Hotspots often occur in places where a component, such as a CPU or GPU, is under heavy load and generating a large amount of heat. As chip designs become more complex and transistor sizes continue to shrink, full chip manufacturing becomes increasingly difficult and costly. As a result, semiconductor manufacturers have paid attention to chiplets, which reduce costs by connecting multiple chips into one package. As chiplet technology evolves, there is a growing need for thermal management that can handle multiple hotspots. When multiple hotspots exist, this can make thermal management more challenging because the heat needs to be effectively dispersed from several locations. A vapor chamber (VC) is a highly efficient heat spreading device that is widely used in electronic devices, especially in high-power applications. The VC provides better thermal performance than traditional heat spreaders due to its ability to spread heat evenly across a large surface area.

We developed an ANN-based multi-objective optimization process to predict the performance of the VC with a liquid supply layer, designed for high heat flux applications. We used a numerical model to study the thermal performance of the VC and its structural components. An artificial neural network (ANN) model was developed to predict VC performance based on the surrogate model's results. Our approach resulted in a VC with significantly reduced thermal resistance and enhanced critical heat flux (CHF), primarily due to a well-delineated role division between the evaporator wick and the liquid supply layer.