The Promised Land Is Still 300 mm Away

After an early flurry of activity, multinational companies have begun to exercise caution when installing additional manufacturing capacity using 300 mm wafers. The combination of lagging global semiconductor demand, excess wafer manufacturing capacity and reductions in feature size have prompted companies to question the need to spend $3B to expand production.

Analysts and industry observers in recent years have significantly changed their forecasts for conversion to 300 mm wafer manufacturing. Global semiconductor forecasts issued in 2000 predicted that the capacity available to deliver leading-edge CMOS logic designs, microprocessors and memory would experience significant constraints. These forecasts spurred 300 mm factory installations by both IDMs and by pure-play foundry service providers.

Figure 1 presents manufacturing wafer size transitions starting at 100 mm and moving through 300 mm wafers. The present iSuppli forecast for the 300 mm conversion follows a more conservative rate of acceptance for the technology between 2003 and 2005, but differs significantly from historical conversion rates after 2010 due to a much quicker decline in the conversion rate.

1. Wafer size transitions from 100 mm through 300 mm.

The acceleration of feature size conversion rates for 0.18 µm and smaller geometries is presented in Figure 2. Although the acceleration will lead to improvements in die cost, the real impact of this phenomenon is the significant increase in unit die output per wafer. When a feature size reduction is coupled with a transition to a larger wafer size, unit output can increase by as much as 40%. This increase in output has resulted in underutilization of existing 300 mm facilities and the postponement of the installation of additional 300 mm capacity. Thus, with the advent of 300 mm technology, the key challenge for semiconductor companies is their capability to sell the output of their leading-edge fabs.

2. Conversion rates for 0.18 µm and smaller geometries is accelerating.

How has the foundry model been affected by the combination of 300 mm wafers and the transition to next-generation feature sizes? The original premise that drove the pure-play foundry model was the development of the fabless semiconductor company. Fabless companies develop unique designs targeted at a specific application. Fabless companies initially were unable to compete with IDMs because fabless companies could not fund advanced technology fabs on their own. Pure-play foundries provided the ideal solution.

The foundry model grew by developing leading-edge technology platforms that multiple fabless companies could use. Unfortunately, most fabless companies do not achieve broad market leadership positions with their products in the applications they serve. This limits the quantity of die required to be manufactured. As the larger foundries transition to 300 mm wafers with leading-edge technology, only a few fabless companies have sufficient volumes to afford foundries an adequate run rate.

Foundries initially solved this problem by developing a method of incorporating multiple designs from several companies on the same set of masks. The practice of processing multi-client wafers provided initial small volumes to clients for initial end-customer qualifications. Multi-client processing is now quite standard for foundries. However, as feature sizes continue to shrink, the die output generated even from multi-client wafers becomes a significant issue.

With the transition to 300 mm wafers, the major foundries are beginning to develop a manufacturing model that is more closely related to a traditional IDM model — key customers consume the majority of capacity and the residual capacity is sold to second-tier clients. Unfortunately, the 80:20 rule has surfaced in the foundry business — 80% of the volume is derived from 20% of the designs. This has translated into underutilized capacity at 300 mm fabs.

As the semiconductor market looks to rebound after several years of dismal performance, a new issue has surfaced: Only a few companies have installed leading-edge technology. This issue will become more acute as the anticipated recovery takes shape.

So why aren't companies installing additional capacity in anticipation of the recovery? The answer is simple economics. After suffering through two years of financial losses, companies are unwilling to spend capital for capacity — even though they realize that it will eventually lead to additional revenue. By creating an imbalance between the supply and demand for technology, companies will drive up prices for their products. After two years of being unable to load their factories, manufacturers are more than ready to tip the scales toward constricted supply.