Grant: Discovery Grants Program - Individual

This project addresses the challenges in modeling and optimization of high-frequency electronic components and packages which are fundamental building blocks in next-generation wireless networks in communications, IoT and autonomous systems. The drive for more functionality, performance and bandwidth in these systems leads to more complex and stringent design requirements, necessitating the modeling of electromagnetic and multiphysics behavior in electronic components and packages while at the same time dealing with increasing dimensions of design parameter spaces. The subsequent challenges in next generation design easily overwhelm today's modeling and optimization capability. Built on top of our recent advances in neural network and neuro-space mapping technologies for radio frequency (RF)/microwave design, this research explores new frontiers in fast design automation algorithms for modeling, optimization, and knowledge-based design. New machine learning based methods employing deep learning for fast parametric models in wider input range and higher dimensional design spaces for electromagnetic and multiphysics design will be developed. New active learning methodologies with high-dimensional adaptive data sampling algorithms will be developed, reducing the cost of data generation for high-dimensional microwave modeling. Incorporating the knowledge of available semi-analytical/equivalent circuit models into different depths of deep learning structures, and exploiting the mapping relations between electromagnetic and multiphysics behaviors will lead to further advances to conquer the expense of parametric modeling in high-dimensional space. Many microwave design problems are inherently inverse of the simulation problem. New inverse modeling approaches will be developed to handle more complicated inverse relationships in high dimensional spaces, addressing non-uniqueness in inverse solutions with more design variables. A further investigation into electromagnetic topology optimization is planned by exploring quantum computing formulations of electromagnetic simulation and optimization, potentially conquering the prohibitive computations of electromagnetic topology optimization which requires exploration of huge design spaces. The long-term direction will be a new paradigm for modeling and optimization unifying knowledge-based design, and deep learning to perform high-frequency electronic design at much larger scale, much faster speed and much higher efficiency than possible today. The long-term impact will be smarter design methodology, faster design cycle, lower design cost, and better design quality. It contributes to creating new knowledge and training of highly qualified technical personnel in areas of high-frequency electronic design. Applications in designing RF/microwave/millimeter-wave circuits, and high-speed integrated circuits and packages also contribute to strengthening the competitiveness our high-tech economy.