Simultaneous logic decomposition with technology mapping in FPGA designs

Conventional technology mapping algorithms for SRAM-based Field Programmable Gate Arrays (FPGAs) are normally carried out on a fixed logic decomposition of a circuit. The impact of logic decomposition on delay and area of the technology mapping solutions is not well understood. In this paper, we present an algorithm named SLDMap that performs delay-minimized technology mapping on a large set of decompositions and simultaneously controls the mapping area under delay constraints. Our study leads to two conclusions: (1) For depth minimization, the best algorithms in conventional flow (dmig + CutMap) produce satisfactory results with a short runtime, even with a fixed decomposition; (2) When all the structural decompositions of the 6-bounded Boolean network are explored, SLDMap consistently outperforms the state-of-the-art separate flow (dmig + CutMap) by 12% in depth and 10% in area on average; it also consistently outperforms the state-of-the-art combined approach dogma by 8% in depth and 6% in area on average.