Multi-objective optimization of 3D printed liquid cooled heat sink with guide vanes for targeting hotspots in high heat flux electronics

Abstract

The data center industry is currently faced with surging energy demands and a growing carbon footprint. Beside these, the industry is also faced with increasing chip power densities and thermal stresses, which can be attributed to the integration of multiple cores into a single chip. This paper presents a novel hot spot targeted cooling approach, the goal of which is to reduce both the cooling device's thermal resistance and chip thermal stresses while also lowering the pressure drop and pumping power required to cool the chip. The approach allows for the temperature of the liquid leaving the chip to be maximized, which can enable different heat recovery potentials. Multiple techniques were used to improve the heat sink's thermohydraulic performance, including a direct chip-attached heat sink, distributed water jet impingement inlets, and embedded guide vanes. A multi-objective optimization study was conducted to minimize the chip's temperature non-uniformity, thermal resistance, and pumping power. For optimization, a Bayesian-based learning technique was coupled with a genetic algorithm, which was then employed. The results show that the optimal design was comprised of variable fin density and profile in the hot spot and background regions of the chip. The optimized design showed a benchmark thermal resistance of 0.036 °C/W and a chip non-uniformity index as low as 0.81 °C. It outperforms the other hot spot targeted heat sinks that have been presented in previous literature, meanwhile also consuming relatively low pumping power.