Unsupervised design and geometry optimization of high-sensitivity ring-resonator-based sensors.

In this study, we introduce a technique for unsupervised design and design automation of resonator-based microstrip sensors for dielectric material characterization. Our approach utilizes fundamental building blocks such as circular and square resonators, stubs, and slots, which can be adjusted in size and combined into intricate geometries using appropriate Boolean transformations. The sensor's topology, including its constituent components and their dimensions, is governed by artificial intelligence (AI) techniques, specifically evolutionary algorithms, in conjunction with gradient-based optimizers. This enables not only the explicit enhancement of the circuit's sensitivity but also ensures the attainment of the desired operating frequency. The design process is entirely driven by specifications and does not necessitate any interaction from the designer. We extensively validate our design framework by designing a range of high-performance sensors. Selected devices are experimentally validated, calibrated using inverse modeling techniques, and utilized for characterizing dielectric samples across a wide spectrum of permittivity and thickness. Moreover, comprehensive benchmarking demonstrates the superiority of AI-generated sensors over state-of-the-art designs reported in the literature.