Fast Stress Analysis for Runtime Reliability Enhancement of 3D IC Using Artificial Neural Network

Lang Zhang1, Hai Wang1, Sheldon Tan2
1University of Electronic Science and Technology of China, 2University of California at Riverside


Heat dissipation and the related thermal-mechanical stress problems are the major obstacles in the development and commercializing process of 3D ICs. Dynamic thermal manage- ment (DTM) techniques can be used to alleviate such problems and enhance the reliability of 3D ICs. However, the time varying stress information is hard to obtain at runtime which limites the effectiveness of DTM. In this paper, we propose a fast stress analysis method for runtime usage. The new method builds artificial neural network (ANN) model by training offline using thermal and stress data. Next, the ANN model is used to generate important stress information, such as maximum stress around each TSV, for DTM methods at runtime. In order to improve the stress estimation accuracy and speed, specially designed input selection plans are proposed and implemented for ANN model generation. Experiments with different configurations of ANN models show that the new method is able to estimate important stress information at extremely fast speed with good accuracy for runtime 3D IC reliability enhancement usage.