DFG Partitioning Algorithms for Coarse Grained Reconfigurable Array Assisted RTL Simulation Accelerators

As the complexity of circuit design increases, verification of these circuits through simulation also becomes extremely challenging. This creates a bottleneck in the IC design process. Distributed simulation is one way of solving this problem where the simulation workload is distributed among the parallel processors involved in the simulation. However the design has to be carefully partitioned for this purpose. In order to perform distributed simulation, many efficient partitioning algorithms have been proposed till date. These algorithms mostly partition gate level netlist or logic circuits and reduces inter-processor communication by minimizing cutsize for a given constraint of load balance. In this paper, we present two different partitioning schemes for performing distributed simulation. They are: a Discrete Particle Swarm optimization (DPSO) based partitioning algorithm (DPSO-PA) and an effective partitioning heuristic. These algorithms partitions Data Flow Graph (DFG) for a recently proposed Coarse Grained Reconfigurable Array assisted Hardware Accelerator (CGRA-HA). We also propose an improved version of the original DPSO methodology through careful selection of initial partition sets. It is found that the proposed heuristic based partitioning algorithm outperforms the modified DPSO-PA in terms of lesser cut-edges.