Titolo della tesi: Thermal management strategies for contemporary and next-generation computing devices - A Computational Fluid Dynamics Study with High Fidelity DNS Simulations for Coupled Conjugate Heat Transfer
The exponential growth of computational demand, driven by advances in Artificial Intelligence (AI), scientific simulation, and data analytics, has made efficient thermal management an urgent engineering challenge, particularly in High-Performance Computing (HPC) systems. With processors reaching power levels exceeding 1200W, the heat flux density has surpassed the capabilities of traditional cooling technologies, necessitating novel solutions to maintain system stability, energy efficiency, and performance.
This doctoral thesis addresses the critical issue of thermal management in advanced computing architectures by employing high-fidelity Conjugate Heat Transfer (CHT) simulations and experimental validation. Using the open-source CFD platform OpenFOAM, this work presents a multi-faceted investigation into thermal control strategies, aligned with the broader objectives of Italy’s Piano Nazionale di Ripresa e Resilienza (PNRR) and focused on industrial applicability and digital-ecological transition.
The core of this research comprises three interconnected studies. First, a comprehensive study of microchannel cooling systems for HPC applications is undertaken. Through Direct Numerical Simulations (DNS) and experimental validation, the effects of fin geometries, pin arrangements, and flow disruptions are analysed. A full-scale microchannel heat block is fabricated and tested, offering design insights with real-world applicability. Furthermore, in collaboration with Gemateg srl, the integration of such systems with thermoelectric modules is explored to enhance thermal control capabilities further.
Second, the thesis briefly examines a bistable system driven by displacement flows inspired by bistable natural ventilation dynamics. DNS is used to explore complex fluid behaviours, such as flow flipping, with potential applications in edge computing and passive thermal regulation for remote sensors and IoT devices.
Finally, a short outlook is provided on the foundational principles of cryogenic cooling systems. This is primarily informed by the experience gained during a secondment at CERN, which allowed for the study of these systems in a high-energy physics context. This section serves to explain the nature of that collaboration and to highlight the broader applications of such ultra-low temperature technology.
Collectively, this thesis presents a methodology for designing and optimising next-generation thermal management systems. It offers practical engineering solutions for immediate application in HPC, insights into passive systems for low-power electronics, and preliminary frameworks for future technologies. These contributions aim to enable higher CPU/GPU performance by safely and efficiently pushing thermal limits, thereby supporting the continued evolution of computational infrastructure.