An ultra-fine processor for FPGA DSP chip multiprocessors

State-of-the-art FPGA host programmable datapaths which enable trillions of mathematical computations per second. However current FPGA design technology cannot harness these resources in an effective manner. This paper introduces the FPGA Streaming Element (fSE) processor, an ultra-high performance, low cost programmable processor designed to exploit these resources. This enables the performance of dedicated hardware solutions with reduced implementation effort, due to the reduction in effort made possible by working a higher level of abstraction. The performance and resource utilisation of the fSE is benchmarked for a 16 point FFT and compared with a dedicated hardware solution, proving that these processors can achieve performance and cost in excess of dedicated hardware solutions, whilst enabling a simpler software based design solution.