Parallel Circuit Simulation on Multi/Many-core Systems

SPICE is widely used for transistor-level circuit simulation. However, with the growing complexity of the VLSI at nano-scale, the traditional SPICE simulator has become inefficient to provide accurate verifications. This thesis tries to accelerate transistor-level simulation on multi/many-core systems, and we will solve 3 problems: 1) develop a parallel sparse LU factorization algorithm for circuit simulation, 2) implement the matrix solver on GPU to further accelerate the solver, 3) develop a circuit partitioning based parallel simulation approach on distributed machines to obtain better scalability. The experimental results show that the proposed parallel LU factorization algorithm effectively accelerates the matrix solver for circuit simulation on both CPU and GPU.