Here is the abstract you requested from the Thermal_2018 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|The Necessity for Thermal-Electrical Multiphysics for Board Heating in a Server Rack Unit|
|Keywords: Multiphysics, Multi-board, Thermal|
|Often in forced convection engineering system simulations, an engineer can get by with a simplified version of a PCB board. In these cases, surface mounted heat sinks and relatively high flow rates cause most of the heat to transfer directly into that heat sink and into the surrounding air flow. While this may work when the current load is well distributed or relatively low in most regions of the board, there are instances in industry where concentrations of current flow due to large-draw components creates a resistive or Joule heating effect in areas of the board metal, increasing the total power that needs to be dissipated in the system. If the total generation of power in the board layers can be calculated, this value can also be added as a general distribution of power on the board, but this lacks the specific areas of high concentration. In addition, modern boards are specifically designed for optimal usage and copper distributions and layers are not necessarily equally distributed. As such, the heat can be drawn through the boards along traces and through vias in very specific flow patterns. In this paper, we show a comparison of four separate modeling techniques of the same rack unit system. All four models are created and run in ANSYS Icepak, an electronics-oriented CFD analysis tool. Initially, we run the boards with orthotropic material properties for conduction and no heat generation in the interior of the boards. This gives a base-line or worst-case heat transfer for the hottest components. All power input in the system is in the surface-mounted components. Secondly, we run the model again with traces and vias mapped through the volume of the boards creating localized conductivity on a layer-by-layer basis. This shows how a redistribution of metal content will affect maximum component temperatures and the effects of different components on one-another through the board. Third, we re-run the initial problem with a 3D distribution of total internal power distributed through the board to show the effect on the final temperatures. This power is calculated directly from a DCIR run in the ANSYS SIwave electromagnetics tool. Lastly, we add the localized power maps from ANSYS SIwave directly into the detailed copper layers of the boards in ANSYS Icepak. This will concentrate the power to the interior areas of the boards that have greater generation rather than dissipate the heat throughout the entire volume. The connection between ANSYS SIwave and ANSYS Icepak also allows the products to pass information back and forth in a steady-state analysis. This full system with multiple boards is run in ANSYS Icepak. When the first run completes, the temperatures from the individual boards are mapped back into their respective SIwave projects. The projects run in a queue and each adjusts the resistivity of the metal properties based on the localized temperatures calculated in Icepak. When completed, all power maps are re- read into the Icepak design and the loop continues until a specified convergence in temperature. This test shows how powers will distribute and increase differently when taking into account the multi-physics effect of Joule heating on a high-current board model and why it’s more accurate to take all aspects into account.|
|Jared Harvest, Lead Applications Engineer