Physics-Driven Digital Twins for Two-Phase Cooling in AI Data Centers (SmartCool)

The rapidly increasing energy density in AI industries is pushing for a major paradigm shift in cooling solutions from traditional air cooling to direct-to-chip liquid cooling. Hereby developing energy effective cooling. Two-phase immersion cooling (IMC) can capture all the waste heat efficiently with a supreme heat transfer coefficient. However, the phase change process is associated with stochastic dynamics of microscale bubbles and a critical point for maximal heat flux. SmartCool is aiming for controllable and predictable operations of IMC processes by establishing a framework connecting fundamental multiphysics models, advanced microfabrication of boiling surfaces, and stochastic greybox modeling. The greybox models will improve our understanding of physical mechanisms of microscale heat transfer, at the same time, a squeeze-shift-adapt principle will be established for robust industrial implementation of IMC processes. The outcome of the project will pave the way for a broad spectrum of two-phase cooling solutions.