Optimization of 3D printed liquid cooled heat sink designs using a micro-genetic algorithm with bit array representation

The advancement of additive manufacturing enables significantly greater freedom of design compared to conventional techniques. Of significant interest is the potential improvements in the design of cold plates and heat sinks for electronics cooling. Greater design freedom could enable new designs that reduce thermal resistance and hydraulic resistance, enabling the usage of higher power systems while maintaining an equivalent heat sink volume. Parts made by using additive manufacturing are built layer by layer, and can be conceptualized as a bit array in a 2D plane. Therefore, this work explores the feasibility of combining computational fluid dynamics and micro-genetic algorithms to produce optimum shapes for liquid cooled heat sinks represented as bit arrays. A 50 mm x 10 mm area is discretized into a grid space. An initial bit array geometry is developed using a heat flux dependent probability function based on two representative chip power distribution profiles. One distribution is a symmetric flux map, and the other uses a non-symmetric map, more representative of a real processor. The performance of generated designs is determined using a commercial computational fluid dynamics code, after which an optimization algorithm is applied. Solution ranking in each generation is based on the system entropy generation rate, which is dependent on the fluid pressure drop and overall heat sink thermal resistance. Crossover, mutation and elitism operations are used to create new generations of designs. Results after one hundred generations are compared with a baseline straight finned cold plate to determine the advantages of designing for an unrestricted manufacturing process. Results indicate that the optimization methods reduced entropy generation rate by 26.4% and 21.7% for the symmetric and non-symmetric power maps, respectively.