Immersion two-phase liquid cooling of data cards

• Main Author: Lioewanta, Sebastian Markus
• Other Authors: Fei Duan
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Mechanical engineering::Fluid mechanics
• Online Access: http://hdl.handle.net/10356/65830
• Institution: Nanyang Technological University

The development of data centres requires fast and effective heat removal from the processors in the data cards. As with most electronics, excess heat is detrimental to the performance of the equipment and as such, cooling systems for electronics are very important and an area that has been often considered in development of hardware. Traditional air cooling has long been the method of choice for heat removal, however, designing and maintaining it can be complex and costly. Furthermore, with the exceeding rate of improvement in performance of electronics, traditional air cooling may not be able to remove the excess heat quickly enough. As such, more efficient methods have to be looked into. This report looks into the possibility of two-phase immersion cooling as a means for thermal management of such electronic systems. Instead of simulating the electronic components with heaters, an actual printed circuit board (PCB) is used as the heat generating component. The PCB is immersed in the commercial engineering liquid Novec 7100 inside a sealed chamber with a condenser unit. Various PCB component temperatures are taken under different conditions such as varying levels of CPU utilization and number of cores being ran. Initial tests showed that the two-phase immersion cooling system managed to maintain temperatures below maximum junction values, however, they were considered as high temperatures especially when compared to normal air cooling values. It is postulated that the high temperatures yielded are due to high wall superheat values and a lack of microcavities to initiate the nucleate boiling regime. The addition of a heat sink during immersion cooling vastly reduced temperature levels due to the increase in surface area in contact with the liquid. However, there was still a lack of vigorous pool boiling due to the high boiling point of the Novec 7100. Further studies can be made into reducing wall superheat values while increasing critical heat flux and heat coefficient values, and also look into different component orientation and the choice of dielectric liquid used.