Intel foundries to manufacture ARM-based smartphone chips
Intel on Tuesday announced it has reached an agreement with rival ARM Holdings to produce 10-nanometer ARM chip designs at its advanced fabrication facilities, a strategy targeting smartphone chipmakers.
Under the agreement, Intel Custom Foundry will be able use its upcoming 10nm FinFET platform to fabricate chip designs based on ARM's Artisan Physical IP, the companies said in joint press releases. Intel officially announced the collaboration at today's Intel Developer Forum in San Francisco.
For chipmakers like Qualcomm, and potentially Apple, the Intel-ARM partnership opens the door to new foundry options beyond industry stalwarts Samsung and, more recently, Taiwan Semiconductor Manufacturing Co. As noted by Ars Technica, ARM's IP license appears to be somewhat narrow, meaning complex custom designs might not eligible for Intel's fab.
"The initial POP IP will be for two future advanced ARM Cortex-A processor cores designed for mobile computing applications in either ARM big.LITTLE or stand-alone configurations," ARM said.
South Korea's LG will be one of the first customers to use Intel's new service. The consumer electronics giant plans to produce a "world-class mobile platform" of its own design based on ARM IP, likely destined for release in a smartphone. LG traditionally turns to Qualcomm or smaller manufacturers for its chip needs.
With LG already on board, Intel might be angling for bigger fish, namely Apple. The iPhone maker's A-series mobile chips are also based on ARM architecture, albeit heavily customized by in-house engineering teams. Apple has exhibited a willingness to diversify, shifting a large percentage of orders from Samsung foundries to TSMC.
Apple is unlikely to make the switch to Intel's process anytime soon, however, as standing contracts preclude meaningful diversification. For example, TSMC is said to be taping out the design for a next-generation 10nm "A11" processor expected to go into production in 2017.
That being said, Intel is said to have won a spot in the upcoming "iPhone 7," taking a rumored 50-percent share of modem orders away from longtime supplier Qualcomm.
Interestingly, Cowen and Company analyst Timothy Arcuri in July said Intel could be looking to ultimately integrate iPhone's baseband chip directly into Apple's A-series SoC, a design that would save space, offer higher operating efficiency and lower manufacturing costs. To do so would mean a switch to Intel's foundry, a possibility brought within sight after today's ARM licensing agreement.
Under the agreement, Intel Custom Foundry will be able use its upcoming 10nm FinFET platform to fabricate chip designs based on ARM's Artisan Physical IP, the companies said in joint press releases. Intel officially announced the collaboration at today's Intel Developer Forum in San Francisco.
For chipmakers like Qualcomm, and potentially Apple, the Intel-ARM partnership opens the door to new foundry options beyond industry stalwarts Samsung and, more recently, Taiwan Semiconductor Manufacturing Co. As noted by Ars Technica, ARM's IP license appears to be somewhat narrow, meaning complex custom designs might not eligible for Intel's fab.
"The initial POP IP will be for two future advanced ARM Cortex-A processor cores designed for mobile computing applications in either ARM big.LITTLE or stand-alone configurations," ARM said.
South Korea's LG will be one of the first customers to use Intel's new service. The consumer electronics giant plans to produce a "world-class mobile platform" of its own design based on ARM IP, likely destined for release in a smartphone. LG traditionally turns to Qualcomm or smaller manufacturers for its chip needs.
With LG already on board, Intel might be angling for bigger fish, namely Apple. The iPhone maker's A-series mobile chips are also based on ARM architecture, albeit heavily customized by in-house engineering teams. Apple has exhibited a willingness to diversify, shifting a large percentage of orders from Samsung foundries to TSMC.
Apple is unlikely to make the switch to Intel's process anytime soon, however, as standing contracts preclude meaningful diversification. For example, TSMC is said to be taping out the design for a next-generation 10nm "A11" processor expected to go into production in 2017.
That being said, Intel is said to have won a spot in the upcoming "iPhone 7," taking a rumored 50-percent share of modem orders away from longtime supplier Qualcomm.
Interestingly, Cowen and Company analyst Timothy Arcuri in July said Intel could be looking to ultimately integrate iPhone's baseband chip directly into Apple's A-series SoC, a design that would save space, offer higher operating efficiency and lower manufacturing costs. To do so would mean a switch to Intel's foundry, a possibility brought within sight after today's ARM licensing agreement.
Comments
Good for them.
While integrating the modem as part of the SoC allows for certain advantages, TSMC has other advantages over Intel.
Time will tell, but it is doubtful that Apple moves to Intel foundries anytime soon.
The real loser in this may eventually be Samsung's chip division. Intel and TSMC will battle it out for future A-series... Samsung better hope they keep their lead in the display tech Biz over LG... If not, well... Who cares...
Intel getting Apple's chip orders would be insurance in Apple moving all its portables to A-chips.
Does this foretell the ability to run legacy x86 Intel apps on ARM chips?
im pretty sure intel can handle the complexity of the A-chips.
The single biggest improvement to the Apple Watch would be to end the need to tether to an iPhone. Adding gps and lte without reducing battery life will take Apple Watch to the next level in terms of sales. The single biggest factor in making that happen is the fab process. Intel is still ahead of the industry, even if that lead is narrowing. Intel can also offer the lte modem.
In in my mind, the big questions are:
1. When will Intel's 10 nm process be available to customers?
2. How does Intel 10nm compare to TSMC 7 nm?
Now, where can I get a Z80 and a BAZIC interpreter?
