Martin van den Brink, EVP, Marketing & Technology, ASML, Veldhoven, The Netherlands

As the global economic tide starts to recede, it exposes many high technology manufacturing challenges that were otherwise hidden. In the semiconductor industry, these challenges are often opportunities for innovation — and the situation we face in lithography today is an excellent example.

At 32nm and smaller, the design for manufacturing (DFM) approach employs ever-more sophisticated off-line manipulations of illumination shapes and mask patterns (through optical proximity correction and other resolution enhancement techniques), to accommodate fab-floor lithography. Is this the right strategy going forward? What if we took a different approach? What if we questioned the DFM concept?

I believe the coming year will bring a shift toward manufacturability for design (MFD). Fortunately, the industry has the technology in place and the ability to accomplish this shift in production almost immediately. Crucial to this ability is computational modeling and analysis of patterning. These techniques, coupled with double patterning, will allow current systems to address the early 22nm node.

While computational design techniques have been leveraged very successfully, they are by definition using a generic representation of scanner performance. Manufacturing was left with the task of getting the real-world mix of scanners in the fab to match these ideal generic designs. Now, however, the margins required for critical dimension uniformity and overlay have become tight enough that individual scanners must be optimized for each layer of each design in the fab — placing even more importance on MFD.

Computational lithography can be leveraged to apply full-chip modeling and analysis to fleets of optimized scanners on the fab floor, by modeling the unique performance characteristics of each scanner. By combining these models with existing knowledge of tunable scanner parameters, each system can be quickly optimized for each layer and design.

Computational lithography is no longer limited to simply making the mask design suitable for a general scanner type in the fab — DFM. It is now ready to optimize each scanner in the fab to each incoming design — MFD.