TI’s Evolving Manufacturing Strategy
By David Lammers
Don’t even talk to Paul Fego, the vice president of worldwide manufacturing at Texas Instruments, about TI being among the ‘fab lite’ semiconductor companies. With 24 front- and back-end facilities in operation, TI’s management team has created a unique mix of new and slightly used to support its fast-growing analog and power management IC businesses.
Fego said TI is “only a few tools away” from expanding production at the 300mm RFAB in Richardson, Texas. The phase-three expansion will increase RFAB production to about 935 wafers per day, which would represent $2 billion in annual revenues. Fego said that leaves TI “positioned very nicely” to expand production of the high-volume analog and embedded processor ICs which now form the “core technologies” at the Dallas-based company.
Fully loaded, RFAB is capable of $5 billion in annual production.
Fego spoke at the ISMI Symposium in Austin, Texas, describing the challenges of managing fabs which TI operates worldwide, supplemented by OSATs (outsourced assembly and test providers) and wafer foundries. About 95 percent of TI’s analog wafer fabrication is in-house, compared with roughly 75 percent of its digital IC production. Partly because TI expanded its assembly and test operation at Clark, Philippines at an opportune time (just as some OSAT’s were shutting down capacity), TI gained share against rivals which did not have enough assembly and test capacity.
“Because of Clark we gained market share at a time when lots of outside vendors were shutting down. We were happy to buy assembly and test equipment relatively cheaply,” Fego said. Expanding RFAB with used tools from Qimonda Richmond and TI’s purchase of Spansion’s 300mm fab in Aizu, Japan is another example of Fego’s admonition: “Never let a good downturn pass you by.”
In Chengdu, China, TI now owns a former SMIC/city-owned fab, which sits across the street from a huge new Foxconn manufacturing site which Fego said eventually will employ 100,000 people.
During Fego’s ISMI keynote, he praised the manufacturing engineers at National Semiconductor, now part of Texas Instruments. “National has three factories, and they have people with 30 to 40 years of experience building analog parts. We are very fortunate, because they are showing us ways to do things that we hadn’t thought of at TI.”
“Analog is not the sexy part of the industry. But when you think about the fact that about a billion motors are sold every year (with chips worth about three dollars per motor), that is a three billion dollar market. We will take that,” Fego said.
The total analog market is about $42 billion per year (TI claims 17 percent of the analog TAM) and the embedded processing market is about $18 billion, providing TI with new opportunities after the company announced it would withdraw from the cellular baseband sector.
“Moore’s Law does not apply directly to analog,” he said, but customers do require specialized analog processes, manufacturing controls, and plenty of capacity when demand ramps up.
Fego said TI maintains a centralized database which connects the 24 TI facilities, six foundries, and 70 assembly and test subcontractors that TI does business with.
“We have 24,000 people, and we try to identify the people with high potential and teach them how to fix problems faster,” he said. The company has an in-house tool which describes manufacturing problems, how they were resolved, and which TI employees can serve as mentors to others trying to solve the same problems.
“Our big challenge is how to get our people to get humble, to keep their individual egos in check, so that they can work in collaboration. We tell everyone that there is someone in the TI organization who knows how to solve your particular problem,” Fego said. If a person fails to consult the database, and pick up the phone to consult with someone in the manufacturing organization, it reflects on the person’s work record, he added.
“If there are language problems, we will even fly someone from, say, Japan to Dallas or vice versa so they can have a face-to-face meeting if that is what it takes,” Fego said.
The in-house tool takes a good bit of work to maintain. Employees must describe the issue, who fixed it, and how. “We have 24,000 people who can tap into that database, and we charge the fab managers with figuring out if they need to put someone on a plane to solve a problem or get people talking on the phone,” Fego said.
The expansions in Texas, Japan, and China are the latest phase in TI’s ever-evolving manufacturing strategy. Eleven years ago, in 1999, the company began working with foundries. As CMOS technology moved from 130 to 90 to 65 nanometers, TI executives became “more comfortable with what the foundries had,” said Kevin Ritchie, TI’s senior vice president of technology and manufacturing.
