Vapour-Compression Assisted Thermal Management of High-Powered Electronics

Abstract

The research presented in this thesis investigates the evolving thermalhydraulic and energy challenges associated with cooling modern high power Central Processing Units (CPUs), culminating in a unified modelling, optimisation, and system integration framework for next generation active cooling and heat recovery concepts. Through a sequence of experimental, theoretical, and mathematical studies, the work develops a systematic understanding of the interplay between heat transfer performance, fluid-mover penalties, and system level optimisation strategies for hybrid-liquid, vapour-compression refrigeration and heat pump systems.

The first stage of the research characterised the cooling performance of traditional Fan-Fin Air-Cooling (FFAC) and Hybrid Liquid-Air (HLA) systems to characterise their performance and establish a realistic baseline against which all subsequent active cooling concepts could be evaluated. This study quantified the full thermal-hydraulic envelope of both systems, demonstrating the clear limitations of FFAC and the superior performance of the HLA approach, while also showing how coordinated fluid-mover control can minimise power consumption penalties. In terms of pre-emptive engineering for the escalating heat fluxes of CPUs, a major limitation was found to be that of the ambient temperature heat sink. Building on this baseline, the second phase investigated Vapour- Compression Refrigeration (VCR) assisted liquid-cooling, first experimentally and then through validated full system modelling. Experimental results confirmed that miniature VCR-assisted liquid-cooling can deliver sub-ambient coolant temperatures and significantly extend CPU cooling capability while maintaining stable operation across wide ranges of heat load, flow rate, and compressor speed. However, the dominant role of compressor power in governing fluid-mover energy consumption penalties emphasised the necessity of coordinated control strategies to achieve favourable system Coefficient of Performance (COP) levels. To address this, a comprehensive steady-state mathematical model of the VCR-assisted liquidcooling system was developed, validated against experiments, and integrated with a Design of Experiments (DOE) methodology. This modelling framework enabled rigorous parameter sensitivity analysis and system-level optimisation,

demonstrating that properly selected combinations of coolant flow rate, compressor speed, and heat exchanger sizing can achieve target CPU die temperatures with mathematically minimized net power consumption, achieving up to COP = 6.5. The final stage of the research expanded the vapour-compression modelling framework to a dual-purpose refrigeration and heat pump system capable of simultaneously maintaining suitably low liquid coolant temperatures for CPU cooling while upgrading waste heat to ≥ 80°C, making it suitable as a heat recovery system for Direct Air Capture (DAC) applications. A new heat recovery mathematical system model, including a high-pressure ratio compressor and a water-cooled condenser heat exchanger, was developed and embedded into a DOE-driven optimisation workflow. Case studies demonstrated that, with appropriate sizing of the evaporator and condenser heat exchangers and coordinated compressorcoolant control, the miniature system could jointly satisfy CPU thermal constraints and DAC-suitable heat recovery temperatures, achieving COPs between 3.8 and 5.9 under optimised conditions. This work represents the first demonstration of an integrated miniature vapour-compression system simultaneously optimised for CPU cooling and high-temperature heat recovery. Overall, this research provides a cohesive and experimentally grounded framework for analysing, comparing, and optimising advanced CPU cooling architectures. By establishing baseline liquid-air performance, developing validated VCR modelling tools, and extending these to heat recovery applications, the research delivers new insight into the fundamental coupling between heat transfer, thermodynamics, and fluid-mover penalties in hybrid cooling systems. The outcomes offer a pathway toward high performance, energy efficient, and multifunctional thermal management solutions tailored to the rapidly evolving demands of modern high performance computing components and systems.