Multi-objective placement optimization of power electronic devices on liquid cooled heat sinks

Abstract

The widespread use of power semiconductors has given rise to a host of thermal design issues. Traditional cooling techniques, such as natural or forced convection air-cooling, are inadequate at such power levels. Liquid cooled heat sinks or cold plates are increasingly used. Along with the development of novel thermal management techniques, there is also a growing interest in thermal design methodologies. One key issue facing the packaging designer is the selection of an appropriate cold plate and optimal placement of components on it. This study investigates the multi-objective placement optimization of power electronic components on liquid cooled heat sinks. The two main components involved are an optimization algorithm and a heat transfer solver. A multi objective genetic algorithm (MOGA) (Narayanan and Azarm, 1999) is chosen as the optimizer. The actual heat transfer in the system is usually complex due to the presence of multiple materials and coupled thermal paths and may require time intensive 3D heat transfer solvers which are inefficient and impractical in this optimization framework. We are concerned with the primary heat transfer path from the device junction and extending to the system ambient. Reduced thermal models, or compact models, are thus more suited for accurately but inexpensively capturing the multimode nature of the heat transfer in this rapid calculation framework and are thus implemented as the heat transfer solvers. Two methodologies for developing reduced thermal models capable of handling coupled convection and conduction, and their implementation within an optimization framework are discussed.