In an interview earlier this year at TI’s Dallas headquarters, Ritchie said digital CMOS “was driving an incredible amount of capex spending, and technology was churning very fast.” Meanwhile, the cyclical cellphone industry prompted concerns by TI’s management about “leaving a lot of stranded capacity behind” if it spent heavily on internal digital IC capacity. (With Nokia’s subsequent loss of market share, that proved to be a prescient concern.)
The company had a brief debate in the first half of 2006 about doing the front end of the line in-house, while farming out the more-generic metallization steps. Then TI’s management made a major decision – announced in January 2007 — to focus its internal digital CMOS resources on designing for low power and performance, putting CMOS technology development largely in the hands of its foundry partners.
“We came to realize that we weren’t going to get ahead in design rules, or our ability to pattern, or in getting the tools ahead of anyone else. And it took a lot of dollars to change to the next-generation technology,” Ritchie said.
Today, foundries account for 25% of TI’s digital CMOS production. Of that, about 60% is “advanced” digital CMOS wafers, a percentage that TI expects to increase over time. For analog, 90-95% is built internally, a ratio expected to remain steady. Foundries will continue to supplement TI’s internal analog capacity, and provide specialized, low-volume production.
During the same 2006-2008 time period, TI’s management watched as analog revenues, including power management devices for mobile systems, began to grow steadily. Though half of TI’s current revenues still come from digital CMOS, including wireless SoCs, the company now emphasizes its strategic directions toward analog and embedded processing.
The shift to CMOS foundry production left the question of what to do with RFAB, in the Dallas suburb of Richardson. The building was finished in May 2006, garnering several awards for its environmentally friendly design. Speculation – never supported by TI itself – arose that the RFAB shell would be sold.
Instead, TI managers began discussing how to use RFAB for analog production. “In the logic space we were playing in, we knew that we had little ability to differentiate. But in analog, the processes create differentiation,” Ritchie said.
As the debate ensued, some proposed filling RFAB with 300mm equipment. Initially management was split 75-25 split towards the 200mm wafer size. “Going to 300mm was more out of the box” than the decision to outsource digital CMOS development to foundries, Ritchie said.
Analog devices often require a 20-micrometer-thick epitaxial silicon layer, which the wafer manufacturers were not supplying at the 300mm diameter. And there were worries about 300mm wafers warping due to the high temperatures (in the range of 1200 ºC degrees) involved in anneal, implant drive-in, and other analog processing steps.
Ritchie said Qimonda’s bankruptcy swung TI towards manufacturing analog devices on a 300mm toolset. “What came together at that time is that the Qimonda 300mm toolset became available, and for pennies on the dollar,” he said.
The Qimonda tools were being used for 65-nm DRAM manufacturing, while TI’s mainstream analog process, called LBC7 (Linear BiCMOS) is at 250-nm design rules. Ritchie said tool suppliers were readily able to disassemble, move, and reassemble the tools without problems. Qimonda didn’t have epi tools, so those were purchased new, as were some furnaces. (Qimonda used all vertical furnaces, but the high-temperature steps in analog make it difficult to avoid wafer warpage in a vertical furnace.)
“We could use all of the 248nm and i-line lithography tools, and all the plasma tools. All told, we averaged about 15, 17 cents on the dollar,” Ritchie said, meaning that the company paid roughly $150,000 for tools that would have cost a million dollars if purchased new. TI calculates a 30-40 percent productivity gain from the shift to 300mm wafers from the 200mm diameter, he said.
TI also upgraded its Freising, Germany and Miho, Japan fabs to 200mm processing, in part by moving equipment from the closed K-fab in Dallas that at one time had been the company’s digital CMOS development center. (Some of the approximately 200 tools acquired from Qimonda also went to Freising and Miho).
To prepare RFAB for production, TI began working with its wafer suppliers to develop the 20-micrometer-thick layer of first epi with the right oxygen content and defect levels. Since then, TI has purchased epi tools and grows the first epi layer in-house, freeing up epi capacity at the wafer manufacturers.
