NEATRouter: A New Method for 2D Global Routing

More and more components are integrated into a single chip. This advance makes difficult the task of placement and connecting these components, since more components means more connections to be performed. To address the place and route problem, there are automatic design techniques which allow to find near optimal solutions. In this work, we present NEATRouter as an alternative algorithm to optimize 2D routing of 2-pin nets. Our method focuses on complementing or replacing the MazeRouter algorithm, which is used in several state-of-the-art global routing algorithms. NEATRouter uses the Neuroevolution of Augmenting Topologies (NEAT) algorithm to generate neural networks capable of finding the shortest and most resource-efficient path for a 2-pin net. The results of our experiments suggest that this method can successfully compete with traditional approaches such as MazeRouter, generating quality routing in terms of wirelength and congestion management.