Foundry Taiwan Semiconductor Manufacturing Co. Ltd. has announced that its 16nm FinFET Plus (16FF+) process is now in risk production and that it expects volume production to being around July 2015. That should be about six months behind the original 16FF process, according to earlier statements. TSMC has also quoted seven customers of the 16FF+ process in a press release, presumably hoping to demonstrate that 16FF+ is a safe bet and to encourage yet more customers turn away from the blandishments of the Samsung-Globalfoundries and Intel FinFET offerings at 14nm.
Some are arguing that this acceleration and mass support for 16FF+ means that TSMC is not going to lag Samsung in FinFET process introduction and provides encouragement that TSMC can win the next-generation Apple A9 application processor business. TSMC is currently riding high financially, helped by its win of the A8 and A8X processor contracts using its 20nm planar CMOS. The company announced October monthly sales of NT$80.74 billion (about US$2.63 billion) that were up 56 percent on the same month a year before and a clear sign that TSMC is enjoying booming sales right now.
Others argue that the push on 16FF+ is indicative that TSMC has already lost out on A9 processor manufacturing, which may have gone to Samsung and its 14nm FinFET process. Samsung was the supplier of all application processors for Apple's mobile products until TSMC gained the win in the latest generation.
The argument runs that in effect the 16FF process was a mis-step thathas cost TSMC a vital six months, as a ramp that starts in July 2015 would only yield processors in the market in about November 2015 and that is too late for Apple's expected launch timetable for the iPhone 7.
Apple of their eye
Either way, TSMC is sending signs that the 16FF process ramp at TSMC will be a minor event and that 16FF+ was what customers wanted all along. This is something that TSMC chairman Morris Chang hinted at in July when he said that his company would fall behind its competition in foundry FinFET market share in 2015 (TSMC to Fall Behind Rivals in FinFET Market Share).
The 16FF+ process is an enhanced-transistor version of 16FF that is intended to achieve a higher performance at the same power or lower power at the same performance, at a level similar to that of Intel's 14nm FinFET process.
The announced schedule means that the original 16FF process looks set to have a relatively little uptake. The previously announced timetable said 16FF ICs are likely to come into volume production early in 2015. HiSilicon Technologies Co. Ltd., the chip design arm of telecommunications giant Huawei Technologies Co. Ltd. (Shenzhen, China), has produced an ARM-based 32-core, 64-bit networking processor using the 16FF process, although no disclosure has been made a volume production (China's HiSilicon First on TSMC FinFET Process).
The 16FF+ process uses the same back-end of line (BEOL) design rules but the transistor offers 15 percent performance improvement over the 16FF transistor. Although the metalization design rules are the same the use of the Plus transistor does provide opportunities to reduce standard cell size and therefore reduce chip size, Taiwan has said in the past.
Safety in numbers
The "risk" production comes at the very end of process development and the running of multiple test chips. It is basically the running of full customer designs or partial designs to allow final optimizations of process or optimization of designs to the process and to climb the yield curve to acceptable levels. Because the process is not quite finalized such production is not guaranteed by TSMC and is at the customers' risk.
TSMC said the 16FF+ process is expected to pass full reliability qualification in November 2014 and that nearly 60 customer IC designs are scheduled to tape out during 2015.
When compared with the 20nm planar CMOS (20SoC) that is currently TSMC's leading-edge volume production process,. and the process thought to be used by Apple for the A8 processor, the 16FF+ process operates either 40 percent faster at the same power, or consumes 50 percent less power at the same speed. Notably designs are not likely to show much area saving – and therefore cost saving. Reducing the size of die has previously been a key reason for moving down to smaller geometry.
Rather the 16/14nm FimFET generation is being adopted by companies that care about achieving higher performance and/or lower power consumption, rather than cost savings. The process is therefore targeting high-end mobile computing and networking. The process has achieved 2.3GHz clock frequency for ARM's "big" Cortex-57 processor core while the "little" Cortex-A53 can consume as little as 75mW. TSMC added that design support for 16FF+ includes more than 100 IP cores that have been validated in silicon.
Senior executives from Avago, Freescale, LG Electronics, MediaTek, Nvidia, Renesas and Xilinx were quoted in a TSMC press statement saying that they are using the 16FF+ manufacturing process and that it is a good thing.
"Our successful ramp up in 20SoC has blazed a trail for 16FF and 16FF+, allowing us to rapidly offer a highly competitive technology to achieve maximum value for customers’ products," said TSMC Co-CEO, Mark Liu, in the same statement. "We believe this new process can provide our customers the right balance between performance and cost so they can best meet their design requirements and time-to-market goals."
Related links and articles: