Selection, Analysis and Development of Greenfield Sites for 300 mm Fabs
Daren Dance, David Lauben, WWK, Pleasanton, Calif. Danny Lam, FHI Research.com Inc., Wilmington, Del. -- Semiconductor International, 11/1/2000
 |
At a Glance | | |
|
| ||
| |||
One estimate of the demand for new fab sites, based on forecasted semiconductor industry capital spending,1 is illustrated in Figure 2. During previous upturns, this level of growth and capacity utilization would have resulted in frenzied activity by companies seeking to find new fab sites and adding new capacity that would be "on stream" by 2003 or 2004 to meet anticipated demand. However, most companies have not done so during this upturn.
Companies cut back severely on site analysis, development and selection activities as a result of the severe, double-dip slump between 1996 and 1998. At present, only one semiconductor company in the world can be said to have preserved its global site analysis, selection and development capabilities at pre-downturn levels, and have an "inventory" of viable sites for 200 mm fabs. Hence, capacity needs during this upturn, rather than being met by solid, well-funded, long-term site development programs, have been created by ad hoc measures. These include integrated device manufacturers (IDMs) using outsourcing strategies, purchasing unneeded fab shells from other vendors and filling them with equipment, and heavily investing in technological upgrading of existing fabs rather than building greenfield capacity. These were practically unthinkable in previous years but a necessary measure for the industry to conserve cash.
While ad hoc measures have been able to fill the need for 200 mm fab sites, no company has invested in developing good 300 mm greenfield fab sites. In all likelihood, when the industry finally determines that they have hit a capacity wall that can only be overcome by the addition of greenfield sites, sites for 300 mm fabs will be chosen not because the site is good, but because of expediency. Many of these sites will be sub-optimal both in terms of profitability and risks. The purpose of this article is to look at some of the technological and market changes that have occurred since the last upturn; to look at how these changes might impact the analysis, selection and development of greenfield sites; and how we have shortened the cycle time needed to bring a candidate greenfield site on stream using a set of software tools and techniques developed during the past three years.
Lessons from the last upturn
1. Capacity utilization rose from a low of
81.6% during 3Q98 to 95.5% in 2Q00. (Source: SIA, IC
Insights)
The upturn that ended in 1995 was a defining moment in the semiconductor industry. For the first time, the average cost of a new fab surpassed $1 billion, causing many sites that were marginally qualified in previous competitions to become decidedly unqualified. At these investment levels, the sheer cost and risks of failure, or normal teething problems encountered in any major construction project, become prohibitively expensive. From a company perspective, the lesson learned is that the cost of bringing an unqualified site in a remote location rapidly eliminates the benefits from even lavish government incentives. What appeared to be extremely good financial incentives — as much as 50 cents on the capital dollar invested — did not make up for the costs or risks involved.
At the same time, governments and communities have also learned bitter lessons that even the best companies can be severely impacted by an industry downturn. For a democratic government that is accountable to an electorate, problems with fab programs that put hundreds of millions of tax dollars at risk can rapidly turn an apparently successful industry recruitment program into a severe political liability. Accordingly, both industry and governments have learned that semiconductor programs, even if they are well conceived and executed, involve risks on both sides. It is not at all certain that rewards will follow just because one or the other party is willing to invest a few billion dollars in a program.
Despite these caveats, there are indeed successful programs. Singapore and Dresden are two examples of locations that have become successful in this industry. Assuming that there are two capable and competent partners, the single common factor that makes semiconductor industry programs successful involves a long-term commitment by the industry and government partners to make the program work over the long haul. This commitment must be supported by industry (not just one company), all levels of a candidate site's government, and the local business community. Once made, this commitment is firm, unwavering and sufficiently strong to weather not only short-term problems, but also such catastrophic events as a double-dip slump as experienced during the last industry downturn.
|
|
| 2. The demand for new fab sites, based on forecasted semiconductor industry capital spending. Note: Capital spending is likely to fall lower than this forecast in 2003 when the industry is expected to hit a slump. |
New technology, new needs
As the industry transitions to 300 mm, the cost of a new fab built between 2002 and 2003 will likely rise to $2 billion. As costs rise and technical specifications tighten, industry needs are rapidly fragmenting, with the needs of a 200 mm trailing edge foundry being drastically different from those of a 300 mm state-of-the-art facility that manufactures a high-value-added product like DSPs.
