When he asked new CEO Krzanich whether he would do foundry for ARM-based products the answer he got back was along the lines of 'if they are willing to pay a premium for our process.'
I'd bet that way.
Intel has two lines of business. They design chips, and they make chips. As part of that effort, they spend enormous amounts on R&D of next-generation processes at shrinking geometries, and build or upgrade fabs to use them.
The part of Intel that makes chips looks at the cost of building and operating a fab, and wants the fab to be operating 24/7, generating revenue by making chips for customers.
Intel has leading edge process technologies. If you want what Intel can currently do, where else can you get it? (I don't see Intel being quick to license their latest and greatest to third party foundries when the fact that Intel has it and they don't is a competitive edge in the chip making business.)
Ottelini's "We won't make chips for competitors" was understandable. But Intel is lagging way behind in mobile where ARM currently rules,l and ATOM is not yet competitive. The X86 market is increasingly stagnant. What do those leading edge fabs make? If the customer is willing to pay enough, the answer is probably "What the customer wants."
I think one reason Intel happily produced ARM chips back in the 1990s is they were in a completely different market than x86, so there was no way they could be viewed as a threat. (No pocketPC could be run off an x86.) Instead, Intel was merely extending their domination.
But now, there's politics involved, and Intel seems more interested in helping x86 succeed in the low power market than becoming a leading ARM producer. An obvious non-political reason for this is economics - ARM chips typically only cost $10 at the high end. x86 means far more money for Intel. And the more people that use x86, the larger an audience Intel enjoys where they are an exclusive maker.
In short, making ARMs means competing on a level playing field for Intel, while x86 means serious profit.
Analyst Nathan Brookwood of Insight64 told me at dinner last night that former Intel CEO Otellini had a policy about not making competitive products in its fledgling foundry business. When he asked new CEO Krzanich whether he would do foundry for ARM-based products the answer he got back was along the lines of 'if they are willing to pay a premium for our process.'
I donot think that its obvious that Intel will open its Fab to design companies. ARM has challenged Intel on its turf but Intel has failed in its affort to promote low power chips. Looking at ARM strategy Intel might rethink its position on closed doors.
Intel used to have a division making ARM architecture chips under the StrongARM label. StrongARM has originally been a collaboration between Digital Equipment Corp and ARM, Ltd., and was sold to Intel in 1997. Intel produced new versions of StrongARM in their X-Scale line. Intel sold that operation to Marvell in June 2006, in a reorginization intended to let them focus on X86 products and the desktop and server markets.
And back in the old days, "second sourcing" on Intel X86 chips was common, as OEMs didn't want to be vulnerable to problems at a single component supplier. There was a lot of grumbling when Intel released the 80386 and chose not to allow second sourcing, believing it had the production capacity to meet demand.
Now we see Intel producing ARM chips for a customer using advanced foundry techniques, and stories about Intel attempting to sell X86 licenses.
It's amusing in many respects, but unsurprising. Intel straddles the fence, and is both a designer and manufacturer of chips, and invests enormous amounts in R&D to develop and implement new process technologies in ever shrinking geometries. The costs involved are staggering, and the question is how you recoup them. The only way you can is for the fab incorporating leading edge technologies to be making chips for customers that use those technologies.
The X86 side of Intel might be screaming "ARM! Not Invented Here! Heresy!", but the fab operations will respond "The customer is willing to pay us a lot of money to make these chips, and these fabs cost like crazy. Deal with it!"
I just wonder when we might see Intel negotiate a full ARM license from ARM, Ltd. and go head to head with people like TSMC as a supplier in the ARM marketplace. I'm sure ARM, Ltd. would be happy to do the deal. They don't make chips, they license designs, and while Intel's Atom line competes with ARM, ARM doesn't have a lot of reason to be worried. It will require a rather wrenching change of mindset at Intel.
Blog Doing Math in FPGAs Tom Burke 2 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...