A concurrent approach to clustering algorithm with applications to VLSI domain

Abstract

Circuit partitioning plays an important role in physical design automation of very large scale integration (VLSI) chips. In this brief we present a new connectivity based top down as well as bottom up approach to clustering algorithm for VLSI circuit partitioning. The proposed clustering algorithm partitions the circuit by focusing on highly interconnected cell groups. This clustering algorithm leads to a parallel implementation in which multiple processors are used to identify clusters simultaneously. The process starts with forming clusters by grouping the cells that are tightly connected and as well as the cells that are loosely connected. Considering both types of groups has the advantage that clusters formed from this technique will be highly connected and compact too. Therefore the proposed clustering method can reduce the size and also speed-up the large-scale partitioning problem without loosing partitioning solution qualities. The performance of the proposed clustering algorithm is evaluated on a standard set of partitioning benchmark-ISPD98 benchmark suite.

An efficient VLSI circuit partitioning algorithm based on satin bowerbird optimization (SBO)

“…They choose cells to shift, aiming to shift clusters that sit astride two partition subsets to form a single subset. A connectivity-dependent clustering method can be used to resolve partitioning of the circuit [8,9]. However, this clustering algorithm usually suffers from poor capacity for local optimization, so its results are commonly applied as the starting point for other algorithms.…”

“…The velocity V i (t) of each particle is described by Eq. (8) The velocity V(t) of the population X(t) is described by Eq. (9) where N is the size of the population.…”

Hybrid Particle Swarm Optimization Algorithm for VLSI Circuit Partitioning

“…目前,已有一些学者针对电路划分的最小割优化问题进行算法构造研究.Fiduccia 和 Mattheyses 于 1982 年 提出的 FM 算法 [3] 作为电路划分问题的一种经典实用算法,是最具代表性的基于模块移动的迭代改进策略.该算 法在每次迭代过程中都是通过移动局部范围内的最佳模块来完成.FM 算法自提出以来,许多学者在此基础上 相继提出了许多改进算法,且都能得到优于原算法的结果.但这类算法同属于局部寻优算法,尤其在划分规模较 大的电路时,体现出明显缺乏全局寻优能力的缺陷.文献 [4,5]采用聚类算法进行电路划分,这类算法一般是通过 根据某一给定的评价方法将一些模块聚集到同一集合中来实现.这类方法主要是缺乏细粒度的寻优能力,通常 情况下,这类算法所找到的解可作为其他算法的初始输入,然后再进行寻优.文献 [6,7]则是采用了基于光谱的划 分算法,它们主要是在将图转化为 Laplace 谱后,通过分析 Laplace 特征向量并得到划分结果.这类算法的划分结 果具有全局性,然而由于算法涉及求解特征值和特征向量问题,计算复杂度较高,不利于进行超大规模的计算. 另外,文献 [8]采用模拟退火算法解决二划分问题,算法主要采用了随机两模块交换策略.文献 [9,10]则是采用遗 传算法(genetic algorithm,简称 GA)来解决电路划分问题,文献除了对遗传算法参数设置进行研究以外,还通过 考虑特定电路划分问题的特点,设计并测试了多种不同适用于该问题的编码、交叉和变异的策略,并取得了较 好的划分结果.这也表明智能优化算法在该问题中具有较好的应用前景.…”