Abstract: Accurate phased-array antenna design depends on full-wave electromagnetic simulations, whose runtime scales sharply with antenna array size. Conventional numerical methods for antenna array simulation trade speed for accuracy, failing to capture electromagnetic radiation patterns at critical regions with enough dynamic range. Machine Learning (ML) is an established method for replacing expensive and time-consuming numerical simulations in design automation. Most existing ML approaches used as surrogate models for antenna radiation modeling support only a fixed array size. This requires retraining on thousands of expensive simulations if the array size changes. Moreover, they neglect the spurious radiation terms (mutual coupling of antenna elements, feed radiation, array edge radiation, unwanted surface waves), leading to inaccuracy in radiation pattern prediction. In this talk, Morteza Fayazi will present RaPID, a physics-informed surrogate model that performs zero-shot prediction of radiation patterns for arbitrarily large two-dimensional antenna arrays, trained only on inexpensive small-array data. Test results show that for several planar antenna sizes, compared to state-of-the-art approaches, RaPID achieves 6x–12x more accuracy while using at least 1,575 fewer large-array full-wave simulations. Additionally, RaPID achieves a 2x–55x reduction in the time needed to procure training data.
Bio: Morteza Fayazi was born in Tehran, Iran. He received a bachelor’s degree in electrical engineering with a minor in computer science from Sharif University of Technology (SUT), Tehran, Iran, in 2017. He also received his master’s and doctoral degrees in electrical engineering and computer science from the University of Michigan, Ann Arbor, in 2020 and 2024, respectively. In 2024, he joined the University of Utah, Salt Lake City, USA, as an assistant professor of electrical and computer engineering. His research interests include electronic design automation (EDA) and artificial intelligence (AI). He has designed and developed multiple AI-based tools for analog, radiofrequency (RF), and system-on-chip circuit design automation, including MuaLLM, SINA, AnGeL, and FuNToM.
