In Situ Health Monitoring for 3-D Heterogeneous Integrated Systems: Embedded Current Sensing and Adaptive Self-Diagnosis Approach

The rapid progression of micro-system technology has given rise to heterogeneous integration technology. This innovative packaging solution provides new opportunities for system-level miniaturization and multi-functionality, surpassing the limitations of traditional integrated circuit technology and rejuvenating the development trajectory of the semiconductor industry. However, the complex interconnection structure of micro-system integration intensifies mutual interference and coupling among circuits, making the testing of internal critical signals increasingly challenging. Moreover, the advanced functionalities and superior performance of these micro-systems necessitate heightened precision in testing. Traditional external testing methods struggle to accurately, comprehensively, efficiently, and in real-time monitor and diagnose the internal health status of micro-systems. To address these challenges, this paper proposes a customized heterogeneous integration technology that incorporates embedded AC/DC sensors. By monitoring the current output of key chips and the data from interface circuits, the status information of current signals in the micro-system during operation can be acquired. Based on this information, the operation mode and fault state can be accurately assessed. The diagnostic performance for all modes was evaluated using six machine learning-based models: Naive Bayes Model (NBM), Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), CNN-LSTM, and Ensemble models. Notably, the Ensemble, LSTM, ANN, and CNN-LSTM models achieved perfect accuracy across all three data splits (training, validation, and test). The remaining NBM and CNN models also demonstrated strong performance, achieving accuracies exceeding 97 percent on the validation and test sets.