A Survey of Methods and Architectures for Improving the Thermo-Electro-Mechanical Responses of MEMS Resonant Accelerometers via a Novel Bidirectional Long Short-Term Memory

Abstract

Microelectromechanical systems (MEMS) resonant accelerometers have gained significant attention due to their high precision, stability, and suitability for advanced sensing applications in aerospace, automotive, and biomedical domains. However, their performance is highly influenced by thermo-electro-mechanical interactions, leading to nonlinearities, drift, and reduced sensitivity under varying environmental conditions. Traditional compensation techniques often rely on physical modeling and calibration strategies, which are limited in handling complex nonlinear dependencies. Recent advancements in artificial intelligence, particularly deep learning, have introduced data-driven approaches for modeling and mitigating such effects. This paper presents a comprehensive survey of methods and architectures aimed at improving the thermo-electro-mechanical responses of MEMS resonant accelerometers, with a particular focus on Bidirectional Long Short-Term Memory networks. The ability of Bidirectional Long Short-Term Memory models to capture temporal dependencies in both forward and backward directions makes them highly suitable for modeling dynamic sensor behaviors. The survey critically analyzes existing approaches, including physics-based models, machine learning techniques, hybrid frameworks, and optimization strategies. Furthermore, it highlights the advantages of Bidirectional Long Short-Term Memory in enhancing prediction accuracy, reducing drift, and improving robustness against environmental variations. Trends, challenges, and future research directions are also discussed, providing insights into the integration of intelligent algorithms with MEMS devices for next-generation sensing systems.