低功耗驱动的电路网表多级划分算法

Power-Driven Circuit Netlist Multilevel Partitioning Algorithm

在划分阶段因得不到实际线长值而无法精确计算功耗值.通过组合使用互连线的通路级数、通路级差和基本线长,提出一种新的独立线长预测方法.使用预测线长和开关活动性的乘积度量划分阶段的动态功耗,并将这一乘积作为权重赋给每条互连线;在聚类和细化处理阶段,尽量避免权重较大的互连线被分割,以实现低功耗驱动的多级划分.实验结果表明,该算法可有效地减小电路的功耗,并且对其他技术指标影响不大.

In order to obtain the approximate wirelength during partitioning phase, a priori wirelength estimation method without actually performing placement and routing is proposed, which combines basic wirelength and two measures of node levels present in a circuit as prediction. During clustering and refinement phase, our power-driven multilevel partitioning algorithm attempts to encapsulate high activity and long wire within partitions, where the connections dissipate less dynamic energy. Experimental results show that our techniques can effectively reduce power.