Thermal-Stress Analysis of 3D-IC Based on Artificial Neural Network

Three-Dimensional Integrated Circuits (3D-IC) is a new processing technology emerging in the field of microelectronics in recent years. Through Silicon Via (TSV) technology is used to connect different wafers vertically. Due to the increasing number of transistors in 3D-IC, its heat dissipation and thermal-stress mechanical problems are increasing. These problems can cause performance degradation of components even fail. When the chip is working, the working conditions of various components are constantly changing. The thermal stress of each point is also changing. It is very difficult to get real-time data, so the dynamic thermal stress management technology will be severely limited. The paper mainly proposes a method for quickly predicting the maximum thermal stress at various points in the work of TSV-based 3D-IC chips. Firstly, the finite element simulation software COMSOL is used to establish the overall chip finite element model of TSV based 3D-IC. TSV temperature and chip thermal stress were calculated to find the maximum thermal stress near each TSV and obtain the temperature distribution within a certain range of TSV. Then, an RBF neural network model is built through sample training based on the derived temperature and thermal stress data. RBF neural network model is used to provide some information support for dynamic thermal stress management technology of working chip. Through the extraction of data characteristic information, the thermal stress around TSV is predicted accurately, which provides an accurate and efficient solution for dynamic thermal stress management when the chip works.