Deep Learning for Spatial Supply Noise Estimation in a Processor Chip

To detect voltage emergency caused by supply voltage fluctuation, people have proposed linear models to estimate the supply noise at a hotspot based on the readings of limited on-chip noise sensors. But such linear model only works well in a power delivery network with linear elements. In this work, we propose to use deep learning to build an accurate noise estimation model for each hotspot in a real chip with nonlinear circuit loads. In our experiments we test the performance of our deep learning based approach on a set of benchmarks with different degree of non-linearity, and we also compare its performance with prior work.