Satyam Saini

Graduation Semester and Year




Document Type


Degree Name

Doctor of Philosophy in Aerospace Engineering


Mechanical and Aerospace Engineering

First Advisor

Dereje Agonafer


The last two decades of 2000 have seen an explosion of utilization and reliance on digital technologies and platforms. During this time, electronic devices have undergone intense miniaturization and a corresponding rise in power densities. In parallel to these developments, data center proliferation has been driven by the increasing utility of revolutionary technologies such as Machine Learning, Artificial Intelligence, e-commerce and, the Internet of Things. Data centers, thus, have become the core of the modern digital age. The primary bottleneck, however, due to extensive device miniaturization and data center proliferation has always been thermal management. Many of the aforementioned technologies require high-performance computing CPUs and GPUs platform clusters, that consume large amounts of power and dissipate an equally large amount of heat. Air-cooling has been the most popular method of data center thermal management but is limited to low power processors since the cost and energy consumption to pump air for high power density racks becomes very large. As a result, data center administrators have resorted to energy-efficient air-cooling techniques like Direct/Indirect Airside Economization and Free Air-cooling. Both these techniques have proven to significantly reduce the Power Usage Effectiveness (PUE) of air-cooled data centers in geographies with favorable climatic conditions. Although these technologies aid in alleviating the concerns of high energy consumption they also have an inherent risk of exposing the IT equipment to harmful airborne particulate and gaseous contaminants. While the impact of gaseous contaminants has been well documented in the literature, the same can’t be said for the impact of particulate contaminants. This is because conducting controlled experiments where particles can be generated and behave similarly as in nature is extremely tough. This further makes it tough to assess the risks associated with particulate contamination and also thwarts the development of mitigation strategies. The first part of this research investigates the distribution of these airborne particulates inside the IT equipment and data center space with the help of particle tracking using Computational Fluid Dynamics. This investigation aims at carrying out simplified modeling 3-D and 2-D models of data center space and IT Equipment with known boundary conditions of the airflow. Particle tracking models of a commercial CFD code ANSYS FLUENT are used to simulate the behavior of a known particle mass in the airflow. Based on the literature review on the most pervasive particulates, a range of particle sizes are simulated and tracked using the Lagrangian scheme. Analysis of the most vulnerable locations of particle deposition is made by assessing the regions with higher particle concentrations. Similarly, regions of high particle concentration are identified inside the IT Equipment for different server configurations and different velocity inlet boundary conditions. Mitigation strategies are suggested based on the analysis of particle concentrations for different configurations of IT Equipment. The CFD methodologies presented in these studies can be leveraged during the data center infrastructure design phase and also during the server mechanical design phase. The second part of this study focuses on the design optimization of heat sinks and the analysis of the thermal performance of different server configurations at the tank level in single-phase immersion cooling. It has been seen that when data centers move from air to immersion cooling due to performance demands or energy constraints, they typically use the same server hardware for immersion cooling as it was in air cooling. Using air cooling hardware, especially air-cooled heat sinks in immersion cooling, can become a significant bottleneck in achieving peak performance due to a limit on case temperatures. An in-depth numerical study on different multi-parametric optimization methodologies for heat sinks is conducted using OptiSLang for forced and natural convection for an open compute server design. The objective of this study is to address the heat sink optimization from dif The optimization is done at a constant pumping power and iterating the combination of pressure drop and thermal resistance minimization as objective functions. Heat sink parameters like fin count and fin thickness are varied first for a constant base thickness and then the base thickness is also varied with the other heat sink geometric parameters. This is done for both copper and aluminum heat sinks to analyze the differences between the two heat sink materials and choose the best option from both economic and thermal performance points of view. The optimization results also correlate the dependency of each of the geometric parameters of the heat sink types on the objective functions. It is observed that this dependency varies for both natural and forced convection. The results of this study will help develop standard methodologies for optimizing heat sinks for immersion cooling. The second study in immersion cooling deals with analyzing the impact of flow boundary conditions of the immersion tank on the thermal performance of the immersed servers using CFD. Owing to the low flow rates in immersion-cooled servers, natural convection may play a significant role in cooling the primary heat-dissipating components. Under different flow rate conditions, it becomes extremely difficult to predict the balance between natural and forced convection cooling modes. This study quantifies this balance for different inlet flow boundary conditions and at various CPU power utilization levels. The flow path of the coolant is determined under these boundary conditions for a shadowed and a spread-core version of a commercially available high-power server. The temperature and velocity fields obtained using this numerical study will be, in the future, compared with the experimental results using Particle Image Velocimetry for validation.


Data center, Thermal management of electronics


Aerospace Engineering | Engineering | Mechanical Engineering


Degree granted by The University of Texas at Arlington