IncPIRD: Fast Learning-Based Prediction of Incremental IR Drop

GridNet

“…A number of numerical techniques have been well developed and can perform IR drop analysis well on power grids, such as hierarchical methods, random walk methods, Krylov-subspace methods, multi-grid techniques, and vectorless veri cation methods. To further speed up the IR drop analysis, several machine learning based IR drop estimation/prediction methods have been proposed [14,17,21,28]. Those methods typically aim to replace the standard full-chip IR drop analysis tool such as ANSYS RedHawk, via data-based learning and feature selections.…”

“…Xie et al [28] proposed a CNN-based model transferable across di erent designs that is able to incorporate design-dependent features during preprocessing. Ho et al [17] focused on incremental IR drop prediction and mitigation. It uses more electrical and physical features for the training based on the gradient boosting framework.…”

“…Hence, there is a need for fast incremental EM-induced IR drop analysis for ECO like xing for the PDN layout via proper sizing of some wire segments [8]. We want to point out that such fast EM-induced IR drop analysis is di erent than the traditional fast static and dynamic IR drop analysis methods [14,17,21,28] as those methods mainly target at accelerating the design time dynamic IR drop analysis tool such as ANSYS Red-Hawk based on the power and timing information of the standard cells.…”

Dynamic IR-drop ECO optimization by cell movement with current waveform staggering and machine learning guidance

“…Compared with time-consuming IR-drop analysis (a chip-level analysis usually takes over 12 hours for a multi-milliongate design), a fast and accurate IR-drop prediction model shortens the turnaround time between IR-drop optimization and IR-drop analysis. Therefore, as summarized in Table 1, state-of-the-art IR-drop prediction works rely on machine learning instead of approximation by analytical formulae [5][6][7]. Ho and Kahng in [5] propose an XG-Boost model to predict static IR-drop for incremental modification of placement and PG network.…”

“…Therefore, as summarized in Table 1, state-of-the-art IR-drop prediction works rely on machine learning instead of approximation by analytical formulae [5][6][7]. Ho and Kahng in [5] propose an XG-Boost model to predict static IR-drop for incremental modification of placement and PG network. Their model achieves a good quality for static IR-drop prediction but cannot handle dynamic IR-drop.…”