NSF Award
2430603
This project aims to revolutionize the field of microwave circuit design by developing an advanced inverse design (InvDes) framework. Microwave circuits are essential in communication and sensing systems critical for consumer electronics, healthcare, and defense, and their design significantly impacts the overall performance, cost, and efficiency of these systems. Current design methods primarily rely on human engineers leveraging their training and experience to craft circuit layouts. In contrast, InvDes lets a computer algorithm find designs that maximize desired performance within specific constraints, discovering new topologies and shapes beyond human intuition. This project will create a systematic framework for the InvDes of planar microwave devices -- filters, splitters, baluns, antennas, and more -- meeting complex performance requirements and shape/size or fabrication constraints demanded by diverse application. The project will support the education of the nation's next generation of electronic engineers, preparing them to excel in the era of advanced computing by involving them in cutting-edge research and developing new curriculum modules. The designs generated by the proposed approach will be fabricated and validated alongside industry partners, solving outstanding challenges in microwave engineering and immediately benefitting a wide array of applications from autonomous vehicle radars to next-generation wireless communication.<br/><br/>This project seeks to create a unified framework for the inverse design (InvDes) of planar microwave and millimeter wave devices, moving beyond conventional circuit design techniques to explore a substantially broader design space. Realizing this goal requires parameterization techniques that give the algorithm maximal freedom to design devices of any necessary topology and shape, powerful optimizers pioneered for use in machine learning tasks, and highly efficient simulators such as GPU-accelerated finite difference methods and versatile finite element and boundary element solvers. For the first time, multilayer devices will be fully machine designed, with both metal layers and via placement controlled by the algorithm, enabling InvDes of the full range of modern microwave components. To address the non-convexity of the design landscape, which generally demands many trials with random initial conditions to find high-performing devices, our initializations will be pre-optimized by first minimizing a convex dual problem with relaxed physics. The designs generated by this algorithm will be fabricated using both macroscopic printed circuit board (PCB) and nanoscale complementary metal-oxide semiconductor (CMOS) technology. Experimental verification of device performance will utilize broadband network analyzers, as well as microwave impedance microscopy which permits mapping of the local electric field distribution in operando with exquisite spatial resolution.<br/><br/>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.
