Rapid In-Line Identification of Defects in Electronics Manufacturing

This research project addresses a critical challenge in semiconductor manufacturing: the inability of current inspection technologies to detect defects hidden beneath the surface of complex electronic packages. As devices become smaller and more densely packed, undetected subsurface defects can lead to costly yield losses and reliability issues. By pioneering a new approach that combines millimeter-wave imaging, artificial intelligence, and digital-twin modeling, the project aims to enable real-time, non-destructive inspection of electronic packages. This advancement promises significant societal and national benefits, including improved manufacturing reliability, reduced waste, and enhanced production efficiency. Furthermore, the project supports the U.S. semiconductor ecosystem by integrating cutting-edge digital manufacturing technologies and fostering workforce development. Through hands-on training and educational outreach, students from K-12 to graduate levels will gain valuable experience in advanced electronics manufacturing, helping to build a skilled workforce for the future. The research outcomes will be widely disseminated through publications, conferences, and collaborative industry projects, further amplifying its impacts across academia and industry. The project seeks to overcome the limitations of surface-only optical inspection by developing a novel framework that integrates millimeter-wave electromagnetic subsurface imaging with physics-informed AI and digital-twin models. The goals include creating a physics-informed neural network methodology that embeds Maxwell’s equations, material properties, and boundary conditions into the learning process, enabling accurate reconstruction of subsurface scattering images and defect identification. The approach leverages new activation functions to improve convergence and numerical accuracy, achieving manufacturing-grade precision. A scalable multi-fidelity convolutional framework will be developed to localize and classify defects by coupling low-fidelity physics-based reconstructions with measured data and digital-twin models, examining discrepancies to pinpoint defects. The project also involves designing a millimeter-wave imaging hardware platform with ultra-wideband antenna arrays and beamforming networks operating across 100 to 300 GHz, optimized for manufacturing environments. Validation will be performed by demonstrating sub-wavelength defect identification, down to 10 micrometers, surpassing conventional resolution limits. This integrated approach is expected to advance knowledge in non-destructive inspection, high-density microelectronics manufacturing, and the application of AI in physics-based imaging, setting new standards for reliability and precision in semiconductor production. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.