Slack Optimization of Timing-Critical Nets

Abstract

The construction of buffered Steiner trees becomes more and more important in the physical design process of modern chips. In this paper we focus on delay optimization of timing-critical buffered Steiner tree instances in the presence of obstacles. As a secondary goal, we are interested in minimizing power consumption.

Since the problem is NP-hard, we first study an efficient method to compute upper bounds on the achievable slack. This leads to the interesting subproblem to find shortest weighted paths under special length restrictions on routing over obstacles. We prove that the latter problem can be solved efficiently by Dijkstra’s method. In the main part we describe a new approach for the buffered Steiner tree problem. The core step is an iterative clustering method to build up the tree topology.

We provide a case study for the effectiveness of the proposed method to construct buffered Steiner trees. Our computational experiments on four different chip designs demonstrate that the proposed method yields results which are relatively close to the slack bounds. Moreover, we improve significantly upon a standard industry tool: we simultaneously improve the slack and largely reduce power consumption.