Efficient acceleration of sparse MPIE/MoM with graphics processing units

In this paper, graphics processing unit (GPU) is used to accelerate both the impedance matrix calculation and the linear system solution in a Mixed-Potential Integral-Equation (MPIE) formulation of the method of moments (MoM). We consider a three-step process. First, the MoM matrix is computed on the GPU by using a quasi-one-dimensional approximation of the MPIE formulation. Second, a sparse representation of the matrix is obtained by thresholding. Third, a GPU-enabled biconjugate gradient (BiCG) method is used to solve the linear system. We validate the accuracy of our approach by analysing a microstrip branch-line coupler and comparing the results provided by the GPU code with a well-known commercial simulator. A performance analysis proves that the developed algorithm, running on a cheap off-the-shelf GPU, is about 25 times faster than the CPU version.