Abstract
Rapidly expanding computing, storage, and networking requirements have increased the quantity and energy density of modern data centers. Air-cooled high-performance servers often require low air-supply temperatures as well as high air-flow rates, making air-cooling inefficient above certain thermal design power levels. Single-phase liquid immersion cooling (Sp-LIC) addresses this challenge by providing a much higher thermal mass and a high percentage of heat dissipation owing to the direct contact of dielectric fluids with all powered components in the server. It also improves reliability by shielding the ITE from pollutants and hostile environments, and it lowers OpEx by eliminating fans and computer room air handling units. Unlike direct-to-chip liquid cooling, Sp-LIC does not need a complex liquid distribution manifold design as the dielectric liquid can be in direct contact with all components of a server, making it an ideal choice for hyper-scale, edge, and modular data center applications. There are some unique thermal challenges to implementing Sp-LIC such as determining whether to use natural or forced convection and customizing heat sinks that were designed for air-cooling that need to be optimized for dielectric fluids with higher fin efficiency and component reliability (active and passive), which must be addressed and thoroughly researched. These challenges can be addressed using CFD simulations including “Multi Design Variables” and “Multi-Objective Function” Optimization with “Constraints” to help in the design of appropriate extended surfaces and cooling approaches. Experiments are done to verify the CFD models for inlet temperatures greater than 40°C. Also, there are very limited studies related to the reliability of such cooling technology. The accelerated thermal cycling (ATC) test given by ATC JEDEC is relevant just for air cooling but there is no such standard for immersion cooling. The ASTM benchmark D3455 with some appropriate adjustments was adopted to test the material compatibility because of the air and dielectric fluid differences in the heat capacitance property and corresponding ramp rate during thermal cycling. Material characterization findings such as modulus, CTE, creep, fracture toughness, elastic modulus, stress relaxation, cracking, dislocation nucleation, and the viscoelastic properties of the samples will be compared before and after immersion using a TI-980 Nano-indenter, Dynamic Mechanical Analyzer (DMA), Thermomechanical Analyzer (TMA), and Scanning Electron Microscopy (SEM). In all, a comprehensive guide to Sp-LIC for the thermal management of hyper-scale data centers will be presented.
