply
rick.merritt
5/1/2012 5:28 PM EDT
I think TSMC's biggest fear is not Intel but it is getting stuck with too much capacity in a down market.
I think it's second biggest fear is having too little capacity in an up market and losing or disappointing customers, something it is currently experiencing with Qualcomm.
Sign in to Reply
Bruzzer
5/1/2012 11:07 PM EDT
Considering all the inputs, technology explorer, yes Intel always loves to learn off other's developments to expand their own. Including how to efficiently manufacturer an x86 SOC. There’s obviously learning curve associated with producing $15,000 revenue wafers compared to $150,000 revenue wafers. On foundry guerrilla attack why not it’s in Intel's blood. On cut off others that too. Open the flood gates, well, if there's a 28 nm capacity crunch Intel's ready to supply that capacity at a tidy profit. And no skin off the 22nm projects including 3D stack at a monopoly profit. On market acceleration and Intel selling its own parts faster than they can be make them consider the lure of $150,000 margin values and a channel conditioned to take the risk of stocks. Multi geographic presence the hub and spoke behemoth of an Intel Nation production aggregate?
Mike Bruzzone
Camp Marketing
Sign in to Reply
ibm221
5/3/2012 2:08 AM EDT
TSMC 's biggest problem is it is stucked in a tiny island with 30million population (smaller than texas?)
their IQ need to be 5 times of US to put up a real fight.
Samsung has been enlightened and made up it's mind on a 30billion $ facility at inner china, if it works out, intel will be history, as Kodak.
Sign in to Reply
peter.clarke
5/3/2012 5:53 AM EDT
@ibm221
You say $30 billion facility but the last I heard it was $7 billion on NAND wafer fab complex.
Not saying that Samsung would never make logic there but it will take a while.
And with regard to IQ. When you consider how many engineering graduates Taiwan and China produce and how many engineering students in the western hemisphere are from the eastern hemisphere, perhaps the collective IQ is greater, at least in terms of engineering.
Sign in to Reply
ibm221
5/3/2012 9:58 PM EDT
$7b for phase 1, only.
you need chinese capability to keep on pace,
http://finance.sina.com.cn/chanjing/gsnews/20120417/145011846605.shtml
the IQ potential of china is huge, the future for US could be like france or Germany, utilize your EQ, artistic talent etc.
Iphone is a good example, shining metal boxes for big bucks.
Sign in to Reply
PV-Geek
5/4/2012 9:10 PM EDT
For Intel the foundry business is a small side project. It can afford to experiment. For TMSC it is their entire bread and butter. It is all about volume and handling a diverse array of designs and products. Intel can tune their main fabs for very few custom built products. That makes it easier to move to a more advanced technology like FinFets when you don't have to worry about the catering it to the masses. I would like to see how well their yields would hold up if they let 50 different customers flood a fab with a wide array of product types simultaneously. When Intel has a yield problem they can make their design teams change the design. At a pure play foundry you have to cater the process to the designs instead.
Sign in to Reply