Multi-material cold plates for data center cooling: Thermomechanical design, fabrication, and testing

Growing demand for artificial intelligence is driving the development of graphics processing units with greater thermal design power and die size, posing critical challenges for data center thermal management. Lidless direct-to-chip liquid cooling with cold plates reduces thermal resistance by eliminating the integrated heat spreader but introduces reliability concerns due to the coefficient of thermal expansion (CTE) mismatch between the metallic cold plate and silicon die. This work introduces a bimetal cold plate that combines a low CTE copper-tungsten (CuW, 90% W and 10% Cu) base with high thermal conductivity copper (Cu) fins. We investigated nine cold plates with three different fin geometries made from Cu, CuW, or bimetal. Thermal-hydraulic performance of all nine cold plates was measured at 1 kW heat dissipation on a 75 mm × 75 mm heater area with single-phase water flow rates ranging from 0.5 to 3.8 liters per minute. The diamond-shaped pin fins with 0.5 mm spacing have the best thermal-hydraulic performance, with 6.9 K/kW in Cu, 9.0 K/kW in CuW, and 7.8 K/kW in bimetal at 9 kPa pressure drop. Variable focus microscopy was used to quantify cold plate deformation under realistic mounting constraints at an 80°C surface temperature. Measured cold plate deformation ranges from 43 µm to 102 µm depending on material, with CuW cold plates deforming 17% less than Cu on average. The bimetal cold plates exhibit geometric sensitivity, with the straight fins deforming 43–61% more than the diamond-shaped pin fins. Results suggest that bimetal fin design may provide a mechanism to engineer multifunctional cold plates for direct-to-chip cooling.