How far and fast Intel moves to enter the foundry business remains unknown, especially since is secretive about its long-term strategy. But that's not to say that the company is undecided about the potential opportunity in the foundry business.
I would say au contraire. We should never bet against Intel.
Lately, we've been asking industry sources what they know about Intel's foundry business, and what worries them if it takes the leap. For now, our unscientific survey shows that many remain largely skeptical about Intel being their potential foundry.
One of the most common arguments against Intel as foundry boils down to this: "How could Intel, whose process technology is optimized for its single product (its microprocessor architecture), help meet more complex [design and production] requirements of my chips?"
Many we spoke with implied that they either know or have heard about several big players--in addition to niche startups like Achronix--already using Intel as their foundry. Asked who are those big players, they all demurred. Of course, Intel isn't talking.
When I spoke with NXP Semiconductors CEO Rick Clemmer last week, I asked him the same question—who are the so-called "big players" already using Intel as their foundry? "I've heard of that, too," replied Clemmer. "But I can tell you it's not NXP."
Looking to the future , where Intel may play a critical role as a foundry, we decided to dig deeper. We tapped our most experienced reporters on the semiconductor beat—Peter Clarke, Rick Merrit and Dylan McGrath—to weigh in.
Clarke argues that Intel is not going to become the next big foundry. "But it is a self-confessed technology explorer that is happy to disrupt the fabless/foundry system which supports such competitors as Qualcomm, Broadcom and others," Clarke says. "It's not your father's foundry business anymore." Perhaps, but it's now time for Taiwan Semiconductor Manufacturing Co. (TSMC) or GlobalFoundries to rethink the way they run their foundries.
McGrath, while also skeptical about the possibility of Intel becoming a major player in foundry, nevertheless believes it would be foolish to rule the possibility out. Nobody knows how far Intel is willing to go in foundry, even Intel, McGrath argues.
I would argue that Intel will become a sizeable player in the foundry business. Therefore, I looked for hurdles Intel needs to overcome if it is indeed serious about entering the foundry business. I picked the brain of Semico Research analyst Joanne Itow to discuss Intel's strength and weakness as foundry.
First, Intel is the only company with a logic 22-nm process in production today. This is important especially since TSMC is ramping 28 nm, one generation behind Intel.
Second, Intel's strength is that it is "a true IDM," Itow noted. "Intel can provide services from design to manufacturing and packaging, one-stop shopping." Similar observations were shared by Achronix Chairman John Lofton Holt in a recent interview with EE Times.
Third, let's not forget Intel has plenty of resources ("both personnel and dollars," said Itow) and the company operates fabs worldwide. Intel''s fabs can "lower the risk from natural disaster or political instability," she added.
Fourth, there are clear business reasons that motivate Intel to enter the foundry business. Intel transitions its processor technology every two years. "In the 1990's, Intel migrated their older fabs to chipset production or flash production in order to extend the life of their fabs," observed Itow. "Today, they have to upgrade fabs or build new fabs. The foundry 'sweet spot' is typically one or even two nodes behind Intel's leading-edge processor technology," Itow added.
As the industry prepares for 450-mm production, Intel will have a significant amount of 300-mm capacity still viable at the foundry sweet spot, the Semico analyst noted. "Intel may be willing to provide foundry capacity at a very competitive price in order to bring new life to their 300-mm fabs."
So it is clear intel is not getting into the foundry biz so discussions about that are sort of irrelevant. These are "business relationships" of some kind.
The other thing to keep in mind is the end of moore's law. When you consider this it becomes apparent that at some point sooner rather than later we will go through a contraction followed by consolidation. Fact is Moores law will end, carbon nanotubes, 3d chips, etc will not mitigate that. Given that, you start thinking out 5 to 10 years then you see mobile chip sales decline due to lack of new features b/c lack of more silicon space, ditto for everything else. But you can see that there is still lot's of $ to be made and one of the consistent demand areas will be FPGAs since it is the one place the majority of companies can still play since tapeouts will still be expensive. It's all about consolidation and it appears to me they are on the ball. Now they have to figure out there mobile missteps.
Wafer processing has entered an entirely different phase – can anybody explain to me why these speculations on if/when will Intel enter foundry business?
Intel is now ahead one FULL node of the industry – it is in HV 22nm while “all the leading foundries” (a crowd of one) are still ramping 28nm. This is an entirely new situation – even way before the arrival of 450mm wafers.
In 2014 Intel will introduce Airmont CPU core for its 14nm platform – this will be entirely new situation of almost two nodes ahead. All ARM-based vendors of processors for Smartphones are no longer laughing at Intel as a major application processor vendor for Smartphones. Which foundry can even dream to keep the pace? Another IDM, Samsung, very likely can, however.
The company has laser-focused on the leading phone and infrastructure vendor in China and on the largest smartphone market (China again) – see today’s formation of a joint lab with Huawei for accelerating implementation of LTE (4G) TD technology (a China standard at par with FM standard in 4G).
Why would Intel want to enter the foundry business – except for “pipe-cleaning” with small FPGA start-ups or with a single mega-company (Apple) with whom it already has very close relationship? Same logic for Samsung – except for mega-customers a la Apple or Qualcomm.
There is a perhaps more pertinent topic – how far will foundries expand into IC packaging.
True. Actually, Intel's entry in the foundry market does make a good "long-term" business sense, in my humble opinion.
Intel transitions its processor technology every two years.
As Semico's Itowh said, "In the 1990's, Intel migrated their older fabs to chipset production or flash production in order to extend the life of their fabs.
"Today, they have to upgrade fabs or build new fabs. The foundry 'sweet spot' is typically one or even two nodes behind Intel's leading-edge processor technology," Itow added.
So, why not getting into the foundry business now -- in order to pay for the future investment?
But it takes a lot more than process technology to succeed in the foundry business. Most particularly, it requires pre-packaged, proven IP libraries -- everything from standard cells to I/O cells to memory compilers to common analog blocks like PLLs, LDOs and data converters.
You can't just hand a customer a PDK and a DRC deck and say "call us when you're ready to tape out."
Clearly everyone is missing the boat on this. Intel is ahead on process technology, behind on mobile. There desktop market is declining, there server market will decline as the cloud build out ramp completes. What can they do? Get into the fpga business. So Achronix is a takeover candidate, they are just waiting to see if it works. think 5 years from now Xilinx and Altera are struggling with TSMC to get the next gen devices out, while intel can dominate in power/performance b/c of there process lead.
I think that the consensus is correct. For now Intel is just dabbling in the foundry business. They can dictate the terms of their current foundry relationships. But who knows what the future holds. Intel can and should leverage their leadership in process technology and manufacturing know how to achieve strategic advantages and improve the corporate bottom line.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.