A matrix product coprocessor for FPGA embedded soft processors

This paper present a matrix coprocessor specialized for computing matrix product. The proposed circuit allows a significant reduction in the computational time required by a general-purpose processor for the sequential execution of the matrix product. The coprocessor has been integrated with a RISC processor and a field programmable system on chip has been realized. The new circuit uses p processor elements to perform a parallel computation of the matrix product during the data exchange with the host processor. Thanks to this, the communication time overhead is minimised and the whole computational time is reduced from O(n³) to O(n²) when n ≤ p.