TSMC Rolls finFET Design Kit Amid 28nm Ramp
By Mark LaPedus
At the TSMC 2012 Technology Symposium, Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) began to set the stage for the finFET era by announcing the industry’s first finFET design kit for foundry customers.
By year’s end, TSMC will provide finFET design kits for the 14nm node. The silicon foundry giant also made several other announcements at the event in San Jose, Calif. on Tuesday (April 17). Among those announcements included a big ramp at the 28nm node, new bipolar-CMOS-DMOS (BCD) processes and an eventual shift to 300mm wafers on the analog/mixed-signal front.
In another move, TSMC also plans to combine technologies and offer one process at the 20nm node. In other words, the company appears to be pushing customers towards finFETs sooner than later.
At the event, Morris Chang, chairman and chief executive of TSMC, also gave a rather somber outlook for the overall semiconductor industry, saying that the IC market would grow by a mere 2-to-3 percent in 2012 over 2011. During a keynote at the event, Chang predicted much higher growth — about 10 percent — for the fabless industry. He added that TSMC is also seeing strong growth from customers in the mobile business.
In an interview with SemiMD after the presentation, Chang said he also sees a new and competitive landscape in the foundry world, especially from one vendor. “Samsung is coming on strong,” he said. At the leading-edge, “GlobalFoundries and UMC are there.”
When asked about Intel Corp.’s recent efforts in the foundry business, Chang said: “Intel is dabbling in the foundry business. They are our indirect competitor.”
By indirect competitor, Chang said that Intel itself competes against TSMC’s customers, such as AMD, Nvidia, Qualcomm and others. During his keynote, Chang was quick to point out that TSMC does not compete against its customers — unlike the IDM foundries like Intel and Samsung Electronics Co. Ltd.
28nm boom
At present, GlobalFoundries, Samsung, TSMC, and to a lesser degree, UMC, are ramping up their respective 28nm processes. Chang dismissed reports that TSMC is struggling with its yields. “There are some rumors that we are struggling with our yields,” he said during the keynote. “We have had some difficulties. But the difficulties are that we don’t have enough capacity. The yields have been good. The yields have been at or above plan.”
Dean Freeman, an analyst with Gartner Inc., agreed with Chang’s assessment. “Everything we’ve heard is that it’s not a yield issue” at TSMC, Freeman said. “It’s a product mix issue.”
To boost its capacity, TSMC is ramping up its 28nm process in several fabs and is also looking to raise its capital spending. Earlier this year, the company planned to spend about $6 billion in capex for 2012, compared to about $7 billion in 2011. Chang said TSMC will exceed the $7 billion figure.
In any case, 28nm is becoming a smash hit in the market. “It looks like 28nm will be a long-lived node,” Freeman said. “If you look at what Nvidia and some others are saying about the costs for the 20nm node, 28nm is going to have a long life.”
The 20nm node will introduce double-patterning to the mix, which could add cost and complexity in the chip-manufacturing flow. For this reason and others, “28nm will become an enormous node,” agreed Walden Rhines, chairman and CEO of Mentor Graphics Corp., during a recent interview.
In fact, some 18.6 percent of the world’s silicon capacity revolves around the 32nm/28nm node right now, but the revenues do not reflect this percentage, Rhines said. This implies there are still yield issues at the node, meaning there is a major upside when the foundries resolve the problems, he said.
Another foundry vendor, GlobalFoundries Inc., has seen a major improvement in its yields at 32nm/28nm. This is especially true at one of its key customers: Advanced Micro Devices Inc. (AMD). “We’ve met or exceeded their expectations” in terms of yields, said Michael Noonen, senior vice president of worldwide sales and marketing at GlobalFoundries, in a recent interview.
20nm vs. finFETs
Following the 28nm node, TSMC will move into risk production for 20nm by year’s end, as previously announced. Both the 28nm and 20nm processes are based on planar transistors. While 28nm could be a long-lived node, 20nm may see the opposite effect — at least at TSMC. In the past, the company offered multiple processes at various nodes.At 20nm, however, TSMC plans to combine technologies and offer one process at 20nm, said Shang-yi Chiang, TSMC’s executive vice president and co-chief operating officer.
Perhaps 20nm is an important node for TSMC. On the other hand, TSMC appears to be sending a signal to customers: There are fewer process choices — and less of an incentive — to migrate to 20nm. After 20nm, TSMC plans to roll out finFETs at the 14nm node by the 2014 or 2015 time frame. TSMC disclosed that its finFET process roadmap is not set in stone, saying it is mulling plans to re-adjust the target date for the technology. Nvidia and others want finFETs sooner than later.
In the fourth quarter of 2012, TSMC will provide the foundry industry’s first finFET design kits, said Cliff Hou, vice president of research and development at the company, during a separate presentation. The kit will include the key components to get the ball rolling in the finFET era, such as a design rule checker (DRC), layout versus schematic (LVS), process design kits (PDKs) and other tools, he told SemiMD.
For finFETs, TSMC plans to use bulk technology — and not silicon-on-insulator (SOI). One SOI wafer vendor, Soitec, believes the foundries will re-think their bulk finFET strategies once they assess the advantages of SOI. “We are still in the early R&D process” in the finFET era, said Steve Longoria, senior vice president of strategic business development at Soitec. With process simplicity offsetting the approximately $500 cost (less in high volumes) of the SOI wafer, semiconductor companies can actually save 3-10 percent compared with bulk technologies, Longoria said.
Meanwhile, there were other announcements at the TSMC event. The company rolled out new versions of its 180nm and 130nm BCD processes for analog and mixed-signal designs. Right now, TSMC provides mixed-signal and analog foundry services within its 200mm fabs. Over the next 18 to 24 months, the company plans to provide these services on 300mm substrates, said George Liu, director of the Power/MEMS Business Development division at TSMC.
The move is in response to Texas Instruments Inc., which has been ramping up the world’s first and only 300mm analog fab. TI believes it has a competitive advantage over its rivals in terms of cost within its 300mm analog fab, dubbed RFAB, which is located in Richardson, Texas. Many of TSMC’s analog/mixed-signal customers compete with TI and are looking for 300mm capacity to drive down costs.
