Sparse Matrix-Vector Multiplication for Finite Element Method Matrices on FPGAs

The paper presents an architecture and an implementation of an FPGA-based sparse matrix-vector multiplier (SMVM) for use in the iterative solution of large, sparse systems of equations arising from finite element method (FEM) applications. The architecture is based on a pipelined linear array of processing elements (PEs). A hardware-oriented matrix "striping" scheme is developed which reduces the number of required processing elements. The current 8 PE prototype achieves a peak performance of 1.76 GFLOPS and a sustained performance of 1.5 GFLOPS with 8 GB/s of memory bandwidth. The SMVM-pipeline uses 30% of the logic resources and 40% of the memory resources of a Stratix S80 FPGA. By virtue of the local interconnect between the PEs, the SMVM-pipeline obtain scalability features that is only limited by FPGA resources instead of the communication overhead