SAN DIEGO, Calif. – Production of next-generation FinFET chips is underway at Samsung and TSMC. Although Samsung announced mass production of its 14nm LPP process technology with a major customer win in Qualcomm, TSMC may have the last laugh with Apple using its process.
In January, Samsung announced its 14nm Low Power Plus (LPP) process with 14% more performance than its LPE process, 0.8V power profile, and a 0.55x smaller die size over 28nm -- 10% smaller than FinFETs from other foundries, Low noted. Qualcomm will build its Snapdragon 820 chips in 14nm LPP, and GlobalFoundries has licensed the process for its fabs.
“This is a great endorsement by a tier one company,” said Samsung Semiconductor's Kelvin Low, senior director of foundry marketing. “Especially many years ago when we started foundry business, there was always a concern about Samsung competing with our customers.”
Low said Samsung has had “overwhelming success” in the mobile and consumer spaces, with future plans to move to network and server segments that require dense logic. Analyst Handel Jones believes the timing of 14nm LPP is on point and having GlobalFoundries as an alternate supplier bodes well, but he isn’t convinced that the technology itself is enough to drive demand.
“Right now Apple is really the key driver for new generations of technology,” Jones, founder and CEO of International Business Strategies, told EE Times. “The number of customers is really very small. Apple is driving it, Qualcomm may be a bit behind, Nvidia is driving it too, but not in big volumes and Xilinx is a tech leader but not with big volumes. If you don’t get Apple and you don’t get Qualcomm, what’s left doesn’t get you a lot of capacity utilization.”
When it comes to Samsung, success will depend on how well its smartphones sell and how innovative they truly are. So far, Jones is leaning toward Apple and TSMC as the more successful next-gen process node supplier.
“We see continued strengthening of the Chinese [in the second half of 2016]. The Chinese companies produced more than 50% of the smartphones shipped last year and the percentage will go up,” he said.
Where TSMC’s plan is easy to follow, Jones said Samsung’s roadmap is a bit up in the air. “They are competing with Qualcomm in terms of modems, Apple in terms of smartphones,” he said. “Their memory business is superb so why not commit to memory?”
Samsung is moving forward with 10nm technology with production expected by the end of the year. Low believes 10nm shipments will be driven by high-end consumer mobile products to start, followed by networking and server players who desire more advancements. TSMC is said to have won a deal to make Apple's next-generation application processors in a 10nm process.
“We would like to make sure we can prove that 14nm is not a single event, that we can repeat this at 10nm,” Low said. “As for long term foundry players in the market – there won’t be many. The [winner will have the] ability to invest not only at 14nm but beyond 14nm, and into capacity.”
Jones said the low cost of 28nm technology and improvements in chip architecture will reduce the need for more advanced process nodes. Apple’s dual-core processors achieve good performance, Jones noted, and performance could be improved by moving to quad- or octa-core SoCs.
TSMC is putting a 16nm fab in China "that probably can last a fairly long time unless you get a significantly lower cost reduction,” he said. “The structure of the mobile market is becoming lower-end and don’t need this advanced technology.”
— Jessica Lipsky, Associate Editor, EE Times