The gap between design and manufacturing already costs $4 billion to $6 billion a year, according to representatives of the Semiconductor Equipment and Materials International (SEMI) consortium. And the situation is going to get much worse. Once feature sizes go below 100 nanometers, many chips may become economically unfeasible because mask-making and manufacturing costs will be prohibitive.

The problem is that the mask-making and device-manufacturing people aren't getting the data they need in a reasonably compact format, and design teams have no visibility into manufacturing constraints or process variations. A manageable, two-way flow of intelligent data is the only way out of this looming crisis.

The first step in addressing this problem is the new Oasis stream format, which promises a ten- to fiftyfold data reduction over GDSII. Design teams should support this new format as soon as Oasis readers and writers are available. It won't change the design process significantly, but it just might help make masks affordable enough so that chips are actually worth designing and manufacturing.

The second step is a "universal data model" (UDM) that can bring more intelligent data into manufacturing, as well as bring process-related data back to designers. With geometric formats like GDSII or Oasis, mask shops are basically getting a bunch of polygons. They spend time fixing errors in what turns out to be logos or area fills, driving up the cost of the masks and chips.

A SEMI working group has proposed that the OpenAccess data model and application programming interface be used to create this UDM. It's a grand vision that will take time, and it will run into politics, given that OpenAccess doesn't have much EDA vendor acceptance beyond Cadence Design Systems.

Chip designers can support the UDM concept by asking their EDA vendors to get behind it. Chip design teams also must be willing to transmit design-intent data to mask shops and to take manufacturing data back in and do something meaningful with it, such as accounting for process variations during simulation.

Whether a UDM based on OpenAccess will succeed is anyone's guess. But I do think that within a few years, design-for-manufacturing will be just as important as design-for-test is today.