The high concentration of capital in 300 mm facilities leads to much more stringent requirements for business interruption risks, vibration, and pool of labor quality. Company requirements are also fragmenting. The needs of a flash memory facility drastically differ from those of a DSP or microprocessor manufacturing facility.
Candidate sites have also changed greatly. Some greenfield fab sites in the Pacific Rim that were formerly strong candidates have yet to recover from the Asian crisis and are unlikely to be serious contenders in the near future. Eastern Germany, in the meantime, has greatly improved its facilities and infrastructure. Should the euro remain at its present depressed levels, it can potentially make Europe a preferred location for new fab sites. China, the perennial promising candidate, has two solid facilities — Tianjian (Motorola) and Shanghai (NEC) — that are to begin or have begun production. The proposed pure-play foundry joint ventures in China have all the indications of being winners.
On the other hand, the booming economy in the United States means that there are few locations where semiconductor manufacturing — a mid-margin business — is not in direct competition with higher-margin and higher-profit activities like telecommunications, Internet companies and other hot businesses. While these higher costs may not concern semiconductor facilities that are geared toward making higher-margin products, it is a serious issue for commodity products such as memories and low-end foundry services.
However, the apparent cost penalties of a U.S. site can be mitigated when there are compelling business reasons for a facility based in the United States. Case in point: If a company has an existing infrastructure already in the United States, this can more profitably jump-start a greenfield facility in that location over a comparable site several thousand miles away. Likewise, the risks in U.S. sites are generally much better known and quantifiable compared with offshore sites, which makes it more likely that a program will meet budget and schedule.
Nevertheless, the inherent problems caused by the saturation of sites in the United States will mean that a majority of greenfield fab sites will, all other things being equal, go to lesser developed regions of Asia and Europe where costs are comparatively lower over the long haul.
Lifecycle profitability
Site analysis and selection, from a company's perspective, should be concerned with only two major issues: profitability and business risks. However, the idea that fab sites can be quantitatively assessed with financial implications and detailed risk profiles is a relatively new concept for the industry.
Prior "state-of-the-art" techniques for assessment of sites involved using a static point-scoring model. Points were scored for each candidate site, identifying potential showstoppers and determining ways to work around them (with costs attached). This method included a crude discounted cash flow analysis for the project that does not quantitatively factor in critical issues such as project risk.
This assessment methodology, while capable of sorting out clearly "good" from clearly "bad" sites, does not fully account for the profitability and risk profiles for each site. For example, this methodology often totally ignores the severe profit penalty incurred when an engineering team is split in two to start up a heavily subsidized greenfield fab in a distant country. Such division of a company's human resources often means that the fab "losing" the crew takes a substantial dive in operational performance until such a time when the greenfield crew can stand on its own. Similarly, it is nearly impossible for a static point-scoring model to predict the substantial differences in facility productivity and ramp-up times at different sites or to make mid-course adjustments as project needs change.
Table 1 is from a previous article that presented a set of tools and techniques for guiding site analysis, selection and development process using Factory Commander as the "engine" for financial assessments.2 This analysis accounted for the unique nature of the product being proposed for a facility, and the profit opportunity that such a product may face in the marketplace.
| Table 1. Facility Implementation Times | |||
| Facility | Firm A | Firm B | Firm C |
| Announcement | May 1996 | August 1995 | May 1995 |
| Groundbreaking | October 1996 | November 1995 | December 1995 |
| Completion | October 1997 | December 1996 | June 1997 |
| Construction | 11 Months | 13 Months | 18 Months |
| First tool install | October 1997 | January 1997 | Not Available |
We began by recognizing that different products have different needs. A facility that manufactures DRAMs, which tend to have sharp spikes of profits followed by long periods of small profits or severe losses, has quantitatively different site and facility needs than a DSP facility with a more consistent profitability profile. Assuming all three facilities in Table 1 have the same cost, Figure 3 illustrates the impact of delay on profitability. The delay is additional time in the construction phase relative to Firm A (best case).
