A Comprehensive Review of Improving the Thermo-Electro-Mechanical Responses of MEMS Resonant Accelerometers via a Novel Bidirectional Long Short-Term Memory

Abstract

Micro-Electro-Mechanical Systems (MEMS) resonant accelerometers have emerged as highly sensitive and stable sensing devices widely used in navigation, aerospace, and biomedical applications. However, their performance is significantly influenced by thermo-electro-mechanical coupling effects, leading to drift, nonlinearity, and reduced accuracy under varying environmental conditions. Recent advancements in data-driven modeling, particularly deep learning techniques, have opened new avenues for compensating such complex nonlinearities. This paper presents a comprehensive review of methods aimed at improving the thermo-electro-mechanical responses of MEMS resonant accelerometers, with a special focus on the application of Bidirectional Long Short-Term Memory (BiLSTM) networks. The study examines existing modeling techniques, including physics-based approaches, machine learning models, and hybrid frameworks, highlighting their strengths and limitations. Special emphasis is placed on the ability of BiLSTM architectures to capture temporal dependencies and bidirectional dynamics inherent in sensor data. The review further explores the integration of thermal compensation, electrical noise mitigation, and mechanical response optimization within unified frameworks. Comparative analysis reveals that BiLSTM-based approaches outperform conventional techniques in terms of accuracy, robustness, and adaptability. The findings suggest that combining MEMS sensor physics with advanced sequence learning models can significantly enhance system performance. This work provides valuable insights for researchers aiming to develop intelligent, high-precision MEMS accelerometers capable of operating reliably in dynamic environments.