Deep Learning for Power and Switching Activity Estimation

This chapter covers the topics related to achieving efficient switching activity estimation, in terms of speed and accuracy, which in turn provide efficient power estimation. It first introduces the topic of power estimation, switching activity estimation, and their role in VLSI design. The chapter describes conventional tools to perform power estimation: gate-level simulation and switching activity estimators, as well as some drawbacks that these conventional tools and methods pose. Then, the chapter delves into tried-and-true modeling methods in the past to alleviate some of the conventional methods’ drawbacks. An overview of the new trend of using deep learning models and techniques for power estimation is given, before a more in-depth treatment of the topic is shown using two case studies. The two case studies focus on state-of-the-art deep learning models such as convolutional neural networks and graph neural networks.