A Fast Thermal Simulation Method Based on the Asymmetric Extended Krylov Subspace

Effectively addressing thermal concerns is essential for accurately predicting the performance and dependability of electronic systems. However, the discretization of heat conduction equations in VLSI devices, that is needed for accurate thermal simulations, tends to produce very large-scale thermal networks. As a result, the extraction of a dynamic compact thermal model (DCTM) via model order reduction, even if it is a onetime calculation, also requires time-consuming computations. In this paper, we propose a moment-matching (MM) methodology that utilizes the Asymmetric Extended Krylov Subspace (AEKS) for the fast extraction of DCTMs, by leveraging the sparsity structures of the discretized equations. Experimental results on several benchmark circuits demonstrate that the proposed methodology can significantly reduce runtimes of the extraction process compared to an established MM method, by introducing a negligible overhead in the reduction error.