Due to the increasing complexity of VLSI circuits, power grid simulation has become more and more time-consuming. Hence, there is a need for fast and accurate power grid simulator. In order to perform power grid simulation in a timely manner, parallel algorithms have been developed to accelerate the simulation. In this dissertation, we present parallel algorithms and software for power grid simulation on CPU-GPU platforms. The power grid is divided into disjoint partitions. The partitions are enlarged using Breath First Search (BFS) method. In the partition enlarging process, a portion of edges are ignored to make the matrix factorization light-weight. Solving the enlarged partitions using a direct solver serves as a preconditioner for the Preconditioned Conjugate Gradient (PCG) method that is used to solve the power grid. This work combines the advantages of direct solvers and iterative solvers to obtain an efficient hybrid parallel solver. Two-tier parallelism is harnessed using MPI for partitions and CUDA within each partition. The experiments conducted on supercomputing clusters demonstrate significant speed improvements over a state-of-the-art direct solver in both static and transient analysis.