Table 2 shows the break-even results from the Factory Commander cost model. This basic model also allowed us to estimate the profitability impact of a greenfield semiconductor program on a company's productivity at existing fabs. Starting a fab 500 miles away can have a steeply lower cost structure than starting a fab 5000 miles away.
| Table 2. Break-Even Time Frame | |||
| Facility | Firm A | Firm B | Firm C |
| Delay | 0 Months | 2 Months | 7 Months |
| Break-even | 62 Weeks | 69 Weeks | 81 Weeks |
Moreover, we are able to show that substantial productivity differences can be readily explained by site factors. For example, one foundry company found that its facility, located in what appeared to be a lower-cost location 45 minutes across the river from a cluster of fabs, experienced serious productivity problems caused by high rates of employee turnover. Turnover was strongly related to the distance between the fab and the location where the employees wanted to live. These long-term productivity losses eliminated the lower costs of the site. Our simulation readily identified these potential issues.
|
| 3. The impact of delay on profitability, assuming all three facilities in Table 1 have the same cost. |
Summary
Semiconductor manufacturing is inherently a cyclical industry that faces periodic bouts of boom and bust on the way to the average 16% annual growth rate. Normally, it takes a process of years for a company to gradually pre-qualify and build the necessary infrastructure for greenfield fab sites by working with the community and governments involved. However, because of the severe slump in 1996-1998, the industry has cut back investments in this process worldwide even as we are about to hit a capacity "wall" that can only be met by quality 300 mm greenfield sites.
As the cost of 300 mm fabs continues to rise and the risk of failure becomes more and more unacceptable to industry, it is critical that an inventory of 300 mm qualified sites be available. Such sites will only be available if the industry can craft "win-win" propositions that sustain the industry, government and community partners involved when the industry encounters temporary setbacks. At the very heart of crafting "win-win" deals is the need for the partners to know precisely and quantitatively what is at stake for them financially and what risks are involved up front. Knowing what is at stake allows the partners to undertake appropriate programs to ensure the project's financial viability and to mitigate risks.
By using a set of software tools and techniques developed by the authors, we have been able to reduce the amount of time it takes to bring a qualified greenfield site on stream from a multi-year process to something that can be done in a matter of months. After such a site is qualified and selected, our tools allow the parties involved to make mutually beneficial mid-course corrections to keep the program viable over the long term as circumstances change. Tools and techniques, however, cannot make up for a basic ingredient of a successful program: the shared commitment to build a successful and profitable industry by industry, government and the community involved. •
Danny Lam is an analyst at FHI Research Inc. He received his doctorate from Carleton University. He has an MBA from the University of Western Ontario and a bachelor's degree from the University of Waterloo.David Lauben is the senior technology consultant for Wright Williams & Kelly (Albuquerque, N.M.). He has eight years of industry experience, including an engineering position with Sandia National Laboratories as a developer and chief architect for the Factory Cost Model (non-commercial predecessor to Factory Commander). His primary role at WWK is the product manager for the Factory Commander application. He holds an M.S. in mechanical engineering from the University of New Mexico (Albuquerque).
1-505-856-3910
david.lauben@wwk.com
Daren Dance is vice president of technology for Wright Williams & Kelly (Austin, Texas). He is co-chair of the SEMI Metrics Cost-of-Ownership subcommittee and is the yield model and defect budget team leader for the International Technology Roadmap for Semiconductors.
1-512-349-4950
dldatwwk@aol.com
REFERENCES
- Strategic Marketing Associates, in "Economic Indicator," Semiconductor International, Dec. 1999.
- J. Kanz and D. Lam, "Wafer Fab Profit Opportunities and Costs," Semiconductor International, July 1998.
