Exclusive: Intel Opens Fabs To ARM Chips

[DSC]

http://www.forbes.com/sites/jeanbaptiste/2013/10/29/exclusive-intel-opens-fabs-to-arm-chips/

At the ARM developers’ conference today, Intel partner Altera announced that the world’s largest semiconductor company will fabricate its ARM’s 64-bit chips starting next year. An announcement that sent shockwaves throughout the technology industry as Intel is desperately trying to break ARM’s supremacy in the mobile market.

“It’s huge. Imagine ARM’s most powerful and technologically advanced 64-bits processor built on Intel’s leading-edge fabs. A duo that will be hard to beat,” explains Nathan Brookwood, principal analyst at Insight 64.

But this is just the beginning of a much larger endeavor for the chip giant as Intel is even willing to compete with semiconductor foundry leader TSMC for the business of its fiercest rivals, like Nvidia or Qualcomm.

“Intel will build Apple's A7, Qualcomm’s Snapdragon or the Nvidia Tegra for the right price. Now, the question is, are they ready to pay that premium and feed their direct competitor, except for Apple. But that would actually make business sense for everyone,” adds Brookwood.

 

[Lazy8s]

Well, it's still just, so far at least, for their FPGA and other smaller-market partners, and Intel used to make their own ARM compatible processors in the not-so-distant past. So, it's not quite bizarro-world yet...

 

[Laurent06]

That's indeed for high price, low volume, high-end FPGA. Note that Intel already builds 22nm chips with ARM cores for Netronome.

 

[Grall]

I'd really like to see Apple move manufacturing to Intel instead of crappy ole TMSC (or ho-hum Samsung for that matter.) Then we'd know what kind of performance and battery life we could REALLY get from a modern smartphone.

 

[wco81]

Looks like Intel has given up on their own mobile architectures and cut a deal with ARM to use their foundries for ARM-based SOCs.

But not the market share leaders like Qualcomm or Apple, since they use custom CPUs:

http://arstechnica.com/gadgets/2016...makers-to-use-its-10nm-manufacturing-process/

Maybe Intel is just testing the waters with lower-volume vendors before signing a bigger deal.

 

[Grall]

Intel's business is building and selling chips, not building and selling x86 chips specificially. x86 never made sense in the mobile space anyhow; too much architectural baggage and no genuine advantages over ARM.

Intel should court Apple, they could make a great team working together.

 

[wco81]

There's been talk about Intel dabbing Apple's SOCs for years.

That was at a time when Apple was looking to take their fab business elsewhere, after using Samsung fabs exclusively for the first several iPhones.

But are Intel fabs still well ahead of other fabs? Haven't they had problems keeping up with the tick-tock paradigm in recent years?

 

[3dilettante]

From the aspect of its traditional leading architectures dominated by digital logic and performance, it's managed to stay ahead.
Using its fabs for foundry purposes are increasingly necessary given the cost and that Intel has failed to find another way to generate enough internal consumption to justify fabs that are doubling throughput with each node. Intel's cadence is slowing, but the foundries are making some notable compromises at the most recent nodes Intel has avoided.

Cost, more mixed integration, analog, and wireless IP mattered more for mobile even if Intel's CPUs battled back any architectural problems.

Bringing in ARM IP allows for more of that surrounding value and processors for all those lower-cost SOCs or products that bring specific IP to the product and just need to hang it off of some kind of CPU and IO block without getting into the weeds of knowing how to implement them. ARM already provides this and is a more neutral and established provider. Intel is itself baggage for using x86 in that capacity.

 

[Laurent06]

Intel's business is building and selling chips, not building and selling x86 chips specificially.

Except that during Otellini Intel tried hard to push x86 everywhere they could. This has failed in some segments, and it looks like they are still trying to push x86 in IoT where the x86 tax is even heavier than for mobile.

x86 never made sense in the mobile space anyhow; too much architectural baggage and no genuine advantages over ARM.

The baggage (or tax as I call it) is sensitive for lower end chips; as you push performance that tax tends to be diluted into the complexity of the design. IMHO at the level of performance of higher-end phones (Galaxy S, iPhone, etc.) that baggage is not that heavy anymore. But as you write, x86 brings no advantage and quite the contrary due to existing legacy.

 

[Entropy]

There's been talk about Intel dabbing Apple's SOCs for years.

That was at a time when Apple was looking to take their fab business elsewhere, after using Samsung fabs exclusively for the first several iPhones.

But are Intel fabs still well ahead of other fabs? Haven't they had problems keeping up with the tick-tock paradigm in recent years?

Fab business is about much more than the electronic performance of your lithographic process. Intel has had a foundry branch for some time, and it has been woefully unsuccesful, to the point where they had to buy their only potential major customer. Both TSMC and Samsung have their finger on the pulse of the customer needs in terms of the entire tool chain and post wafer processing in a way that Intel has no history of.
If we isolate the discussion to process tech alone, opinions differ, mostly depending on level of discourse and what application area you are targeting. A couple of trends stand out though - Samsung and TSMC have been competing more aggressively the last few years, and look set to continue doing so in the immediate future. Their pace of process improvement will remain high, they both are well underway with their 7nm nodes, and seem rather confident in their 5nm roadmaps. Beyond that the easily accessible crystal balls grow murky indeed. But it is difficult to see that Intel has something that would bring substantial competitive benefits, particularly in market segments outside their own core business. It is both a highly technical business, and fairly secretive, so caveats apply.
And of course just like in mobile SoCs Intel face market constraints. Samsung won't shift their business to Intel, and they couldn't hook Apple even when they did have a clearer process lead. Qualcomm? Mediatek? Huawai and the other chinese outfits? Just where is the profitable business to be found for intel?
To me, this development simply underscores the issue with excess foundry capacity that intel has faced for some time. It is not sprung out of a need in the ARM eco system.

 

[Laurent06]

And of course just like in mobile SoCs Intel face market constraints. Samsung won't shift their business to Intel, and they couldn't hook Apple even when they did have a clearer process lead. Qualcomm? Mediatek? Huawai and the other chinese outfits? Just where is the profitable business to be found for intel?
To me, this development simply underscores the issue with excess foundry capacity that intel has faced for some time. It is not sprung out of a need in the ARM eco system.

There's also another issue: if Intel process really is that much better, can Intel really afford to accept to fab designs that could compete against their Core/Xeon lines?

 

[Exophase]

Except that during Otellini Intel tried hard to push x86 everywhere they could. This has failed in some segments, and it looks like they are still trying to push x86 in IoT where the x86 tax is even heavier than for mobile.

That's what I would think based on everything I associate the IoT term with but Intel seems to have a non-standard notion of what IoT is. Actually, I'm not sure if they have a notion at all. Their newly ostensibly IoT-focused SoCs look like their typical previous generation Atom apps processor SoCs with quad cores, ~2.4GHz clock speed and upgraded Gen 9 GPUs. All with a higher slated power budget. Not at all fitting for what comes to my mind when someone says IoT, which is mainly integration into appliances or small wearables.

They've demoed it in a tiny PCB with a camera which, given the power requirements, doesn't really make a lot of sense except as a component in a much larger product. I suppose they see the GPUs as being useful for computer vision processing instead of graphics, but I really doubt this is going to be a killer app for IoT.

Not going to be too surprised if they announce a Cannonlake based IoT platform next.