Another way around monster mask costs

EE Times, SAN MATEO, Calif., 12/9/2002 - The issue of spiraling mask costs for 130 nanometers and beyond has been in the headlines long enough for responses to start forming a pattern. There's denial: "Make them stop!'' There are the searches for alternatives, such as the reemergence of gate array technologies and the further penetration of FPGAs into production. And there have been longer-term attempts at solutions: reusable masks, direct-to-wafer technologies and the like.

Yet, for all this, mask sets for large 130-nm chips still hover around half a million dollars, with occasional excursions upward. Analysts still brandish the threat of the routine million-dollar mask set for 90-nm processes. And there are still designs that demand a full mask set, and that cannot be certain of a DRAM-sized production run. Something has to give somewhere.

In this week's contributed article Frank Schellenberg from the Calibre group at Mentor Graphics Corp. surveys the situation, offers some striking quantitative data on the extent of the problem and reaches a most contrarian conclusion.

The article is available online.

Schellenberg traces the evolution of the photomask to show how we reached our current plight. Along the way, he observes that the two hottest technologies in mask enhancement right now-phase shifting and optical proximity correction-in fact date back at least to an 1896 paper by Lord Rayleigh and a 1963 manual from Eastman Kodak, respectively.

Among the sobering data Schellenberg presents is a comparison of mask-set costs to the scale of ASIC production. After warning that the $1 million mask set has already been sighted here and there around the industry at 130 nm, he quotes a Sematech study that found that for some ASIC houses, the average customer production run was only six wafers. That works out to more than $100,000 in mask costs per wafer-assuming the first mask set goes to production.

So, with solid technical solutions still years away, what can be done? Schellenberg's suggestion should prove quite interesting-and motivating-to designers involved in physical design, mask development or process integration.

Author: Ron Wilson