Toward ML-enhanced Design and Fabrication of Resonant Nanoelectromechanical IR Sensor

This study introduces a novel approach for optimizing the design of ScAIN-based tethered nanoplate resonators with enhanced sensitivity, bandwidth, and thermal efficiency for infrared sensing applications. Without active cooling, achieving uncompromised IR sensor requirements necessitates an advanced design strategy to optimize the design parameter space. We used a feature-based ML approach to learn the design parameters of the nanoplate resonator. A data set encompassing layout (geometric) parameters (L<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</inf>, W<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</inf>, W<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a</inf>, E<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">w</inf>, L<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e</inf>, S, C), measured resonator parameters (f<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">s</inf>, Q, $k_t^2$), and Butterworth-Van Dyke (BVD) fitting results (C<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</inf>, R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</inf>, L<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</inf>, R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">s</inf>, C<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">o</inf>) was collected from 288 ScAIN-based IR sensing tethered nanoplate resonators using the automated RF measurement setup, at various RF power levels (-5, -10, -15 and -20 dBm). We demonstrated that the Random Forest model yields the best results among the other machine learning algorithms tested, achieving a mean absolute error (MAE) of 0.59 MHz and a Mean Squared Error (MSE) of 3.36, with an R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) value of 0.98, indicating an excellent model fit. Using the trained Random Forest method, we demonstrated design exploration (interpolation) for target resonator quality factors. These results underscore the efficacy of the ML-guided design optimization approach in advancing next-generation thermal detector performance for IR sensing applications.