The transition from 0.18-micron to 0.13-micron process technology, like all previous design dimension reductions, introduces design problems and mask cost problems.Design problems are pretty much borne by all vendors alike. The problem of mask costs, although also shared by all vendors, creates winners and losers.

A 15-mask set of 0.25-micron masks costs $150,000, 21 masks at 0.18 micron cost $250,000, and a set of 0.13-micron masks costs $600,000. The greater than 2x premium in 0.13-micron mask sets over 0.18-micron versions raises the volume threshold beyond which it is better to do an ASIC or ASSP than an FPGA.

Before FPGA vendors jump up and down too much, $600k is not all that far away from $250k in the context of the $20 million or so the average startup raises to get to first silicon. Things do get interesting, and consequently more favorable for FPGA vendors, at 90 nanometers, where mask sets will cost $1.5 million, and especially at 65 nm, where the price will approach $4 million.

More significant than the design rule shift is the shift from 200-mm to 300-mm wafers. Because of rapid capital depreciation and the competitive nature of the semiconductor industry, it only makes sense to build a fab if the facility can remain near 100 percent utilization. At a fairly optimistic $150 of revenue per square inch of silicon, a single 300-mm wafer is worth around $20,000. Multiply that by 25,000 wafer starts per month and 12 months per year, and it's quickly apparent that a company must sell $4.5 billion worth of semiconductors to justify owning a fab.

What does all this mean? First, the move from 0.18-micron to 0.13-micron process technology is a lot less disruptive than some vendors claim. Second, the move to 300-mm wafers will leave fewer companies owning their own fabs. More vendors will use foundries or ink deals with competitors to build jointly owned fabs. The few that can fill 300-mm fabs will pull increasingly ahead of direct competitors that choose to use foundries. With few exceptions, though, the fabless do tend to compete with the fabless, so all is not lost.

Jeremey Donovan (jeremey.donovan@gartner.com<.A>) is Vice President and Chief Analyst at Gartner Dataquest.