A Clustering Based Linear Ordering Algorithm for K-Way Spectral Partitioning

The spectral method can lead to a high quality of multi-way partition due to its ability to capture global netlist information. For spectral partition, n netlist modules are mapped to n points in d-dimensional space, and then a linear ordering of these n modules is constructed to be used as a basis for partitioning. In this paper, we propose two clustering based linear ordering algorithms taking into consideration the objective function presented by [l]. A general k-way partitioning problem can be defined as follows: Given n modules V = {VI, 212, ...,v,}, construct a k-way partition P k in which V is divided into k disjoint clusters S I ,S2, ...,sk with (SIU S2 U ...U Sk) = V and Si n Sj = 0 for 1 5 i , j 5 k, and a user-defined object function F(P k )is minimized. In this paper, we consider the partition problem that divides the netlist modules into clusters subject to size balance constraint while minimizing cut nets among clusters. Therefore, we will adopt the scaled cost [5] as ou...

• Type: Conference
• Year: 1999
• Where: ASPDAC
• Authors: Shiuann-Shiuh Lin, Wen-Hsin Chen, Wen-Wei Lin, TingTing Hwang
• Keywords: ASPDAC 1999, Global Netlist Information, Hardware, Linear Ordering, Netlist Modules