NUMERICAL INVESTIGATION ON EFFECT OF TARGET COOLANT DELIVERY IN LIQUID-COOLED MICROCHANNEL HEAT SINKS

Rising demand for high-performance processors and increased difficulty in their thermal manage-ment have resulted in the need for advanced and efficient cooling technologies like direct-to-chip liquid cooling. Typical CPUs/GPUs today have multiple cores, and at any given time, not all the cores in the package are utilized, which creates a nonuniform heat distribution or a thermal gradient across the processor. This can lead to reliability issues due to localized thermal transients and cyclic thermomechanical stresses. A practical solution to solve this issue is investigated in this paper by proposing a dynamic cold plate and analyzing its thermal performance using computational fluid dynamics (CFD) modeling. The proposed cold plate design consists of four different fin sections. The flow rate to each of these fin sections is regulated passively using bimetallic strips that respond to the outlet temperature of the coolant. A cold plate that can dissipate a maximum of 360 W was first designed and optimized using multiparametric optimization in optiSLang. Based on the anticipated heat loads, a bimetallic strip was selected and analyzed for maximum deflections using finite-element analysis using a range of cold plate operating temperatures. The results show that the proposed cold plate design can reduce the thermal resistance by a maximum of 42%. Furthermore, a maximum reduction of 62% was observed in the temperature difference value of the power sources as compared to the baseline design.

Automated summary

The rising demand for high-performance processors and increased difficulty in thermal management have led to the need for advanced cooling technologies like direct-to-chip liquid cooling. This study proposes a dynamic cold plate design to solve the issue of nonuniform heat distribution across the processor. By using computational fluid dynamics modeling, the thermal performance of the proposed cold plate design is analyzed, showing a maximum reduction of 42% in thermal resistance and 62% in temperature difference compared to the baseline design.