Symbolic Moment Computation for Statistical Analysis of Large Interconnect Networks
Symbolic Moment Computation for Statistical Analysis of Large Interconnect Networks
复制标题
用于大型互连网络统计分析的符号矩计算
DOI：
10.1109/tvlsi.2012.2197835
复制
发表时间：
2013-05-01
影响因子：
2.8
通讯作者：
Tlelo-Cuautle, Esteban
中科院分区：
文献类型：
--
作者：
Hao, Zhigang;Shi, Guoyong;Tlelo-Cuautle, Esteban
不断缩小的技术特征尺寸和密集的大规模集成使得工艺变化成为最新设计自动化工具面临的一个挑战。工艺变化导致互连网络的严重变化，包括非常大规模的集成互连结构，如时钟树、时钟网、电源-地网络和三维集成电路中的其他布线结构。传统的矩计算技术对这类变分问题的分析只有部分作用，但其计算效率不能满足快速增长的需求，如统计分析。提出了一种用于变分互连分析的符号矩计算器（SMC）。矩计算器是在一个规则的数据结构中构造的，它包含了用于数据存储和计算的二进制决策图。对于一个互连电路，这样的计算图只需要构造一次，并且可以反复调用来计算具有不同参数值的矩。此外，SMC易于互连合成，因为它可以根据对电路结构的修改进行增量修改。讨论了SMC在快速力矩计算、灵敏度分析和统计时序分析中的应用。与其他现有方法相比，该方法具有显著的效率。
The shrinking technology feature size and dense large-scale integration make process variation a challenging issue directly confronting the latest design automation tools. Process variation causes severe variation in interconnect networks, including very large-scale integrated interconnect structures, such as clock trees, clock mesh, power-ground networks, and other wiring structures in 3-D integrated circuits. The traditional moment computation techniques are only partly useful for analyzing such variational problems, however, their computational efficiency cannot meet the quickly rising needs, such as statistical analysis. This paper presents a novel symbolic moment calculator (SMC) for variational interconnect analysis. The moment calculator is constructed in a regular data structure that incorporates binary decision diagrams for data storage and computation. Given an interconnect circuit, such a computation diagram has to be constructed only once and can be repeatedly invoked for computation of moments with varying parameter values. Also, the SMC is friendly to interconnect synthesis in that it can be incrementally modified according to the modifications made to the circuit structure. Applications of the SMC for fast moment computation, sensitivity analysis, and statistical timing analysis are addressed. Significant efficiency is demonstrated comparing to other existing methods.