A CNN-Based Model for Yield Prediction in MEMS Resonator Manufacturing

This work presents a novel method for predicting the yield of Micro-electromechanical Systems (MEMS) resonators by using Convolutional Neural Networks (CNNs) to extract morphological features from cross-sections images of MEMS resonators’ beams produced by Deep Reactive-Ion Etching (DRIE). By establishing a model that maps the relationships between the MEMS resonators’ operating frequency and the morphological variations in beams induced by process deviations, this method can predict the operating frequency more than 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> times faster than Finite Element Analysis (FEA) simulation, achieving over 96.7% accuracy, advancing the standardization of MEMS resonators manufacturing flow.