Tough competition ahead from new Japanese wafer fabs*
10/01/2001
*This article was translated for SST from the June 2001 issue of Nikkei Microdevices, our partner in Japan.
Yasuaki Nagahiro, Nikkei Microdevices, Tokyo, Japan
Japanese manufacturing looks like it may be making a comeback in world leadership. After the high-flying 1980s, Japan's competitiveness plunged to where people were questioning if it still made sense to produce semiconductors in Japan at all. Now, Japanese plants are aggressively benchmarking competition and are no longer behind the rest of the world. They're confident they can take the lead with their new fabs, using new manufacturing strategies ranging from making charge-coupled devices and LCDs on the same line, to going entirely to single-wafer processing (see figure).
Overconfidence from their success in the '80s made Japanese chipmakers reluctant to adopt new ways of doing things. When foreign competitors took the lead, the Japanese "just stood by with their arms folded," says Masahiro Nakajima, GM of NEC Kansai's semiconductor division. They kept thinking in terms of DRAM-like high-volume production plant layout. They also clung to open cassettes when the rest of the world moved to FOUPs.
Improvements in efficiency, speed
But that overconfidence is long gone. Aggressive benchmarking has brought in best practices from other companies worldwide, with big improvements, particularly in the once weak areas of plant design and time to ramp-up. Instead of using downflow cleanroom technology with open cassettes, and bays and automated transport everywhere, companies have redesigned plant layouts to suit particular products. NEC Kansai, for example, cuts LSI driver costs with SMIF pods and manual transport. New approaches like Sony Semiconductor Kyushu's miniline, and Trecenti Technologies' scalable production modules can now make small volumes of product without sacrificing efficiency.
The other big improvement has been in how long it takes to get a fab up and running. In the 1990s, Japanese companies took 18 months to two years to build a fab, while rivals in Korea and Taiwan could get up and running in only one year. Despite much more complex regulation to comply with in Japan, NEC Kansai got its new LCD driver line up to mass production in six and a half months this year, in part by starting with an existing building. Sony Semiconductor Kyushu and Trecenti also ramped up new fabs in less than a year by doing multiple construction and installation steps in parallel.
A company can catch up by benchmarking others' best practices, but it needs new technology of its own to take the lead. Sony says it has reduced costs and improved quality by making its high-temperature polysilicon TFT LCDs on the same new 300mm line as CCDs, transferring CCD technology to improve LCD production. Trecenti says using all single-wafer processing at its new 300mm fab cuts turnaround time down to half or a third of that of batch processing.
Matching Asian production costs is a daunting task, with Japan's sky-high infrastructure costs. Japan's weakest points, says Nippon Foundry president Yuko Sakamoto, are "high infrastructure costs, and managment that can't make decisions." He states the problems of the high cost of land, labor, taxes, and electricity, the length of time it takes to get the many permits needed for construction, and the problem of training workers.
In Japan, taxes are 43%. In the US, they're 35%, in Taiwan 25%, in Singapore 31%. Taiwan offers a 5-year tax holiday, while Singapore offers 10. Taiwan and Singapore governments prepare and lease land.
In Japan, some 80 types of approvals are needed to build a plant, and getting them can take a long time. Electricity costs twice as much as in other countries, and labor costs are 1/3 more than in the US, 2x as much as in Italy or Singapore, and 3x as much as in Taiwan.
A new "fab-rich" business model
The conventional model of the IDM worked when Japan dominated the DRAM market, but doesn't work so well without a strong standard product. When much of the world moved on to focus on only one part of the semiconductor business, with the rise of design shops and production foundries, Japanese companies didn't want to give up either their design or production capability.
Now rivals with strong design technology such as IBM and Motorola use a "fab-light" model, stressing design skills more than production, and contract out other production as demand warrants. So unlike the design ventures, they can design chips taking advantage of the most advanced processes. But this model is better suited to companies strongest in design. Japanese producers have traditionally been strongest in production.
That leaves some Japanese companies attracted to the alternative, a "fab-rich" model, capitalizing on leading-edge production more than design. A "fab-rich" company would focus its limited design resources on a particular field to develop cash cow products, but would also take orders to make other products it didn't offer. Where the fab-light company adjusts the amount it contracts out as demand changes, the fab-rich company adjusts what it takes in as it has available capacity. Companies using this strategy would need to match foundries on production technology and costs, and would have to adjust their product mix to eliminate products that compete with their customers. n
Yasuaki Nagahiro is an editor at Nikkei Microdevices, 2-7-6 Hirakawacho, Chiyoda-ku, Tokyo 102-8622, Japan; ph 813/5210-8311.
________________________________
Case Study 1: NEC Kansai makes new logic fab on the cheap
We greatly reduced costs in our new line, which makes drivers and controller-drivers for our LCDs. We cut building costs by 70%, cleanroom construction by 30%, and equipment costs by 30%, enabling us to match the costs of overseas foundries. When we built NEC's second dedicated logic fab, we focused on reducing production costs as much as possible to competitively supply cost-sensitive products. Then we reduced equipment costs by using manual instead of automated transport, buying used equipment where possible, and not ordering any tool options.
Starting with an existing building shell helped shave some 30% off time to ramp, and we did both remodeling and cleanroom installation in parallel. We divided the plant into twelve sections and began cleanroom installation as construction was finished in each one, completing construction in four months. Installation and qualification took another month and a half, for six and a half months in all.
— Masahiro Nakajima, GM, Semiconductor Division, NEC Kansai
________________________________
Case Study 2: Trecenti cuts turnaround time in half with single-wafer processing
Our new, fully automated 300mm wafer fab cuts processing time down from a half to a third what it takes in other foundries by only using single-wafer processes. Costs are lower than for 200mm wafers at our current production volume of 7000 wafers/month, and should be 30% lower once we reach full capacity of 40,000 wafers/month. With our scalable fab design, we can ramp up production to that level in stages as demand warrants. Trecenti started shipping products made on 300mm wafers in April through its parents Hitachi and UMC.
 Trecenti says that by improving its equipment and process technology, it has almost eliminated excess tool capacity, thus reducing equipment costs. |
Trecenti ramped up to mass production in six months by first installing one relatively small volume line in one part of the building. As demand increases, we intend to add more capacity in stages by adding on more such modules, while continuing production in the first ones using FOUPs. To make these small lines running only 7000 wafers/month economical, we worked with our tool suppliers to simplify the tools and eliminate wasted excess capacity. Our "half-and-half" strategy aimed for half the throughput at half the cost. By solving problems like cross contamination, we can use one tool for several processes. This has reduced our excess tool capacity to almost zero (see figure).
— Atsuyoshi Koike, GM, Manufacturing Technology, Trecenti Technologies
________________________________
Case Study 3: Sony 300mm miniline makes both LCDs and CCDs
Sony has started operations in its new 300mm fab, which will make both LCDs and CCDs. It expects four times the productivity and half the cost of competitors.
Both high-temperature polysilicon TFT LCDs and charge-coupled devices are relatively large. The larger area of the 300mm glass or silicon gives immediate major economic advantage, producing 2.5-3x as many units per substrate. Neither uses the most aggressive geometries, and both imaging products can use similar testing technology.
Besides major improvement in efficiency and cost from using 300mm substrates, Sony looks for advantages from using a scalable low-volume miniline and automated metrology. The plant will also aim for zero emissions. The miniline uses smaller equipment and FOUPs, so added capacity can be added as needed without stopping production on the first module.
Applying a decade of CCD production experience, we worked with toolmakers to improve the process and cut the size of the line down to only 10% of the usual LCD line. Transport will eventually be fully automatic, using mostly overhead hoists. Design of the 1/10 size equipment is almost complete. Sony is also applying the automated metrology it developed for CCDs to TFT LCD production. Some tests of LCDs can take days, and problems may not be found until the panel is finished. But automating and networking the process can get results within hours, and earlier in the production process.
— Hirakazu Soma, GM, Imaging Devices New Fab Project, Sony Semiconductor Kyushu