Proper Orthogonal Decomposition and Long Short-Term Memory Neural Network-Based Multiphysics Digital Twin Model for Electronic Device Online Condition Monitoring

Digital twin (DT), a virtual representation of a real-world object, has great potential in online condition monitoring and operation management of electronic devices. Although numerical simulation methods, such as the finite element method and method of moment, can accurately evaluate the performance of electronic devices, they are usually time-consuming and computationally intensive, making real-time monitoring challenging. In this work, a multiphysics DT modeling approach for electronic devices is proposed, where the proper orthogonal decomposition (POD) method is first employed to reduce the problem order, and then, the long short-term memory (LSTM) neural network is trained based on the reduced-order model to obtain the DT model. Furthermore, to deal with the difficulty of obtaining the initial physical states for DT model, an adaptive compensation method is developed, enabling the online monitoring to start from unknown initial states. Through building the thermal and mechanical DT models of an integrated microsystem and comparing modeling efficiency and model accuracy with traditional approaches, the performance of the proposed modeling method is verified. Finally, a prototype online thermal monitoring system for a multichip module is developed by combining its DT model with a measurement platform. Results indicate that the model performs well in online condition monitoring of multichip module even starting from a random unknown initial state.