Can Intel Dethrone The Foundry Giants?

For some customers, Intel may be an interesting partner. For others, foundries still have some distinct advantages.

popularity

The leading-edge foundry business isn’t for the faint of heart. It requires deep pockets and sound technology to keep pace in the chip-scaling race. And despite pouring billions of dollars into new fabs and processes, foundries are competing for fewer customers at each node.

Given the difficult business conditions, only a handful of vendors can afford to compete in the high-end foundry business. Going forward, though, the business is expected to get even more competitive—and intriguing—as Intel expands its efforts in the foundry arena.

In 2010, Intel officially entered the foundry business, but the company has only offered its service to a small group of non-competitive customers. But in a surprise move, Intel now plans to offer its foundry service to a broader set of customers, including competitors that sell ARM-based chips. “We will open the foundry to any company capable of utilizing our leading-edge silicon,” said Brian Krzanich, chief executive at Intel, at a recent event.

But can Intel dethrone the traditional foundry giants? With its move, Intel will certainly give the industry another option for leading-edge foundry processes, thereby creating more competition, and possibly better prices, in the market. It also will put Intel in more direct competition against GlobalFoundries, Samsung and TSMC. To a lesser degree, IBM and UMC also compete in the sector.

Clearly, Intel has an edge over its rivals. In finFETs, the company claims to have a three- to four-year lead over the rest of the field. And at 14nm, Intel will take the lead over the other foundries in interconnect density, which is critical in area scaling.

On the other hand, Intel’s foundry service will appeal only to a limited customer base, mostly at the leading edge. Today, there are obvious gaps in Intel’s foundry portfolio, such as analog and mixed-signal processes, not to mention third-party intellectual-property (IP).

All told, it will take time before Intel makes a sizable dent in the market. “Intel’s foundry revenues will remain very low or negligible single-digit figures over the next few years,” said Samuel Wang, an analyst at Gartner. “Gaining market share in the foundry business is not determined by better process technology alone. It must also be matched with the right service mentality. Intel is starting to learn this, but it won’t happen overnight.”

Still, the other foundries are not taking Intel lightly. “I never, never underestimate anybody,” said Morris Chang, chairman of Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC), at a recent event. “I would not underestimate Intel.”

Searching for new markets
Intel is scrambling to look for new markets such as the foundry business, and for good reason. Its core PC market continues to decline amid soaring demand for mobile products. In 2013, the traditional PC market is forecast to fall by 11.2%, according to Gartner.

Meanwhile, the foundry business is growing faster than the overall IC industry. The IC market is expected to increase by 5.2% in 2013, followed by 5.6% growth in 2014, according to Gartner. In comparison, the foundry business will increase 13.4% in 2013, followed by 8% growth in 2014, according to the firm.

Overall, the foundry market will be a mixed bag in 2014. “Fab utilization rates (for foundries) will drop by 10% in the next few months, but it will return back to the mid-80% range by mid-2014,” Gartner’s Wang said. “Blended average selling prices (ASPs) for wafers will continue to go up in 2014, just like what happened in 2012 and 2013. This is due to more mixed wafer shipments on advanced nodes, which carry higher ASPs.”

In 2014, demand will be brisk for leading-edge foundry processes, such as 28nm high-k, 20nm, and 14nm finFETs. “The ramp in 20nm will be pretty significant in 2014 and there is a big focus for all of the foundries to move to finFET technology as quickly as possible,” said Gary Dickerson, president and chief executive of Applied Materials, in a recent conference call. “The next battleground for foundries is finFET and customers are pushing hard to begin pilot production in 2014.”

The migration from planar to finFETs is expensive. At the 14nm/22nm nodes, process development costs are about $1.3 billion and a 300mm fab is $7 billion. “It will require big bucks,” said Ajit Manocha, chief executive of GlobalFoundries. “Only a few companies can really afford these expensive fabs.”

Changes in foundry landscape
Up until 90nm there were about a dozen foundry vendors at the high end. But as manufacturing costs increased, the number of leading-edge foundry vendors has dwindled and now consists of IBM, GlobalFoundries, Samsung, TSMC and UMC.

Now, the question is how will Intel impact the foundry market. In fact, Intel has already altered the landscape. In 2011, Intel moved from planar devices at 32nm to finFETs at 22nm. Compared to 32nm planar, Intel’s finFETs provide a 37% performance increase at lower voltages.

Intel also entered the foundry business about that time, by announcing a deal with FPGA supplier Achronix. Under the terms, Intel would make Achronix’ FPGAs, based on Intel’s 22nm finFET technology. Intel also announced similar foundry deals with Microsemi, Netronome and Tabula.

