树拓扑片上网络的低能耗映射

Low Energy Mapping for Tree Based Networks-on-Chip

摘要: 针对树拓扑片上网络(NoC)中通信时延受约束的低能耗映射问题，提出了一种递归的二路划分算法RPM(recursive bipartitioning for mapping)。RPM基于分而治之策略，首先将NoC映射转化为多层次的IP核通信任务图划分问题，并采用带参数的Kernighan-Lin算法实现最小割值划分。实验结果表明，与已有算法相比，RPM可以在较短的时间内获得能耗更低的映射解。通过设置不同的参数，RPM既可以用于生成高质量的优化解，也可用于快速的NoC设计空间探索中。

Abstract: A recursive bipartitioning algorithm, RPM, is proposed for low energy mapping in tree-based Network-on-Chip (NoC) architectures subject to communication latency constraints. The mapping problem is formulated to multi-level IP core communication task graph partitioning problems, and the modified Kernighan-Lin mincut heuristic is used to solve them. Experimental results show that RPM obtains lower energy mapping solutions compared with existing algorithms.