Long-Term Reliability Test on An Air Assisted Liquid Cooling System

The power density of advanced-node semiconductors used in HPC applications has been increasing. Simultaneously, the adoption of advanced packaging techniques such as 2.5D, 3D, and heterogeneous integration has posed significant challenges for cooling these HPC applications. It has become evident that traditional air-cooling methods are insufficient for cooling HPC systems, necessitating the use of liquid cooling. However, liquid cooling technologies come with their own set of challenges which are currently being identified and understood by the industry. Air-assisted Liquid Cooling (AALC) system is one of the solutions that can be adopted into the data center without the need of the infrastructure-wise upgrade. To gain confidence in this technology, ensuring the reliability and sustainability of this system is of utmost importance. To achieve this, we have conducted an extensive long-term reliability test over one year duration, continuously monitoring the system's operation and performance in an accelerated environment. In this paper, we will discuss the setup of the long-term reliability test, examine the degradation of system performance over time, and explore our methods for monitoring the test while it is in operation.In addition, one notable challenge is the reliability of materials that come into direct contact with the coolant, as it can lead to potential corrosion and material incompatibility issues. This can result in cooling efficiency degradation in the event of microchannel blockages by corrosion byproducts and more critically can lead to catastrophic failure in the affected CPUs, racks, and possibly the whole facility. In this study, we will also show how we identify and address the root causes of issues encountered during this collaborative effort and lesson learned.