Efficient Standard-Cell Legalization for Minimization of Total Movement

Abstract

In this paper, given a global standard-cell placement, based on the total weighted Manhattan movement as the disturbance metric, a two-phase legalization algorithm is proposed to minimize the total movement of the legal cells with satisfying the capacity and non-overlapping constraints. In the proposed algorithm, firstly, all the placed cells can be initially allocated onto original rows and some placed cells can be reallocated onto adjacent rows to satisfying the capacity constraint with minimizing the total vertical movement. Furthermore, all the cells inside one specific row can be legalized to satisfy the non-overlapping constraint with minimizing the total horizontal movement with. Compared with Abacus and HiBinLegalizer in the standard-cell legalization, the experimental results show that the proposed algorithm can decrease 80% and 54% of CPU time to reduce 32% and 8% of the total movement and 59% and 4% of the maximum movement on the standard cells for the 7 tested circuits on the average, respectively.