KEYNOTE: AI/Machine Learning Technologies for Electromagnetic/Multiphysics Based Modeling and Optimization

AI and machine learning are unconventional technologies with unique capability to address challenges in electromagnetic-based analysis and optimization in high-speed/high-frequency electronic components/packages and systems. With phenomenal progress in electromagnetic based computation algorithms, along with dramatic changes in the computing environment, high-fidelity electromagnetic models are now an important part of high-speed/high-frequency electromagnetic design automation. However new design challenges continue to rise. Electromagnetic structures and circuits are becoming more complex, and frequency is getting higher. More sophistication in multi-physics modeling and design are becoming increasingly necessary. Meaningful design problems easily become computationally prohibitive. In this talk, we present AI/machine learning technologies for electromagnetic /multiphysics based modeling and optimization, and their applications to signal/power integrity analysis of high-speed/high-frequency electronic packages and subsystems. We will highlight emerging directions of knowledge-based, cognition-driven design. Incorporating domain-specific design knowledge/engineering equations into artificial neural networks, knowledge-based and deep-learning based computational technologies are producing fine-grained modeling and design solutions for problems which are otherwise computationally very expensive. New formulations of inverse neural network training algorithms allow instant solutions to electromagnetic inverse modeling problems addressing the technical challenges of non-uniqueness in inverse modeling. Emerging machine learning structures and optimization algorithms for electromagnetic based design and signal/power integrity analysis will be discussed.