A numerical simulation model of embedded liquid microchannels for cooling 3D multi-core chips is established. For the thermal management problem when the operating power of a chip changes dynamically, an intelligent method combining BP neural network and genetic algorithm is used for distribution optimization of coolant flow under the condition with a fixed total mass flow rate. Firstly, a sample point dataset containing temperature field information is obtained by numerical calculation of convective heat transfer, and the constructed BP neural network is trained using these data. The “working condition–flow distribution–temperature” mapping relationship is predicted by the BP neural network. The genetic algorithm is further used to optimize the optimal flow distribution strategy to adapt to the dynamic change of power. Compared with the commonly used uniform flow distribution method, the intelligently optimized nonuniform flow distribution method can further reduce the temperature of the chip and improve the temperature uniformity of the chip.

Embedded liquid cooling is a preferred solution for dissipating the heat generated by high-power chips. The cooling capacity and pump power consumption of embedded liquid cooling heat sinks differ significantly between different structures. To achieve an accurate match between cooling capacity and heat dissipation requirements, the selection of a liquid-cooled heat sink should be carefully considered in conjunction with the heat dissipation needs of heat sources in real-world thermal management issues. Based on the manufacturing limitations on chip temperature and microchannel pressure, a composite performance index function was developed to assess the cooling capacity and cooling cost of the heat sink. This allowed for the establishment of an evaluation standard to determine the suitability of embedded liquid cooling and heat sink for the heat source. In this study, the suitability of four microchannel heat sinks with the same feature length and fin volume was evaluated under various thermal load conditions. The results show that the best-suited heat sink changes with variations in the thermal load of the chip. In the example, when the heat source was homogeneous at 100 W, the circular section pin fins have an optimal suitability of 0.928 for Re = 500. When the heat source was a heterogeneous heat source with a power of 100 W, the value of Θ was found to be 0.389. Additionally, the optimal suitability of drop section pin fins for Re = 971.5 was determined to be 0.862.