A scalable distributed memory architecture for Network on Chip

Besides processors, DSPs and other IP blocks, Network on Chip (NoC) also integrates lots of memories. However, the research of on chip memory subsystem for NoC has not been undertaken thoroughly. In this paper, a distributed None Uniform Memory Access (NUMA) memory architecture for NoC is developed; the performance and the programming mode of this system are discussed. Two kinds of parallel algorithms are implemented, the pipelined matrix multiplication and the full-parallel FFT. Simulation results verify the proposed architecture can achieve high parallelism and provide flexible programming scheme.

[1]  Axel Jantsch,et al.  The Nostrum backbone-a communication protocol stack for Networks on Chip , 2004, 17th International Conference on VLSI Design. Proceedings..

[2]  Axel Jantsch,et al.  Networks on chip , 2003 .

[3]  Gianluca Palermo,et al.  Exploration of distributed shared memory architectures for NoC-based multiprocessors , 2007, J. Syst. Archit..

[4]  Nikil D. Dutt,et al.  A Framework for Cosynthesis of Memory and Communication Architectures for MPSoC , 2007, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems.

[5]  Ran Ginosar,et al.  The Power of Priority: NoC Based Distributed Cache Coherency , 2007, First International Symposium on Networks-on-Chip (NOCS'07).

[6]  Alain Greiner,et al.  On Cache Coherency and Memory Consistency Issues in NoC Based Shared Memory Multiprocessor SoC Architectures , 2006, 9th EUROMICRO Conference on Digital System Design (DSD'06).

[7]  Mahmut Kandemir,et al.  Memory Systems and Compiler Support for MPSoC Architectures , 2005 .