SAN JOSE, Calif. — Taiwan Semiconductor Manufacturing Company (TSMC) announced plans to roll out a compact, low-power version of its 16nm FinFET process and revealed its road map for smaller process nodes. The company will begin volume production of its 16nm FinFET Plus (16FF+) in the middle of this year and break ground on a new 10nm fab next year.
A year after volume production of 20nm chips, TSMC announced it will begin volume production of its 16FF+ in the middle of 2015. TSMC claims the chips made using FinFET Plus have 10% better performance than competing silicon, consume 50% less power than a 20nm SoC, and have a cycle time twice that of 20nm chips.
The foundry will have more than 50 tape-outs by year’s end, covering applications processors, GPUs, automotive, and network processors, said TSMC President and Co-CEO Mark Liu said at the TSMC 2015 Technology Symposium Tuesday.
“We are in a critical juncture — today we not only have to drive the growth of our respective companies, but drive the search for new companies that didn’t exist before,” said Liu. “Our consumer product cycle hasn’t changed much. What has changed is the cadence of product design and tech development. More and more work must be accomplished in the same time frame.”
Liu added that TSMC has collaborated with ARM on a Cortex-A72 that leverages the 16FF+ to achieve 3.5x better performance than a Cortex A-15 while consuming 75% less power. TSMC and ARM will continue to collaborate on next generation process nodes, he added.
The company also developed a compact version of its 16nm FinFET process, called 16FFC, for mid-to-low-end smartphones, consumer products, and wearables. The process reduces power consumption more than 50% to 0.55 volts and product tapeouts are expected in the second half of 2016.
“In terms of the 16FF and 16FF+, there have been some significant challenges with cost and we projected that cost per gate would go up,” International Business Strategies CEO Handel Jones told EE Times. “I think they acknowledged it now with the 16FFC. The 16FFC will have some good traction especially when they have the low-power version.”
TSMC has long been engaged in a 16/14nm FinFET node battle with Samsung — TSMC’s 16nm is similar to other 14nm offerings — and the Korean chip giant announced that it will have a 14nm chip in its Galaxy S6 at Mobile World Congress. While TSMC officials declined to comment on competition, the victor will be determined by chip volume.
“Samsung claimed they were in production now, but we haven’t seen anything,” Jones said, pointing to the Exynos chip. “If Samsung does produce high volumes at that point they’re ahead of TSMC.”
Taking it to 10
While TSMC’s 16nm technology is due this summer, the company announced a road map for its much-discussed 10nm process. The 10nm process will have 2.1 times the logic density of the 16nm node along with a 20% speed gain and 40% power reduction. The company demonstrated a 256 Mbytes SRAM made in the process. TSMC expects its 10nm to be in production by the end of 2016 and alluded to more than ten partnerships in various stages of design.
“We think 10nm will be the long-lasting technology node and for TSMC to be accelerating 10nm, I think that is a very good sign for the industry,” Jones said. “With the acceleration of 10nm — which they might end up going 8nm — TSMC will close the gap with Intel. I think TSMC is on a roll.”
TSMC has invested between $11.5 billion and $12 billion in its 16nm and 10nm technology — which means they have to have customers in place, Jones noted. Strengthening its commitment to 10nm, the company will break ground on a new fab with 10nm tools in the second quarter of 2016 and will move 10nm tools into an existing fab this quarter.
Intel will be TSMC’s major competitor in this space, with plans to ramp production of 10nm in the next 12-18 months. The issue may not be TSMC’s timing with Intel, but Intel’s strength of design and its manufacturing problems.
“It’s not completely clear who is ahead at 16/14 but I think TSMC is making a major commitment to trying to be ahead at 10,” Jones said. “If that happens and TSMC has closed the gap with Intel, the issue is then if TSMC’s 10 and Intel’s 10 are the same,” he said.
— Jessica Lipsky, Associate Editor, EE Times