It took roughly six months to develop the 300mm epi wafer supply with TI’s three silicon suppliers. “All of this 300mm epi equipment is available and capable of putting down a thicker film. But it hadn’t been done for that wafer size. Memory needs 1-3 microns of epi, and the wafer suppliers hadn’t gone above that,” he said.
The challenges were largely in process integration, including adapting tools to the double-charge implants, higher processing temperatures, and other unique steps in analog IC production.
Tom Weichel, RFAB manager, said equipping a new fab with used 300mm equipment is “very unique, perhaps an industry first.” Most of the tools acquired from Qimonda were first purchased by the now-defunct German company in 2004, Weichel said, aimed at 70-nm DRAM production. “We had to adapt it to an analog flow, but most of the equipment was in good shape, for the most part. It was not perfect, by any means,” he said.
As TI was equipping the building and adding the gas and chemical delivery systems, it also was building the human resources needed to operate the facility. Again, TI’s timing was fortunate, as several companies (Maxim, National Semiconductor, and STMicroelectronics) were closing smaller fabs in the Dallas area, allowing TI to hire some experienced technicians and engineers. Most RFAB workers transferred in from other TI facilities, while others were hired from nearby colleges and technical schools, providing a mix of new hires and experienced personnel, Weichel said, adding that “finding skilled labor is something every manufacturer has to keep in front of now.”
“We still have the flexibility to grow this facility as necessary,” Weichel said, adding that TI’s policy is to “provide capacity ahead of demand.” Those supply assurances are important to customers large and small, he added.
With the initial RFAB line in production, TI made two other major acquisitions in 2010 to broaden its analog manufacturing base. As flash memory manufacturer Spansion Japan entered into a period of financial restructuring, TI purchased the Spansion Aizu site in northeastern Japan in August 2010. The 200mm Aizu fab is operated by TI today, but some equipment from the non-operating 300mm Aizu fab was moved to RFAB to expand the number of wafer starts in Richardson. Other 300mm tools from Aizu were sold at the time of closing to Taiwan-based foundry UMC.
In the fourth quarter of 2010, TI acquired a 200mm fab in Chengdu, China from foundry SMIC and the Chinese government. There is a second shell at the Chengdu site which could be equipped either as a 200mm or 300mm factory.
Slightly more than 90 percent of the tools at RFAB are refurbished tools. Even the Muratec AMHS system is refurbished, with the track and stockers taken down from the Virginia fab, modified to a different ceiling height, and reinstalled in Texas.
“Moving the tools was the easiest part. Most of the work was in process integration. We found that some tools were too good for what we needed – the etchers were too vertical, for example. And every process had to be reintegrated to 300mm – the job of tuning of processes to get them to match the older 200mm equipment, to get the same parametric performance,” Ritchie said.
“You hear about the big tool purchases in deals like the ones with Qimonda or Spansion, but we are out there every day, buying one here, one there.” Early this year, a former TI manager now working at a large DRAM company told Ritchie’s staff about 10 furnaces the DRAM maker no longer needed after a technology upgrade.
TI snapped them up, and is looking for more. “To buy large blocks of tools, to some extent we have to wait for somebody to go out of business,” Ritchie said. “But we are not going to wait for the next downturn to buy used tools. What could happen is that some of the DRAM or flash guys could consolidate. Then there is just the natural turn of technologies, as processor guys move to 45 and 28, a lot of tools are left behind. Same thing with memory, a lot of tools are left behind.”
TI will rely largely on used tools to fill out RFAB, and to populate the Aizu and Chengdu sites. Taken together, they give TI the potential to add $10B in revenues. However, to cost-effectively manufacture analog ICs, with their relatively relaxed design rules, finding used tools is essential.
“To make 300mm work for analog production, we have to buy used tools. If we have to buy new tools, the math doesn’t work. To pull the trigger on Aizu or Chengdu, we have to be able to buy used tools,” Ritchie said.
Epi is a special case, as TI realizes used epi tools are in short supply. “For the rest of the tools, while we may not be getting them for 15 or 17 cents on the dollar, we expect to buy most of them for less than 25 cents on the dollar,” Ritchie said.