The type of patterning to be used at a given technology node is determined layer-by-layer. At 10 nm, SADP is planned for metals, but contacts are looking like they’ll require triple patterning, according to Mentor’s David Abercrombie.
We’ve talked about LELE coloring before, and algorithms exist to automatically color a circuit with linear scaling – twice as big a circuit takes twice as long to color. But when you go to LELELE, it’s no longer linear: it’s np-complete. Beyond 30 features to be colored, the algorithm isn’t going to finish in a lifetime.
So coloring algorithms are instead being approached using heuristics and constraints. For instance, it’s generally considered better to balance the colors so that there are roughly the same number of features on each color. That eliminates all of the gazillion possible decompositions that don’t meet that criterion.
There is also some concern that different decompositions might have different litho effects. It’s not known yet whether this will be an issue, but that would certainly complicate matters, since the algorithm would now have to take lithographic distortions into effect – as if it didn’t already have enough to think about.