| Abstract: | The computer-aided simulation and design of microwave integrated circuits often leads to large-size numerical jobs, especially for modern multiple-device MMIC subsystems such as distributed mixers and amplifiers. When this is the case, the use of large computer systems such as vector processors may become very attractive, because of the dramatic reduction of wall-clock time and of the related improvement of cost-to-performance ratio. This paper discusses the application of supercomputers to the solution of a number of such advanced microwave circuit CAD problems. An overview of the performance of general-purpose vectorized algorithms such as system solvers and Fast Fourier Transforms is first presented, in order to give a feeling of the potential of this class of machines in microwave CAD applications. Then a few typical microwave engineering problems are specifically addressed, including the statistical design (yield optimization) of linear circuits, the analysis of nonlinear circuits under multitone excitation, and the optimization of ultrabroadband nonlinear subsystems. In each case, a possible vectorization strategy is studied in depth, and the structure and performance of a specific highly vectorized algorithm are described in detail. |
