Design and intelligent optimization of TSV-based embedded microchannel heatsinks in 2.5D Packaging

With the continuous advancement of integrated circuits, chip density and miniaturization have increased significantly, resulting in a substantial rise in on-chip temperatures. Customized thermal management strategies tailored to different chiplet packaging configurations have become crucial. At present, microchannel heatsinks (MCHS) are widely employed to improve heat dissipation in multi-chiplets packaging. In order to improve the heat dissipation capability of multi-chiplets packaging, embedded MCHS has been extensively studied in recent years. However, embedded MCHS will inevitably affect the routing within the interposer. In this study, we propose an innovative heat dissipation approach that integrates MCHS within TSV(Through-Silicon-Via) interposer. By embedding TSV into the walls of the MCHS, this method significantly enhances heat dissipation while maintaining routing integrity. The novel structure in this study is defined as T-MCHS. Firstly, we verified that the proposed design exhibits better heat dissipation capability and a higher friction factor compared to the traditional structure. Subsequently, the structure is optimized using an enhanced NSGA-II (Nondominated Sorting Genetic Algorithm II) algorithm. To enhance computational efficiency, the simulation model is simplified during the optimization process, and an ANN(Artificial Neural Network) surrogate model is constructed based on simulation data to replace time-consuming simulations, thereby enhancing optimization efficiency. Finally, performance validation is conducted through extensive simulations. The optimized T-MCHS demonstrated significantly enhanced heat dissipation performance while effectively reducing pumping power. Compared to other jobs, this design is able to achieve a greater convective heat transfer coefficient by sacrificing less pumping power. Additionally, stress uniformity across the substrate is improved, contributing to a significant enhancement in overall reliability.