Generative Design of Liquid-Cooling Channels for Thermal Management of 2.5D and 3D Integrated Advanced Packaging

This study develops a physics-guided generative design method for cooling a 2.7 kW package containing two GPUs and one CPU. A conditional diffusion model proposes symmetric channel layouts based on temperature and pressure-drop targets. Connectivity and branching checks screen the layouts before a calibrated reduced-order thermal-fluids model evaluates their performance. Analysis of 5,000 candidates identifies a design with a predicted maximum GPU temperature of 70.30 degrees Celsius and pressure drop of 89.72 kPa. Independent three-dimensional conjugate heat-transfer simulations in OpenFOAM assess the selected design, illustrating how reduced-order modeling can guide channel generation while reserving expensive CFD for validation.