Physics-Informed Neural Networks for Modal Analysis of Diaphragm-Structured Mems with Experimental Validation

We present and experimentally validate a novel framework that leverages physics-informed neural networks (PINNs) for advanced MEMS eigenmode analysis. Our work advances the field in three key ways: (1) we demonstrate the successful application of PINNs to calculate multiple vibration modes in MEMS diaphragms of arbitrary geometry by proposing and implementing three key physical equation as constraints; (2) we validate our PINNs model through digital holographic microscopy (DHM) measurements on a fabricated Piezoelectric Micromachined Ultrasonic Transducer (PMUT) device; and (3) we develop a highly efficient computational platform that combines our PINNs and already existing analytical models to evaluate PMUT array performance using only geometrical configurations of the array.