Publication
We propose novel algorithms for post-mapping delay optimization, referred to as delay resynthesis. The state-of-the-art approach optimizes delay by rewriting with optimal local substructures. However, it is limited to local transformations, does not exploit don’t cares, and relies on a basic decomposition heuristic to handle subnetworks with more than four inputs. Our method overcomes these limitations by employing modern resynthesis techniques for non-local optimization and integrating don’t care conditions. In addition, we introduce a more powerful decomposition strategy that extends beyond prior methods. Central to our approach is the set of pairs of functions to be distinguished (SPFD), a Boolean function representation that captures functional dependencies between nodes for finer-grained logic restructuring. Experimental results on EPFL benchmarks show a 5.70% delay improvement over the state of the art.