Geometry-engineered heat sinks for enhanced cooling performance in immersion-cooled servers

This study introduces a novel geometry-engineered dual-CPU heat sink framework for single-phase immersion liquid cooling (SPILC) servers, addressing the limitations of uniform designs that ignore the distinct thermal-hydraulic conditions of upstream and downstream CPUs. The originality lies in independently optimizing each CPU's fin height, thickness, and spacing using high-fidelity CFD and Response Surface Methodology, enabling CPU-specific thermal management in SPILC systems. A total of 54 Box-Behnken simulations are conducted to quantify the effects of six geometric variables on average CPU temperature ( T a ), inter-CPU temperature difference (Δ T ), and pumping power ( PP ). ANOVA results reveal fin height as the dominant factor across all responses, while fin spacing shows secondary significance and fin thickness has minimal impact. Flow-field analysis demonstrated that tailored geometries greatly enhance coolant penetration, suppress recirculation, and improve temperature uniformity. The optimal configuration ( h 1 = 19.52 mm, h 2 = 25 mm, t 1 = 0.90 mm, t 2 = 1.50 mm, s 1 = 4.83 mm, and s 2 = 5 mm) achieves T a = 51.26 °C, Δ T = 0.173 °C, and PP = 90.68 × 10 −5 W, confirming the effectiveness of CPU-specific optimization. A comparative coolant evaluation further shows that oils provide the lowest temperatures while fluorocarbons reduce pumping power, with FC-40 offering the best balance. Overall, this work establishes a new pathway for geometry-adaptive heat sink design in SPILC servers, enabling significantly enhanced thermal performance, uniformity, and energy efficiency.