Reduced Order Modelling for the Thermodynamic Design of Electronic Systems

The continued growth in information technology has resulted in a demand for electronic systems that provide higher computational performance at reduced power consumption. At the individual user level, portable electronic devices have increased functionality while duration of battery life is also considered an important selection criterion. At the service provider level, performance must also increase to handle remote computations and communication services at minimal operational costs. Therefore, energy challenges associated with information technology sectors is across all scales of implementation. A significant contributor to the total energy demands of electronic systems is the requirement to maintain individual components at reliable operating temperatures. Historically, thermal management has been viewed as secondary to the design process. However, the importance of electronic device characteristics such as performance, volume and energy efficiency has converged which signifies that thermal management can no longer be considered at the latter design stages. The introduction of computational fluid dynamics in the design process has assisted, however a number of challenges remain. One major challenge is the multiple length scales that are involved. At the component level alone this spans from tens of nanometres to tens of millimetres. In electronic systems such as datacenters, the overall range of length scales can be 10-12 orders of magnitude which make the prediction of junction temperature impractical without introducing simplifying assumptions that can adversely impact the outcome. Another significant challenge is addressing the transfer of information between equipment suppliers and integrators. Thermodynamic optimization by equipment integrators is often restricted to crude representations due to the limited information provided by suppliers to preserve intellectual property. In this presentation, these challenges will be discussed and approached using a reduced order modelling methodology for thermal-fluid systems that include complex non-linear effects. The method is shown to reduce simulation time, provide reasonable accuracy across a range of scales and facilitate the exchange of thermal information. In addition, the use of model reduction to perform informed optimization and design improvements is highlighted. Finally, the future needs for developing a strategic framework to employ multiscale modelling of thermal-fluid systems incorporating reduced order modelling is discussed.