Static IR drop prediction is crucial for power integrity, which is an important metric for ensuring product reliability and performance. However, since conventional methods of IR drop analysis are time-consuming, it has become essential to achieve fas...
Static IR drop prediction is crucial for power integrity, which is an important metric for ensuring product reliability and performance. However, since conventional methods of IR drop analysis are time-consuming, it has become essential to achieve fast and accurate IR drop estimation. In this paper, we propose HiFuseUnet, a hybrid U-Net for static IR drop prediction. The proposed model employs hierarchical feature fusion (HFF) in the encoder and multi-scale blocks in the decoder, thereby effectively integrating local information and global context information. Furthermore, we propose PDN-based features to reflect the physical characteristics of the power delivery network (PDN) to improve the accuracy of the model. Experimental results on the benchmark dataset demonstrate that the proposed model achieves better performance than previous methods, improving MAE by 43% over IREDGe, 25% over Swin-Unet, and 34% over MAUnet. Additionally, the model shows superior performance in hotspot prediction at different scales. Compared to commercial tools that take several hours to obtain results, the proposed model is more efficient as it predicts IR drop in less than 1.0 seconds.