Sherm Fairchild just started spinning in his grave!
Yeesus Marta! I've been focused on Apple's use of the Intel modem in the iPhone ... In the Apple Watch that's a whole 'nother story. A smart phone and phone calls are sooo last decade!
Fabrication is a very capital intensive business, much like chip design, and the moves on the chess board must be made years in advance of the results-- so Apple has spent years getting away from Samsung. Getting dependent on TSMC isn't a huge improvement, Apple has made itself independent- both by in-housing chip design and also by making their designs flexible for manufacturing.
So a few years down the line, Intel could win some business from TSMC. Apple will always keep multiple foundries fighting for its business, and thus they will need to compete on process.
Intel and TSMC both are great at process, though intel is ahead.
Legacy x86 will never run on ARM chips, except in emulation. What makes ARM into an ARM chip is its instruction set (ARM stands for "Advanced RISC Machines" and RISC stands for "Reduced Instruction Set Computing". x86 is CISC or "Complex Instruction Set Computing")
The ARM Instruction set is what gives it superior power-per-watt vs Intel approach of chasing superior-power-at-any-watts. To reduce power per watt , Intel can improve process-- and they have-- or go to another instruction set -- and if they do the latter it will no longer be x86.
But take the superior instruction set of an ARM, and the superior SoC Design of Apple and add Intel's process advantages, and that would put Apple even further ahead of everyone else. Literally nobody would be able to compete, even other ARM licensees (because LG is using off the shelf ARM IP, while Apple is doing better stuff in house.)
Except...
You totally don't get it.
The primary constraint on Intel fabbing custom ARM chips like the A-series is not ARM's willingness to allow it. ARM wants as many ARM chips produced as possible, and they don't care who produces them. Their business model is based on royalties and huge volumes.
The constraint has been with Intel. Intel has been extremely reluctant (to put it mildly) to be a large scale foundry. They really, really wanted to avoid the foundry business and stick with x86 for everything. This deal represents a seismic shift in Intel's strategy and attitude. The existence of this deal strongly implies that Intel is now adopting a long term strategy that is compatible with them fabbing for Apple.
I'm sure at this point you're still not getting it. You're going back and re-reading the article, and you're about to say again "but I don't see that in the article! You're a big dummy!"
So let me try it once more.
It's not this specific deal in and of itself that makes an Intel-fabbed SOC for Apple more likely -- it's what the deal represents in terms of Intel's change in strategy. The existence of this deal implies a much higher probability that Intel will make other deals.
Is any of this registering with you?
Your last statement though is filled with issues. ARM is only superior to x86 because it uses less power and heat, and as a result it is only possible to make x86 processors suitable for the low power/low heat mobile devices if you hobble them. But if it were possible to get good x86 performance within the ARM power/heat restraints, it is what everyone would use. This is the main reason why Apple refuses to even consider ARM for any of its laptops or desktops, even the MacBook Air and Mac Mini, which are not considered workhorses.
LG is not using off-the-shelf ARM IP. LG uses Qualcomm, whose ARM cores are heavily customized just as the Ax are customized. You are free to argue that the Ax series is better if you want, but since there will never be an apples to apples comparison - where both the Ax and Qualcomm 8x processors run the same OS in the same hardware - there is no way to tell. That is why the "nobody would be able to compete" thing is impossible to quantify. First off, it is heavily debatable whether Intel's process is better than Samsung's, and Qualcomm has the option of switching to Samsung if they need/want to. And the debate of whether Qualcomm's custom cores are better than Apple's custom cores is really a debate between the iOS and Android systems that run on a Snapdragon 820 or A10. That debate is a bit more difficult now than it was back when Apple was still using 1 GB of RAM so it was easy to say that Android devices that used 2 and 3 GB of RAM had to do so because the hardware, the OS or both were inferior. But now that Apple used 2 GB of RAM for their last iteration of iPads and using 3 GB of RAM is not out of the question for soon-to-be released iPad Pro and iPhone Plus models (either 7 this year or 7s next year) in order to handle the increased demands that the new releases of iOS require - and also since Apple has silently, subtly decided to not care nearly as much about backwards compatibility for older devices as a result - again it is difficult to state that Apple's Ax chips are going to remain so far ahead of everyone else's.
Intel is going down the slippery slop.
They have their back to the wall with missing out on the mobile market for processor technology, partly because the tied their wagon to the horses called Microsoft. They now think the solution to their problem is to become a commodity FAB house for ARM processors for companies who can not get in toTMSC and Samsung since Apple has those suppliers all wrapped around the Apple wagon wheel.
Yeah this is going to work out well for Intel margins, I wonder how the Intel Investors will feel in a could of years when Intel tells them Margin are suffering because the low cost competitors like TMSC eating their margin lunch. I guaranty the market will said this is the best thing that every happen to Intel get people to buy in and then the big guys dump out before the bad news.
I understand the difference between RISC and CISC, and that the legacy x86 instruction set is CISC. But, I did some research (surfing and reading) a while back ...
The [oversimplified] net of that research is that recent Intel chips use RISC instructions and, through built-in, proprietary hardware, translate the x86 CISC instructions into RISC instructions for execution.
If that is accurate, and there is need/desire to continue to run legacy x86 -- it seems that Apple and Intel could collaborate on ARM chips to do this efficiently.