Hamiltonian-Based Efficient Algorithms for Legalization with Neighbor Diffusion Effect

Abstract

Neighbor diffusion effect (NDE) is a crucial aspect in advanced technology node that is well-known for its infamous consequence of significant performance decrement of the circuit. In this paper, we observe that NDE is caused by different diffusion heights (the number of fins) between two adjacent cells, and consider reducing the number of height differences in single row to reduce NDE violations. Ignoring the movement of the cells, we first propose a Hamiltonian-completion-based algorithm that reorders the cells in the row such that the number of NDE violations is reduced to a near-optimal value. Then, for a given fixed integer [Formula: see text], we devise an algorithm to compute the new positions of cells, such that the number of NDE violations is bounded by [Formula: see text] and the maximum displacement is minimized. Moreover, we extend our algorithm for legalization in multiple rows against mixed-height cells. Experimental results show that our algorithm reduces the NDE violations to a near-optimal minimum without any area overheads while achieving a better practical running time compared to baselines conforming with the theoretical analysis.