Multirate Integration Properties of Waveform Relaxation with Applications to Circuit Simulation and Parallel Computation

Abstract : Because of the high cost of fabricating an Integrated Circuit(IC), it is important to verify the design using simulation. There are a wide variety of techniques for simulating integrated circuit designs. but the most accurate and reliable is to construct the System of nonlinear ordinary di1ferential equations that describe a given circuit. and solve the system with a numerical integration method. This approach, referred to as circuit simulation, is computationally expensive, particularly when applied to large circuits. To reduce the computation time required to simulate large MOS circuits, new numerical integration algorithms based on relaxation techniques have been developed. These techniques can reduce the simulation time as much as an order of magnitude over standard circuit simulation programs. In addition, they are particularly suited for parallel implementation. This thesis covers both the classical numerical techniques and the new relaxation-based algorithms. with particular emphasis on the Waveform Relaxation (WR) family of algorithms. Algorithms in this family are reviewed, convergence theorems are included, and their implementations on a parallel processor are presented.