Is the fabless business model becoming less functional? Some members at International Fabless Forum feel so. At the last IFF, Ronnie Vasishta, CEO of Israeli firm eASIC, said, "The current way of doing things is broken. Innovation is stalled, and the cost of customization is prohibitive. It will cost $50 million to develop a 65nm chip . . . The problem is how to move to a new level of abstraction. We need an era of more designs per engineer, and not more engineers per design."
This thinking contrasts with the mainstream thinking at the 450-member FSA. The FSA forecasts that 50 percent of global semiconductor revenues will come from fabless by 2010 from below 15 percent now. Perceptions about fabless have changed since its start in the 1980s when it was ridiculed. A comparison of foundry/fabless roadmap to ITRS roadmap shows that foundry/fabless technology is at par or ahead of the industry.
RETURN TO IDM MODEL In one aspect the fabless model seems to be returning to IDM model. In the 1980s IDMs were engaged in design, manufacturing and marketing. The 1990s saw desegregation, with manufacturing gradually being outsourced to save costs. With technological problems now being encountered in deep nanometer nodes, some designers feel that desegregation is becoming difficult to sustain.
Now we are seeing resegregation, but with a difference. The resegregation is from the point of view of information flow between the design and manufacturing teams. If the collaborating teams can pull together well, resegregation leads to a successful product; but if they cannot, product cost escalates and chances of product success diminishes. With the number of problems in the design-to manufacturing chain growing, and the two teams in different time and geographical zones, resegregation through information flow is becoming challenging.
DFM continues to gain influence, but I think a major change designers should note is that for company success DFM importance cannot override the importance of applications. While IDMs and fabless/foundries drive technology roadmap, the great success stories are written by companies that have identified an application, and transformed DFM into that application, such as iPods or digital cameras. IDMs and fabless/foundries are enablers of that application success. Each device targeting a wider number of applications for better success is a growing trend.
Another major change pertains to chip life cycle. From idea through design through manufacturing through testing to market takes three years. Most time is spent in testing and qualifying. Four to five years from the idea the chip enters significant volume, and ramps down after 10 to 12 years. This contrasts with chips at higher nodes wherein most time was spent in designing and manufacturing.
On one side, companies that used to be vertically integrated, such as Intel, TI and Motorola, aim more at outsourcing to cut down costs. Fabless and fab-lite companies have enabled a foundry growth rate almost double that of semiconductor.
IP CONFIDENTIALITY On the other side, great complexities are getting added to the foundry business. When fabless became a business model, a primary issue was IP confidentiality. This is no more an issue. Foundries have honored their commitment to keep customer IP confidential.
Today getting silicon right the first time is becoming extremely difficult, but is critical to success. No foundry will openly declare its yield, but even a foundry leader like TSMC is said to have a yield of less than 50 percent during its early days of 90nm product manufacturing. Silicon goes through several iterations, which costs more in terms of money and time to market. The tons of verification add even more uncertainty to the life cycle. Communication between designing and manufacturing teams is now becoming a 24/7 proposition. But with the two teams in different time and geographical zones, which is often the case in fab-lite and fabless models, how far will this proposition succeed?
