Communication costs in a multi-tiered MPSoC

The amount of digital processing required for phased array beamformers is very large. It requires many parallel processors, which can be organized in a multi-tiered structure. Communication costs differ for each of the stages in such an architecture. For example, communication costs from the antenna front-end to the first processing stages is costly because of the amount of connections and data rate. Furthermore there is a trade-off between sequential processing exploiting locality of reference versus exploiting parallelism but adding communication costs. Thus, the optimal architecture depends on the importance that is given to the different measures. A model is presented to determine the partitioning of a (beamforming) system based on communication costs. It is shown that different solutions can be explored based on the cost model and the incorporated quantitative and qualitative measures. Determining the importance of each measure is subjective to the situation and application. In this work a simple beamforming application is used optimised for energy efficiency.