Exploration And Optimization Of A Homogeneous Mesh Of Clusters-Based Fpga Architectures

This paper presents an efficient interconnect network for Mesh of Clusters (MoC) Field-Programmable Gate Array (FPGA) architecture. Compared to conventional MoC-based FPGA, proposed architecture improves the MoC-based interconnect in 2 ways. First, we optimize the intra-cluster interconnect topology by depopulating the intra-cluster full crossbar. Then, we propose a new multi-levels interconnect for the Switch Box (SB) which unifies a downward and an upward unidirectional networks based on the Butterfly-Fat-Tree (BFT) topology. The comparison with the common MoC-based VPR-Style shows that the proposed MoC-based architecture has better area and power efficiency. To optimize the interconnect flexibility of the proposed MoC-based FPGA, we explored and analysed the effect of different architecture parameters on performance, power consumption and density. Experimental results show that architecture parameters can be tuned and adapted to satisfy different specific applicative constraints. Results also show that cluster size 8 presents the best trade-off.