Also about that time, the other foundries were just ramping up their 28nm planar technologies, with 20nm planar still in R&D. But after realizing that Intel was invading the foundry business, the other foundries all came to the same conclusion: They had to accelerate their finFET efforts to close the gap with Intel.

So, GlobalFoundries, Samsung, TSMC and UMC separately decided to take the same approach. Basically, the foundries are in the process of combining a 14nm-class fin technology at the front-end with an existing 20nm planar interconnect scheme. The foundries could devise a 14nm interconnect technology, but this approach is more expensive and would take much longer to implement.

By taking the hybrid approach, the traditional foundries have accelerated their finFET roadmaps by a year. They are now expected to move into finFET volume production in 2014 and 2015. “There are industry and business pressures for the foundry world to move to finFETs,” said Subramani Kengeri, vice president of advanced technology architecture at GlobalFoundries. “But that doesn’t mean we are going to gamble and try to do something risky. And so, we said: ‘Let’s advance finFET first on the existing platform on the front end. Leave the middle-of-the-line and backend exactly the way they are, because there is a huge amount of learning there.’ In other words, we will minimize the risk going from planar to finFETs. This is very compelling to customers.”

Turning the tables
Meanwhile, Intel surprised the industry in early 2013, when the company announced its initial first-tier foundry customer—Altera. Until then, Altera’s sole foundry partner was TSMC. But apparently Altera wanted finFETs sooner than TSMC could deliver, so it turned to Intel as part of a major deal. Under the terms, Altera will gain access to Intel’s 14nm and 10nm finFET processes, giving the FPGA vendor a lead in technology over rival Xilinx. Meanwhile, for finFETs, Xilinx is sticking with TSMC.

Intel has high hopes for its 14nm process. For years, the foundries have been ahead of Intel in interconnect technology. Intel’s 14nm process not only represents its second-generation finFET technology, but it also gives the company the lead in terms of interconnects. Unlike its rivals, Intel’s 14nm finFET process combines a 14nm-class fin and a true 14nm interconnect, which impacts area density. Compared to its rivals, Intel will have a 35% advantage in area density at 14nm and a 45% edge at 10nm, said William Holt, executive vice president and general manager of the Technology and Manufacturing Group at Intel. “(This means) the die size is smaller,” Holt said.

Intel, however, recently delayed its 14nm ramp until the first quarter of 2014, which is a quarter later than expected. There were defect issues with the 14nm process, which have been addressed, he said.

Still, Intel is ahead in process technology. And looking to expand its foundry customer base, the company is now willing to engage with companies that use competitive ARM-based chips. But can Intel translate its manufacturing expertise into a broader foundry service offering?

To be sure, there are some major differences between being a microprocessor supplier and a broad-based foundry provider. Price, delivery and flexibility are key. Dealing with multiple customers with different needs can be a headache. IP protection and access to fab capacity are paramount. And even if Intel provides a sound foundry offering, it can’t overcome one obstacle. It competes against many potential and large foundry customers, such as Apple, Broadcom, Nvidia and Qualcomm, according to Nathan Brookwood, a research fellow at Insight 64.

Simply put, Intel will need to adopt a new mindset if it wants to be a bigger foundry player. “The thing that will need to happen is that you will need to see Intel be a bit more flexible in process and design,” said Dean Freeman, an analyst with Gartner. “(Intel’s) copy exact fab strategy, for example, is fairly rigid. It is difficult to say if copy exact will work well in a foundry environment.”

Intel also will need to become more open and bolster its third-party IP and design flows for customers. “This may mean opening up the kimono, as you will likely need to have Mentor, Synopsys, Cadence and others in the mix as design partners,” Freeman said.

All told, Intel will appeal to some but not all foundry customers. “In the world of foundries, you have some flexibility in how you design your transistors and thus your device. With Intel, you possibly have a bit more inflexibility, as you probably are going to be funneled into an x86 type of design,” he said. “For some companies, this will work out great, and for others it may not be a match. The upside is if you are a fit, you get some of the fastest transistor technology on the planet.”



  • Mark LaPedus

    Hi Ian. Check Inttel’s recent analyst meeting. Bill Holt goes into detail about it…..

  • Dusty

    It seems that the Intel 14nm node will have a Metal1 pitch of 64nm and a Metal2 pitch of 56nm. (The Intel 22nm node has a Metal1 pitch of 90nm and a Metal2 pitch of 80nm.)

    What is the density improvement of having a Metal2 pitch that is smaller than the Metal1 pitch?

  • Pingback: Semiconductor Engineering .:. Looking Beyond Moore’s Law()