Explainable Machine
Abstract
This work addresses the trust deficit in AI-driven EDA through a Thermodynamic Intelligence Framework. By modeling netlists as dynamic graphs using GNNs and integrating SHAP/LIME for Explainable AI (XAI), we provide designer-actionable diagnostics. The model achieves 74.4% higher accuracy and demonstrates Zero-Shot transfer learning from 28nm training to 7nm prediction.
Fields of Study & Subjects
Explainable Artificial Intelligence (XAI)
Advanced Graph Neural Networks
Adversarial Robustness in Machine Learning
Publication details
- Journal / source
- Figshare
- Publication date
- 1 January 2026