Investigation of Flow Restrictors for Rack Level Two-Phase Cooling Under Nonuniform Heating

Conventional air-cooled systems in data centers are failing to address the increasing power densities brought by high-performance computing. Pumped direct-to-chip two-phase cooling offers a potential solution to thermal challenges in next-generation data centers owing to the high heat transfer coefficients and heat capacities associated with flow boiling. However, there are technical challenges to address such as flow maldistribution caused by nonuniform heating in highly parallel two-phase loops. Flow restrictors can be implemented to suppress the maldistribution, however, there is a need to better understand the requirements and performance cost of their use. To this end, this work analyzes flow maldistribution and restrictors in a server-rack level two-phase loop with 34 sleds cooled in parallel. A numerical system-level model is developed which relies on an empirical correlation to capture the pressure drop response of the flow components in the server sleds. Pressure drop data are collected experimentally at various heat inputs and flow rates to calibrate the empirical correlation. The resulting model is used to evaluate flow maldistribution and pressure drop cost of various flow restrictors. In the absence of restrictors, maldistribution due to nonuniform heating causes dry-out in cold plates even though the overall flow rate is adequate to dissipate the total heat input. Orifice restrictors which have quadratic scaling of pressure drop and mass flow rate, suppress the maldistribution below an acceptable limit at the cost of additional pressure drop twice as much as that of the sled itself. Restrictors with high order mass flow rate-to-pressure drop scaling, such as flow regulators, can suppress the maldistribution with significantly smaller pressure drop penalty at the cost of design complexity.