Parallel Processing for Pitch Splitting Decomposition
2009
Decomposition of an input pattern in preparation for a double patterning process is an inherently global
problem in which the influence of a local decomposition decision can be felt across an entire pattern. In
spite of this, a large portion of the work can be massively distributed. Here, we discuss the advantages
of geometric distribution for polygon operations with limited range of influence. Further, we have found
that even the naturally global "coloring" step can, in large part, be handled in a geometrically local
manner. In some practical cases, up to 70% of the work can be distributed geometrically. We also
describe the methods for partitioning the problem into local pieces and present scaling data up to 100
CPUs. These techniques reduce DPT decomposition runtime by orders of magnitude.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
2
References
1
Citations
NaN
KQI