Online proceedings - EDA Publishing Association

Staggered Cartesian meshes tolerate very high cell aspectratios (e.g. 100:1) without impacting result quality, makingthem efficient at handling thin layered structures like PCBs,heat sink fins, etc.Libraries of behavioral models of common parts, such asfans, heat sinks, chip packages, etc. have an important role toplay in efficiently creating a thermal model of an electronicsenclosure. Some progress has been made in getting suppliersto provide flow/thermal models of the parts they sell, notablyheat sinks, fans, filters, interface materials, typical ICpackages and TECs [18]. Heat sinks were one of the firstparts to become available, but the requirement to customizedesigns for each application has since reduced this need.Today heat pipes are a common feature in many products,with liquid cooling being employed in some applications.These and future cooling technologies such as synthetic jets[19] and piezo fans [20] are expected to lead to an increasedemphasis on validated libraries of parts.Cooling adds cost, weight and volume to electronicsproducts generally without improving functionalperformance. The desire to minimize cooling costs against abackground of increasing thermal density has led to anemphasis on design optimization. Early efforts were focusedwas on fan and vent positioning to improve flowdistribution. As space constraints and power densitiesincreased, heat sink optimization became important tominimize weight, system pressure drop and wake effects.The Cartesian nature of the geometry, use of Cartesianmeshes, and robust solution techniques support fullyautomatedexploration of the design space. Addition,movement and removal of objects coupled with space-fillingDesign-of-Experiment (DoE) techniques with objectcollision detection and optimization techniques makes itpossible to optimize component placement, PCB spacing,heat sink design, etc.Over the last 20 years EC CFD codes have had their owndevelopment trajectory, quite different to that of general-24-26 September 2008, Rome, Italypurpose CFD, driven by the needs of a different target userprofile. As electronics products and the cooling technologiesthey employ continue to miniaturize, new challenges willappear: micro-channel cooling pushes the limits ofapplicability of the Navier-Stokes equations requiring a slipcondition at wall boundaries, and the design of MEMSdevices often requires a multi-physics approach. Vendors ofLattice Boltzmann method codes have also shown an interestin EC.Fig. 6: Local Embedded Fine CFD GridV. ELECTRONICS COOLING CFD: CAUGHT BETWEENMCAD AND EDAFor EC applications, CFD sits at the interface between theMCAD and EDA worlds. A CFD model of a completeelectronics enclosure contains both mechanical andelectronic parts, and so needs information from both theseworlds. By the end of the design process, part details anddesign powers, PCB layout, details of the board structure etc.are all available within the EDA system. However, due tothe largely 2D nature of electronics design, necessarymechanical information about the board assembly is oftenlacking, such as component height.The geometric detail of almost all other aspects of theproduct will exist within the MCAD system. Neither systemcontains information about the thermal properties of thematerials used in the product, nor do they contain behavioralmodels of the parts needed for the analysis, such asresistance networks for packages, fan curves, heat pipeeffective thermal resistance, TEC performance data, etc. orinformation about the product’s operating environment,needed to define boundary conditions for the analysis.VI. THE CASE FOR STAND-ALONE EC CFDIn general, building geometry within a CFD pre-processoris a lot less efficient than using a modern feature-basedMCAD tool to do the same job. However, stand-alone ECCFD has served the industry well. In the early days, limiteddesktop computing power forced significant modelsimplification. The experienced thermal engineers that firstadopted EC CFD created models manually and were capableof making appropriate simplifying assumptions andproviding representative values for any missing data. ECCFD pre-processors handled the historically-Cartesianelectronics geometries well, so models could be easilyevolved as more information about the design becameavailable and refined where indicated by thermal concerns.In many EC applications the geometry is still largelyCartesian, or can be treated as such. Creation of suchgeometry as a pre-processing step within the EC CFD toolremains efficient.Today stand-alone tools are heavily supported bysophisticated interfacing software that can import MCADgeometry in various formats like STEP and ACIS SAT.Geometry can be ‘healed’ to remove small gaps, simplifiedand adapted for analysis, for example by replacing theMCAD assembly for an axial fan with the equivalentbehavioral model from a library with just a few mouseclicks.©EDA Publishing/THERMINIC 2008 4ISBN: 978-2-35500-008-9