Research on Methods for Very Large Scale Integration Track Assignment Routing

Abstract

Routing is a crucial stage in the physical design of Very Large Scale Integration (VLSI) circuits, comprising three phases: global routing, track assignment routing, and detailed routing. With the development of VLSI circuits, scholars have proposed various track assignment routing algorithms. However, improving the efficiency of track assignment routing and optimizing conflicting design rules have become bottlenecks in track assignment routing problems. This study addresses these bottlenecks by utilizing single-level horizontal and vertical Steiner trees to extract routability information of local wire nets, resolving the adaptation issue between global routing and detailed routing. The proposed algorithm enhances routability information by an average of 16.07% across ten benchmark circuits. Additionally, a Generative Neural Network model based on Conditional Variational Autoencoder (CVAE) is employed to improve the efficiency of track assignment routing, yielding significant simulation results. Furthermore, a negotiation-based tear-and-reassign approach is utilized to address track congestion issues, resulting in an average optimization of 26.03% in overlap cost, with a tradeoff of sacrificing 6.67% of wirelength on average.