Physics-guided machine-learning enhanced electrostatic actuated method for in-situ measurement of Young’s modulus

Young’s modulus of polysilicon is a vital mechanical parameter highly dependent on sample preparation and growth techniques. In-situ measurement of this property is essential for effective process control monitoring in microelectromechanical systems (MEMS) fabrication. In this work, an innovative electrostatic actuated method without pull-in instability for in-situ test is proposed. Based on the behavior simulated through finite element method, physics-guided neural networks, which integrate the advantages of both data science models and physics-guided ones, are utilized to extract the Young’s modulus and assess the probability of pull-in instability. Moreover, the performance of the structure is evaluated and optimized through Pareto analysis based on genetic algorithms. It is found that the mapping relationship between systematic parameters, excitation, and response of the structure can be modeled accurately by a physics-guided neural network, and the optimization of design facilities convenience of measurement. Moreover, the error of this method is within 5% under most circumstances, and the measured Young’s modulus through this method is close to that by nanoindentation test. This work explores potential applications of machine learning in MEMS design, testing, and optimization.