Illustration by James Yang

One Piece at a Time

Over the last 25 years, the power of computer chips has quadrupled every three years. With more transistors squeezed onto each new chip, semiconductor manufacturers have been able to reduce the cost of computing power dramatically with each new generation of technology.
       But while the price tag on computing power has declined, the cost of the wafer fabrication equipment used to make semiconductors has soared tenfold about every decade. Today, a state-of-the-art chip factory demands a $1 billion capital investment. There is growing concern among semiconductor makers that the future of the industry will be limited by escalating capital costs.
      Unpredictable demand and productivity--typical of the industry--make investments in new fabrication technology extremely risky. If a wafer fabrication plant, or fab, is too large, chip makers are stuck with expensive overcapacity. If a chip factory is too small to meet a sudden, burgeoning demand, there are significant losses in opportunity costs. Securing the huge loans needed for new fabs soon may be too onerous even for companies with a good track record. "A billion dollars is a huge amount of capital to bet that things are going to work out right," says Samuel Wood, assistant professor of manufacturing and technology.
      One way to cope with this risk is with the development of modular capacity, a concept Wood and Alex Angelus, PhD '97, began studying for a project that was initially funded by the Stanford Integrated Manufacturing Association (SIMA), a cooperative venture between the Graduate School of Business and the School of Engineering. Wood has also worked with Evan Porteus, the Sanwa Bank Professor of Management Science, to show that costs and risks can be reduced if a factory's infrastructure "shell" is built first and smaller modules of production capacity are added as needed later on for expansion. Additional buildings and machinery are added only when new production capacity is required, thereby reducing initial investment outlays as well as financial risk. "What the industry would rather do is start with a modular fab, get it up and running quickly, using profits from the first module to fund additions," says Wood. "Semiconductor manufacturers tell me that it is easier to borrow $100 million than to get $1 billion."
      Wood identifies a special category of fixed costs that are relatively independent of capacity. These include the costs incurred to build a fab of any size, such as architectural planning, an uninterruptible power supply, and the fab manager's salary. He has also evaluated critical decisions in planning a modular fab. These revolve around which resources fit into the shell and which are reserved for the modules, as well as the size of the shell and the size and timing of the modules.
      Although capacity is a small part of the total fixed costs associated with a fab, it has a strong impact on its efficiency and expected profit, which makes modular capacity attractive. Indeed, the 1997 National Technology Roadmap for Semiconductors, an industry consensus published by the Semiconductor Industry Association, supported many of Wood's ideas on the transition to modular capacity.
      Modular capacity has immediate relevance for the semiconductor business, but the idea is also applicable in other areas such as the manufacturing of flat panel displays for laptop computers. Wood, Porteus, and Angelus are continuing different avenues of research in this important field of manufacturing technology.

--By BARBARA BUELL

"Optimal Sizing and Timing of Capacity Expansions with Implications for Modular Semiconductor Wafer Fabs," Alexandar Angelus, Evan L. Porteus, and Samuel C. Wood, GSB Research Paper #1479, February 1998