Preliminary application of machine-learning techniques for thermal-electrical parameter optimization in 3-D IC

Abstract

Three-dimensional (3-D) integration technique, a promising integration technique, can increase system density but at the cost of increased thermal and power density, leading to thermal-related problems. Design of three-dimensional integrated circuits and systems requires considerations of temperature and gradients observed across the die, because temperature gradients can vary the delay of clock paths. As we need to analyze a large number of parameters for thermal-electrical design, optimization of those parameters becomes important for achieving efficiency and accuracy. Machine learning methods have been applied in the past for artificial intelligence, data analysis, and for general optimization problems. In this paper we propose the application of machine learning methods for parameter optimization in 3-D systems.