GVF: GPU based Vector Fitting

Abstract

Vector Fitting (VF) algorithm has been widely used for system identification from multiport tabulated data. Particularly, it is of high interest to the design community focused on modeling of high-speed modules such as large number of coupled interconnects, packaging structures and variety of electromagnetic modules. This paper advances the applicability of VF to exploit the emerging massively parallel graphical processing Units (GPUs). Necessary parallelization strategies suitable for GPU platforms are developed. For large problem sizes (increasing number of ports and poles), numerical results demonstrate that the proposed method provides significant speedup compared to both the single CPU based VF as well as existing multi-CPU based parallel VF techniques depending on the number of cores used.