High heat flux immersion cooling in data centers: Quantitative bypass control and fluid-structure compatibility across dielectric fluids and heat sink designs

초록

High-density artificial intelligence workloads have caused escalating heat fluxes in data centers, posing thermal risks that demand advanced thermal management and high energy efficiency. Single-phase immersion cooling systems have been extensively researched, yet existing investigations have been limited to heat fluxes below 30 W & sdot;cm 2, a range that leaves its high heat flux regime and the quantitative impact of fluid-structure interactions and thermal enhancers unexplored. To bridge these voids, this study investigated single-phase immersion cooling under chip-level heat loads exceeding 700 W per chip. The open ratio is introduced as a quantitative metric for bypass control, and its effects on thermal resistance and pumping power are evaluated using FC-40 and GC-5X across plate fin and metal foam-assisted configurations. Despite superior thermophysical properties, GC-5X recorded a chip temperature 23.0 K higher than FC-40 under unconstrained flow. Open ratio reduction from 100% to 0% decreased thermal resistance by 45.6% for GC-5X, 7.5 times greater than FC-40, at the cost of a 37.8% pumping power increase. In metal foam configurations, GC-5X reached 112.3 degrees C at open ratio 100% yet recovered to 59.3 degrees C at open ratio 0%, while FC-40 decreased from 57.3 degrees C to 46.7 degrees C. The metal foam configuration at open ratio 0% demonstrated the most stable thermal performance, with FC-40 below 79 degrees C up to 70 W & sdot;cm 2 while GC-5X approached this threshold at 50 W & sdot;cm 2. Ultimately, these findings demonstrate that thermal performance is governed by the interplay between fluid properties and flow control, providing a quantitative design basis for immersion cooling systems under next-generation high-power conditions.

키워드