Session: Session 02-05: Two Phase Cooling - II

Paper Number: 173209

173209 - Hollow Micropillar Evaporative Cooling for Direct-to-Chip Cooling of High-Power Density Gpus 

The exponential growth of artificial intelligence (AI), machine learning, and other high-performance computing (HPC) applications has driven unprecedented demands for computational power in modern data centers. This surge has accelerated the development of new, high-power graphics processing units (GPUs) that dissipate intense thermal loads, exceeding several hundred watts per chip. As conventional air-cooling systems struggle to meet these increasingly demanding thermal requirements, direct-to-chip liquid cooling is being adopted to enhance reliability and achieve energy savings. However, on a future-looking trend, energy efficiency (Datacenter PUE < 1.02) and high heat fluxes (> 200 W/cm²) are becoming important. To address these challenges, two-phase evaporative cooling has emerged as a promising solution, offering high heat transfer coefficients, near-isothermal performance, and minimal pumping power compared to traditional cooling technologies. In particular, direct-to-chip (D2C) two-phase cooling architectures—where a dielectric working fluid evaporates directly at the chip surface—enable efficient heat removal at the source while allowing for modular scalability and fluid compatibility with sensitive electronics.

In this present study, a novel direct-to-chip evaporative cooling (DCEC) cold plate is introduced that utilizes a silicon hollow micropillar device to achieve evaporative cooling. This structure's micropillar device acts as a precision-manufactured membrane that allows liquid and vapor separation with low surface tension liquids like refrigerants and dielectric fluids, by pinning the liquid at the sharp edges. This allows thin-film evaporation with an exit quality of 1, thus pushing the thermodynamic utilization of the refrigerant to the maximum even at low flow rates. Additionally, the micropillars create a significantly high evaporation surface area due to liquid pinning, leading to higher evaporative heat transfer. This study presents the design, implementation, and performance characterization of this hollow micropillar two-phase evaporative cooling system tailored for direct-to-chip thermal management of high-power GPUs. Emphasis is placed on evaluating the evaporator-to-condenser thermal resistance with varying flow rates, heat loads, and operating pressures, to meet the thermal demands of next-generation AI and HPC infrastructure. The thermal resistance variation with flow rate and heat loads will be correlated to different modes of operation (single-phase cooling, thin-film evaporation, and boiling). Preliminary modeling shows a maximum heat flux of 413 W/cm² at a superheat of 20℃ and a base temperature of 70℃. This corresponds to a pressure drop of about 12 kPa with a total power dissipation of 5 kW over an area of 7.5 × 7.5 cm². The findings contribute to the growing body of research focused on two-phase evaporative cooling solution development and its implementation in data centers.

Presenting Author: Vivek Manepalli University of Maryland- College Park

Presenting Author Biography: Vivek Manepalli is pursuing his PhD at the University of Maryland, College Park in the mechanical engineering department. He earned his B.E. in Mechanical Engineering and M.S. in Physics from Birla Institute of Technology and Science, Pilani in India. Following that, he completed his M.S. in Mechanical Engineering from the University of Colorado, Boulder, in the field of electronics packaging and thermoelectrics. His research is focused on the development of novel two-phase evaporative cooling technology for GPUs and CPUs, and microchannel liquid and hybrid cooling for power electronics devices.