A Comprehensive Review of Enhancing Thermo-Electro-Mechanical Responses of MEMS Resonant Accelerometers with an Attention-Guided Siamese Fusion Neural Network

Abstract

Microelectromechanical systems (MEMS) resonant accelerometers have become essential precision sensors owing to their high sensitivity, compact size, low power consumption, and stable frequency-based output. However, their performance is significantly influenced by coupled thermo-electro-mechanical (TEM) effects, including thermal drift, thermoelastic damping, electrostatic nonlinearities, anchor losses, and mode coupling, which limit measurement accuracy and long-term stability. Conventional analytical and finite element methods provide valuable physical insight but are computationally intensive and often inadequate for modeling complex real-world operating conditions. This review presents a comprehensive analysis of recent advances in artificial intelligence-based optimization for MEMS resonant accelerometers, with particular emphasis on Attention-Guided Siamese Fusion Neural Networks (AGSFNNs). The survey examines deep learning techniques including convolutional neural networks, recurrent neural networks, transformer architectures, attention mechanisms, transfer learning, and Siamese networks for temperature compensation, drift prediction, signal denoising, nonlinear correction, and multi-sensor fusion. Comparative analysis indicates a clear transition toward hybrid physics-informed and attention-based learning frameworks that achieve superior prediction accuracy, robustness, and computational efficiency. The AGSFNN paradigm emerges as a highly effective solution by combining similarity learning with adaptive feature selection for robust thermo-electro-mechanical response modeling. Finally, this review identifies current challenges and future research directions toward developing lightweight, explainable, and real-time AI-enabled MEMS accelerometer systems for next-generation precision sensing applications.