Bridge-Defect Prediction in SRAM Circuits Using Random Forest, XGBoost, and LightGBM Learners

The modern fabrication technology node for devices and circuits is causing several failure analysis challenges of the current state-of-the-art tools to actually find the location of the physical defects. Here, we show an approach of combining SPICE circuit simulations and machine learning for bridging-detect predictions in a 6T-SRAM cell. Our predictive model consists of Random Forest, XGBoost, and LightGBM algorithms which are trained with several electrical attributes of the circuit having faults. We compare the performance of the algorithms and calculate the accuracy in predicting the defect location. We believe that this approach promises to improve the failure analysis, enhancing the cycle of design to product.