When Xilinx was founded in 1984, the company decided it would not manufacture a new class of logic chips it had invented—field programmable gate arrays, or FPGAs. There were many sound reasons for this decision.

As a pioneer of the "fabless" semiconductor company business model, Xilinx did not want to risk the huge capital investment required to build wafer fabrication facilities. Xilinx also did not want the financial burden associated with operating a fab. Nor did the company like the prospect of owning a fab that would be useless within a few years. Moreover, Xilinx did not want to make chip manufacturing a core competency. Instead, Xilinx wanted to stay focused on designing and marketing its chips, developing software to program them, and providing technical support for the customers.

During the 1980s, the cost of building fabs had skyrocketed, and by 1990, a company could pay from $25 million to $500 million for a fab, depending on delivery requirements and process technology—ten times more than a decade earlier. While some industry leaders back then were claiming that "only real men have fabs," Xilinx and a handful of other semiconductor suppliers quickly proved the viability of the fabless concept. Unlike the vertically integrated semiconductor companies of the time, fabless suppliers generated positive cash flow, achieved higher operating income, and often saw a return on equity in excess of 25 percent.

Several years ago Xilinx co-founder and chairman Bernie Vonderschmidt rightly pointed a number of factors companies should consider when choosing to go the fabless route or not. These guidelines still hold true today:

• Companies that manufacture products requiring only a state-of-the-art fab run the risk of owning an obsolete fab. Back in the 1980s and 1990s, new standard manufacturing processes like CMOS came along about every three or four years. But that has accelerated, and now new processes are going online every 18 months to two years. Xilinx thrives on new process technologies because it allows the company to produce larger, faster, lower power, and more highly integrated devices. This in turn allows Xilinx to play in new markets and grow its revenue.
  
  
• Commodity products such as memory that are sold primarily on cost usually require high fab investments.
  
  
• Owning a fab is cost effective for companies manufacturing a large number of devices or a large number of customer products.
  
  
• Economies of scale are a key determinant in going fabless. In order to establish a long-term relationship with a pure play foundry, a fabless company must have a demand for more than 1000 wafers a month. Any amount less than this will result in excessive wafer costs.

By 1994 there were enough true believers to form an industry trade group—the Fabless Semiconductor Association (FSA). Today the FSA is the leading voice for hundreds of companies associated with the fabless semiconductor industry. Members range from pure fabless semiconductor design companies like Xilinx to integrated device manufacturers (IDMs) like Texas Instruments or Motorola that are moving toward a fabless model. The fabless "ecosystem" also includes essential partners such as pure-play wafer foundries, packaging and test companies, investment bankers, consultants, and intellectual property providers.

The fabless model makes even more sense today. For example, the recession of 2001 was the worst in the history of the semiconductor industry, and many large semiconductor companies with their own fabs—and the associated fixed costs of those fabs—were forced to lay off workers. Fabless companies, on the other hand, for whom manufacturing is a variable cost, fared much better. Xilinx, for example, was able to reduce costs significantly elsewhere and has so far been able to avoid layoffs altogether.

It's likely that the fabless business model will be choice of most semiconductor companies in the future. That's because only large companies like Intel or IBM Microelectronics can justify the tremendous investment required to build a single new 12-inch wafer fab, which can now cost in the neighborhood of $3 billion. In fact, a recent study shows that fewer than 10 integrated device manufacturers will be able to afford a 12-inch fab, not only because of the cost, but also because of the volume of silicon that is produced would be too much for their needs. On the other hand, there are currently more than 600 hundred fabless companies around the world, and their silicon is manufactured by fewer than ten independent wafer foundries like United Microelectronics in Taiwan. Of those, 450 are located in North America.

The Fabless Semiconductor Association in the U.S. has well over three hundred members. Similar associations are expected to be formed in Japan and Europe. In the next five years, it's likely that there will be fewer than a dozen vertically integrated semiconductor companies, but if the trend continues, there will be about one thousand fabless companies. The FSA forecasts that by the year 2010 the fabless industry will account for 50 percent of the world's total semiconductor production.

This is healthy for the industry because the more new companies that are being created, the more creativity, innovation and stimulation will occur. This is a development that will foster the fast development of new chip companies since they will require a relatively small amount of capital. They'll focus on the design and marketing. It's a model that has proven its capabilities, and is more stable than the traditional model in a cyclical industry.

My advice for anyone who wants to start a new semiconductor company is this: go fabless. This will provide you with an advantage. From my experience, a company can only be really excellent at a few things. As a company, you cannot be the best everywhere. The fabless model really allows you to focus on a few things and be good at them, and not waste management talent on building and running a fad. The industry will be stronger, because different companies are focusing on different things. That's why Xilinx is a fabless company, and if you look at our history, profitability, and growth rates, I think it has proven that it is a model that works and is viable.

About the Author Wim Roelandts is president and chief executive officer of Xilinx, Inc. He currently serves as chairman of the Fabless Semiconductor Association and is on the board of directors of the Semiconductor Industry Association.