Machine Learning for 3D-IC Electric-Thermal Simulation and Management

Abstract

Thermal management for 3-D ICs is not only important but also challenging. While air-cooled heat sink is agreed to become incapable for 3-D ICs, microchannel cooling has provided a better solution. In this paper, a machine learning method, Bayesian Optimization (BO), is applied in 3-D ICs with a time-dependent power map to intelligently control the flow rates of the tier-specific microfluidic heatsink (MFHS) for dynamic thermal management (DTM).