Worst-Case Dynamic Power Distribution Network Noise Prediction Using Convolutional Neural Network
TimeThursday, July 14th4:30pm - 4:50pm PDT
Location3007, Level 3
Timing and Low Power Design
DescriptionIR drop analysis is significant in signoff stage to ensure the performance of chips. However, dynamic IR drop simulation becomes time-consuming as the chip integration increases. In this work, we propose a fast dynamic IR drop prediction model based on CNN. The model is trained with easily accessible features of instance current and distance to power source, and predicts the maximum IR drop distribution of PDN. We also present an current feature compression algorithm for the model to speed up inference. Experimental results show that the proposed model achieves both high accuracy and efficiency and outperforms the latest ML method.