EMGraph: Fast Learning-Based Electromigration Analysis for Multi-Segment Interconnect Using Graph Convolution Networks
TimeWednesday, December 8th4:30pm - 4:50pm PST
Event Type
Research Manuscript
Virtual Programs
Presented In-Person
Manufacturing Test and Reliability
DescriptionElectromigration (EM) becomes a major concern for VLSI circuits as the technology advances in nanometer regime. VLSI interconnect can be naturally viewed as graphs. Based on this observation, we propose a new graph convolution network (GCN) model, which is called EMGraph, to estimate transient EM stress. Compared with generative adversarial network (GAN) method, EMGraph model can learn more transferable knowledge to predict stress on new graphs without retraining via inductive learning. Experimental results show the model has 1.5% averaged error compared to the ground truth. It also achieves smaller model size, 4× accuracy and 14× speedup over the GAN-based method.