SAN JOSE, Calif. – Intel Corp. will build FPGAs for Altera Corp. using its 14-nm FinFET process technology in a deal that turns up the heat on TSMC in foundry and Xilinx in high-end FPGAs. The deal marks the largest of a string of publicly disclosed foundry deals for Intel to date--and its first at 14 nm--but is not expected to result in products until 2014.
Altera (San Jose, Calif.) declined to disclose details of the deal, including what products it will make when. However, Altera CEO John Daane did say he believes Intel is two to four years ahead of other foundries with its 14-nm FinFET process, which Altera will use initially to give its highest-end FPGAs advantages in density, performance and power.
High-end parts make up about half the FPGA market, with Altera claiming a lead with 40- and 28-nm parts that it aims to extend with the new Intel process. Besides winning more business away from rival Xilinx, the 14-nm parts could help Altera grab more sockets away from ASICs and application-specific standard devices, Daane said.
Intel promised Altera access to the 14-nm process for 12 years to satisfy long-term availability requirements of defense and other customers, Daane said. The multi-year deal allows Altera to use other existing and future nodes, but the FPGA maker initially will focus on high-end parts at 14 nm, he said.
Using multi-die chip stacks, Altera currently ships an FPGA that packs 1.2 million logic elements, lagging a similar chip from Xilinx with 2 million logic elements. However, such parts have relatively high costs and power and take a performance hit due to additional on-chip communications. They are used “for prototyping predominantly—it’s a niche,” Daane said.
Altera surveyed foundries for a year before striking the deal with Intel. It will continue to make chips at TSMC and conduct ongoing evaluations of other processes as they develop.
Daane cited reports that other foundries are grafting a first-generation of FinFETs on to existing 20-nm design rules to create what they are calling a 14-nm node. “Intel’s 14-nm is a second generation FinFET process, while others are just starting to implement their first,” he said.
The deal marks "a significant departure for Altera," said Deutsche Bank analyst Ross Seymore, who doesn't expect Altera to see revenue from it until 2015. It is also "a validation of Intel's manufacturing leadership" that "should help Intel make gains in foundry services," he added.
"It is not Intel's objective to become a general foundry service provider," said Len Jelinek, a chief analyst at IHS iSuppli. Rather it aims "to select a few high volume [foundry] clients [that] provide Intel with an additional revenue stream to help defer the cost of its advanced manufacturing capability," he said.
It makes sense for Intel and Altera.
As you move into progressively smaller geometries, fewer foundries qualify. Altera wants to move to 14nm for new products. Who else *can* make them?
But as you move down into that range, the cost of building a fab rises into the ionosphere. Very few companies can afford to build them at all, and those that can want to be sure a substantial market is there for what the fabs will make.
The deal with Altera provides Intel with a chunk of revenue to help offset the design and construction costs of 14nm geometries, and Altera is not a direct competitor to Intel. The question is what Altera has in the pipeline that it might ask Intel to manufacture for them.
Intel is essentially in two businesses, design and manufacture. Competitors like ARM are in one: design. Intel has historically designed what it made, but as growth in the X86 market flattens, the question arises of what business will keep those oh so expensive foundries operating at capacity.
"It is not Intel's objective to become a general foundry service provider," sounds quite true, but they also need to keep the fabs turning out product and generating revenue 24/7 to justify having them. Juart filling in the corners and mopping up any idle capacity will be attractive.
I'm just wondering how firm Intel's insistence that while it *has* an ARM license, it doesn't intend to *use* it will remain. Intel *used* to make ARM CPUs via its former StrongARM division. If someone like Apple inquired about Intel making ARM CPUs at 14nm geometries for them, what do you suppose Intel might say? I can see the design side of the house turning colors and sputtering, and the manufacturing side calling the janitors to mop up the drool.
On the other hand, Intel has been one of the most reliable companies at consistently hitting their process milestones on time.
Then you add in the huge potential performance/cost advantage of a true 14nm node.
Those two things together likely equal a lot less risk to forward profits than counting on TSMC.
2) With Apple controlling the end product space, it makes sense to target them as a customer. Apple is a big customer to Intel already and while their processors could be considered competitive, they are not a competitor to Intel. That is not true at all of Qualcom who is definitely a competitor. Partnering with Qualcom would mean Intel abandoning a large future market space. That is just not going to happen.
3) Compared to overall Intel revenue, this portion of the market could be small, even at a 100 million+ units/year. A $10 processor at 100 million units is still only $1 billion. Yes a nice adder to the bottom line, but ultimately not huge dollars for a company of Intel's size. The question is whether yield and die size would offer enough cost savings to give them sufficient margin to justify the business. Also need to consider what portion of the power budget is processor and if the impact of Intel allows a premium charge.
Intel has an architectural license to ARM though they have stated in the past they will not use it.
The question for Intel sales and marketing is is this a good back door method for getting into those companies to sell their architecture?
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.