Intelligent Thermo–Electro–Mechanical Modeling of MEMS Devices Using Deep Fusion Learning

Abstract

Micro-Electro-Mechanical Systems (MEMS) devices operate under complex multiphysics interactions involving thermal, electrical, and mechanical domains. Accurate modeling of these coupled phenomena is essential for improving device reliability, performance, and design optimization. Traditional finite element methods (FEM) and analytical models often struggle with high computational cost and limited adaptability for nonlinear MEMS behavior. This study proposes an Intelligent Thermo–Electro–Mechanical Modeling framework for MEMS devices using Deep Fusion Learning (DFL-TEMM). The proposed approach integrates deep neural networks to fuse thermal, electrical, and mechanical feature representations for accurate predictive modeling of MEMS behavior under varying operating conditions. The model leverages multi-branch neural architectures to independently learn domain-specific features and a fusion layer to combine them into a unified representation. Performance is evaluated using prediction accuracy, RMSE, computational efficiency, and generalization ability. Experimental results demonstrate that the proposed model significantly improves accuracy and reduces simulation time compared to traditional FEM-based approaches.