Manufacturable Micro Heat Sink Designs Through Design Filtering and Macro-Scale Optimization

Thermal management of electronic devices concerns the removal of heat from components to prevent premature failure and improve reliability. Current cooling strategies are under continuous pressure to deliver adequate performance, and ongoing component miniaturization and compact packaging techniques have driven the need for further innovation in the field of thermal management. To this end, topology optimization has increasingly been applied to improve the cooling performance of single-phase liquid cooled micro heat sinks. This is achieved by determining the optimal placement of solid material and fluid channels within the heat sink. However, topology optimization is not always able to ensure that an optimized design is manufacturable. This dissertation investigates topology optimization for micro heat sinks, with a particular focus on methodologies which improve design manufacturability. Design manufacturability is evaluated based on how well the size of solid material and fluid channel features in an optimized design can be controlled, and based on how much non-physical material remains present in optimized designs. To improve manufacturability, two methodologies are explored. The first methodology is design filtering in traditional topology optimization, where the fluid-solid material distribution is optimized, and a filter is used to control the feature size. The second methodology is macro-scale topology optimization, where the size distribution of an a priori chosen solid structure is optimized, and an interpretation is given to non-physical material, while also reducing the computational cost of optimization. First, design filtering techniques are studied, which include sensitivity filtering, density filtering, projection, and robust optimization. For each method, the manufacturability of the topology optimized micro heat sink design is assessed, as well as how different filter parameters influence manufacturability. Sensitivity filtering is shown to modify the optimization problem in such a way that optimized designs are inferior to those obtained with other filter methods. Density filtering is shown to introduce non-physical material into optimized designs, which can partially be eliminated when projection is included. Robust optimization is shown to be the most capable in improving the overall manufacturability of optimized designs, and providing adequate control over the feature size. Second, this dissertation introduces a new methodology known as macro-scale topology optimization, which is comparable to homogenization-based topology optimization. Macro-scale optimization is shown to be capable of ensuring manufacturability of the optimized design, due to the way in which the design problem is formulated and interpreted. Macro-scale optimization is explored in the context of both hydraulic and thermal-hydraulic micro heat sink design. To validate the macro-scale optimization methodology, macro-scale simulations are compared to results from direct numerical simulation, for designs containing solid structures obtained at various stages of the optimization. The comparison indicates that the macro-scale modeling procedure overall is a decent approximation of the physical behavior in the designs, except in regions where the local changes in the size of adjacent solid structures are large. Macro-scale optimization is further shown to require much less computation time in comparison to direct numerical simulation. The macro-scale optimization methodology is proposed as a computationally cheap, initial estimator for design improvements, which can be used to replace or complement optimization based on direct numerical simulation.

Jaar van publicatie:2021

Toegankelijkheid:Open