Machine learning based surrogate models for microchannel heat sink optimization

Microchannel heat sinks are an efficient cooling method for semiconductor packages. However, to properly cool increasingly complex and thermally dense circuits, microchannel designs should be improved expanded on. In this paper, with secondary channels ribs investigated using computational fluid dynamics coupled a multi-objective optimization algorithm determine propose optimal solutions based on observed thermal resistance pumping power. A workflow that combines Latin hypercube sampling, machine learning-based surrogate modeling is proposed. Random forests, gradient boosting algorithms neural networks were considered during the search best surrogate. We demonstrated tuned can make accurate predictions used create acceptable model. Optimized show negligible difference in overall performance when compared conventional approach. Additionally, calculated one-fifth of original time. Generated attain temperatures lower by more than 10% under same pressure limits as convectional design. When limited temperature, drops reduced 25%. Finally, influence each design variable power was employing SHapley Additive exPlanations technique. Overall, we have proposed framework has merit viable methodology sink optimization.