A Parallel GPU Implementation of the TimberWolf Placement Algorithm

GPUs have been gaining acceptance in the electronic design automation field as attractive platforms for implementing and accelerating computationally extensive applications. Researchers agree that it is critical that EDA algorithms exploit future platforms and explore the use of parallel algorithms as we move to the manycore era. This paper describes the implementation of the TimberWolf placement algorithm using CUDA and demonstrates the applicability of GPUs in accelerating electronic design automation tools. The algorithm has been implemented on a Xeon Workstation using C, and achieved a substantial acceleration on an Nvidia Tesla C2070 card.