Abstract
Dynamic IR-drop verification, the analysis of transient supply voltage sag caused by currents in resistive supply metal, can become a simulation bottleneck for analog-mixed-signal (AMS) blocks. Layout-dependent current profiles may demand long transient simulation runs, and a custom analog power delivery network (PDN) lacks the structured grid routing found in digital designs. A prediction method addresses these limitations in two stages. A first stage applies graph attention layers to a circuit graph once per circuit and caches the resulting static embedding. From the cached static embedding, a recurrent waveform predictor forecasts transient supply currents without graph operations in a time-stepping loop. A second stage fuses rasterized transient current maps with resistance maps of the PDN through convolutional encoding and graph convolution, and a deconvolution decoder reconstructs a dynamic IR-drop map. The method may approach reference simulation accuracy at reduced runtime, without retraining per topology.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Pendyala, Prateek, "Dynamic IR-Drop Prediction for Analog-Mixed-Signal Blocks Using a Two-Stage Graph Network Architecture", Technical Disclosure Commons, (September 16, 2026)
https://www.tdcommons.org/dpubs_series/11744