The companies that supply Americans with ever-newer versions of cell phones, iPODS and other technological marvels, says David Wu, deal constantly with risk and uncertainty.
It can take two years of lead time and cost $2 billion, says Wu, just to build and equip a fabrication facility (fab) and to train the work force that will produce the wafers of silicon from which computer chips and microprocessors are made.
All this for products that may be obsolete in less than two years and for a fab that may need to be retooled or reconfigured again.
Wu, the Iacocca Professor and chair of industrial and systems engineering, and Rosemary Berger, an assistant professor in the department, have received a $400,000 grant to help the semiconductor industry navigate uncertainty.
The three-year grant from the National Science Foundation (NSF) and Semiconductor Research Corporation (SRC) will enable the researchers to study supply chain planning and optimization for the industry. SRC is a university research management consortium based in North Carolina.
Wu says the project will focus on two areas: demand forecasting for high-tech products with short life cycles, and supply-demand planning for companies that make those products.
Most existing methods of forecasting consumer demand, says Wu, are designed for autos, TVs, commercial electronics and other products whose cycle of demand is, if not stable, at least predictable.
But forecasting the demand for high-tech products that are constantly being improved, he says, is almost like predicting the fall fashions.
"High-tech products go through an initial period of demand growth, reach maturity and then decline - all this in 18 months in most cases," he says.
In addition to the problem of short shelf life, says Wu, the wafers used in high-tech products are often not interchangeable, but differ according to the functionalities that they hold.
Companies thus pay a penalty, says Wu, if they invest hundreds of millions of dollars in new fab infrastructure and employee training only to discover that there is no demand for the product they introduce.
"The consequences of building a fab for the wrong kind of wafer or mix of wafer technologies can be very severe," he says.
Wu and Berger hope to forecast demand for high-tech products by identifying a subset of high-tech products that demonstrate a demand pattern several months ahead of a larger group of products.
"These 'leading indicators' are anywhere from three to eight months ahead of the curve," says Wu. "We have developed a set of statistical techniques to identify these indicators and track their demand pattern."
Wu and Berger then propose to use a stochastic optimization model to determine the most effective way of aligning supply with demand. Such a tool - a stochastic, or probabilistic model - would enable decision makers to take demand uncertainty into account before building their capacities.
"We can reduce prediction errors," says Wu, "but we can't eliminate them completely. That's why these stochastic optimization models are useful.
"We have developed mathematical and computer models to capture this type of problem and the algorithm [set of procedures] to solve the models. We have already developed a prototype. Over the next three years, we hope to improve our prototype so that it can handle a much wider variety of demand types."
Wu is co-director of Lehigh's new Center for Value Chain Research, which recently conducted a two-year project for Agere, developing an analytical process that optimizes new product development portfolios. Researchers developed an advanced stochastic optimization model that optimizes product portfolio while considering the risks and uncertainties due to market demands, business partnerships, and technology lifecycles.
The CVCR, an NSF Industry/University Collaborative Research Center (IUCRC), is part of an NSF consortium in engineering logistics known as CELDi.
Posted on Friday, October 01, 2004