GPU-based PRIMA Algorithm for Passive Model Order Reduction

Analysis of modern high-speed high-density designs with multi-physics-based formulations for signal, power and thermal integrity analysis is becoming increasingly challenging due to the associated mixed-domain and large set of circuit equations. PRIMA algorithm has been widely used for passive reduction of large linear sets of equations for providing computational efficiency. In this paper, PRIMA algorithm is advanced to benefit from the emerging platform of Tensor Core (TC) Graphical Processing Units (GPUs), yielding significant speed-up compared to the traditional approach.