Reviewing Cooling Strategies for Post-Dennard Era Computing and Modern Electrothermal Phenomena

Abstract

Excessive thermal runaway typically manifests itself in computer component damage and other various negative side effects. As a mitigation strategy, end-users and consumers typically implement differing methods of component cooling, including fan-heatsink cooling, heatsink-only cooling, and liquid-based cooling. Different cooling methods remain impactful to modern computing, as it remains a core component in the thinking of thermal design and reliability engineering in the realm of semiconductor devices. The industry-wide acceptance of the end of Dennard scaling and the imminent end of Moore’s law are major factors that are currently impacting CPU power consumption trends and modern cooling philosophies. Transistor packing and process refinement is beginning to push against atomic boundaries in combination with phenomena such as leakage current and high current density, causing a general trend of increasing temperatures generation-to-generation in microprocessors. As a result, thermal mitigation strategies and protections must be in place to reduce damage and catastrophic failure while increasing performance of the die package. In a high-heat scenario, liquid-cooling can provide up to 38% to 48% improvement over fan-heatsink variants depending on the type of workload executed by the processor. Fan-heatsink cooling faces thermal resistance limitations in the form of spreading and air convection resistance as a result of heatsink material composition, the resistance along the path of heat flow impeding conduction rate, and the lower thermal conductivity of air compared to liquid. Currently, the best performing variant appears to be liquid-based cooling while fan-heatsink combinations provide adequate levels of thermal dissipation based on these observations.