Neither Globalfoundries nor Samsung would directly address expectations that the process will carry a higher cost per transistor that the prior node. Such costs are anticipated given the challenges of FinFETs and the need for double patterning lithography on at least some layers of designs at 20 nm and beyond.
"There's a big debate about the increasing cost of transistors, but the good news is there's a huge demand for FinFETs," says Kelvin Low, a senior director for foundry marketing at Samsung. "We'll have a better picture on the costs soon."
The two companies says their collaboration was motivated, at least in part, by common customers that wanted to second-source foundry services and need the latest technology. That could be a reference to Apple or Qualcomm, the two leading mobile application processor designers. Apple has long been rumored to be seeking an alternative to Samsung, given the rivalries between the two companies. Qualcomm could not get enough 28 nm wafers when TSMC first released the process.
GlobalFoundries and Samsung see more potential for developing a strong ecosystem between them than they had under Common Platform deals at 65, 45 and 28 nm. "Previously... we had diverse customers wanting to do different things, but now we have joint customers," says Ana Hunter, vice president for product management at GlobalFoundries.
Her company hasn't stopped development on its own 14XM process, she says, but it has not decided whether to complete it and bring it to market.
IBM's absence from the deal is the clearest sign to date that Big Blue is ending its chase of leading-edge process technology. Losing to AMD in the latest round of ASIC design deals for the Microsoft Xbox, Sony Playstation, and Nintendo Wii was a huge factor in its ability to fund work on the next generation of chips.
"We continue to work with IBM on other things, such as our collaboration in Albany at 10 nm and beyond and other early process research and materials research," says Hunter.
Both foundry execs agreed that the 20 nm planar node will have a short life.
"We are prototyping products in it now, but we don't expect it to be a big node," Hunter says. "We have customers at the leading edge designing at 20 nm, but we expect they won't remain very long and start moving to FinFETs soon."
Jim McGregor, principal of the market watcher Tirias Research, says most customers "are likely to move directly from 28 nm to 14 nm, which is a significant jump" of about 40% in performance and 55% in power. "I think we are likely to see the mobile SoC vendors jump first, and then the PC/server and graphics."
— Rick Merritt, Silicon Valley Bureau Chief, EE Times