Water remains the best liquid for immersion lithography is the conclusion of the recent SPIE Advanced Lithography conference in San Jose.
Despite a massive search for a liquid with a higher refractive index than water, to allow the further reduction of feature sizes using conventional optical lithography, the industry has not come up with an alternative.
The EDA companies are putting an enormous amount of work into double-patterning, which suggests that this is going to be the workhorse lithographic method at 32nm. The EDA guys have to find ways to divide one layer into two masks.
Will a double patterned wafer cost more than an EUV wafer? It’s dependent on how many wafers you get from a mask, meaning that the high volume wafers will cost less. But it was ever thus in the semiconductor business.
Will EUV make it as a production tool at 32nm required in 2009? Probably not, but it may be used for critical layers on 32nm chips.
So that leaves double patterning immersion lithography as the most likely lithographic tool at 32nm.
Will double patterning extend to 22nm? Probably, Yes. But the proponents of EUV would be severely disappointed if their tool didn’t become a major factor in 22nm production.
Meanwhile the always-the-bridesmaid-never-the-bride candidate, direct write e-beam, is still in there with a chance if EUV and double-patterning never make it.
Are there any Perkin-Elmer Aeble machines still around?
Hi David
We're using e-Shuttle's direct-write e-beam for small volume prototyping right now (for 65nm on 300mm wafers) but it has its own issues for smaller geometries -- these are different to those for conventional masks (the fundamental limit appears to be speed rather than resolution) but are also not solved yet:
http://www.edn.com/blog/1690000169/post/960022696.html?nid=3351&rid=448711237
Cheers
Ian
Thanks Ian. That's interesting. And it's interesting that, after all these years, people are still seeing direct write e-beam as one possible route to a volume production tool. But presumably something very dramatic will have to happen to throughput times for that to be possible.