An Efficient Leakage-Aware Thermal Simulation Approach for 3D-Ics Using Corrected Linearized Model and Algebraic Multigrid.

Chao Yan,Hengliang Zhu,Dian Zhou,Xuan Zeng

DOI: https://doi.org/10.23919/date.2017.7927172

2017-01-01

Abstract:Thermal control has become a great challenge for 3D-ICs due to the ever increasing power density and 3D integration. Among techniques to address the problem, fast thermal simulation approach is basically required to accurately characterize the runtime temperature variations of 3D-ICs. In this paper, we propose an accurate and fast leakage-aware thermal simulation approach for 3D-ICs with consideration of both heatsink cooling and microfluidic cooling. First, the proposed approach is based on a corrected linearized model for leakage power approximation, which is proved to be equivalent to the Newton-Chord method for solving nonlinear algebra equations. A convergence comparison is presented in this paper to show that such approach is more efficient than other methods for leakage-aware thermal simulation. Second, an aggregation-based algebraic multigrid (AMG) preconditioned iterative linear solver is adopted that greatly reduces the computation time for solving the linear equations during calculation, which makes the proposed approach even more efficient. Numerical experiments show that the proposed approach can achieve 8x-139x speedup in comparison with the state-of-the-art methods, and with almost negligible average temperature error no more than 0.025K and maximum temperature error no more than 0.095K.