Question and Answers from Web Seminar: Heatsink 201: Even More About Heatsink Design

Q. A. I missed some of the acronyms used in the presentation. Can you recap these? CFD = Computational Fluid Dynamics DoE = Design of Experiment FEA = Finite Element Analysis h = heat transfer coefficient HOPG = Highly-Oriented Pyrolytic Graphite LED = Light Emitting Diode SSL = Solid-State Lighting TIM = Thermal Interface Material How do you model a heatpipe in FloPCB or FloTHERM? We have a SmartPart for that in FloTHERM and FloEFD. Since the effective thermal resistance and limiting heat flux are influenced by the orientation and any bends in the heat pipe, we allow the user to specify these for their specific case. FloTHERM gives a warning if the critical heat flux is exceeded in the solution. Thermal resistance due to spreading is a big problem in a lot of our applications do you have any more info/contacts on embedding copper in an aluminum heatsink? I dont have any particular knowledge of the manufacturing issues regarding embedding a copper slug into an aluminum heatsink. Manufacturers may use other methods, but I know one approach is to bond the copper slug into the heatsink with thermally enhanced epoxy. Id suggest you look at the Buyers Guide on the ElectronicsCooling web site www.electronics-cooling.com and contact the vendors directly. Can you recommend a good reference for DoE? We use a numerical DoE approach its uses a space-filling Latin Hypercube method that maximizes the distance between the points in the design space. Theres a SEMI-THERM paper thats available through IEEE Xplore that talks about the technique and its application to heatsink optimization: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1194333. Assuming that youre going to use this in conjunction with a numerical code, the code vendor should be able to provide information about what they do. We have this information in our Command Center Users Guide on our SupportNet site. For the Infrared radiation, how does the emissivity dependence on the surface finish? Any reference? If by surface finish you mean roughness, Im not aware of any dependence. Generally bright metals have a very low emissivity, the increases somewhat as the metal tarnishes, but this is still lower than anodizing the surface, which increase the value to around 0.8, a typical for many materials. Are there any benefits to using copper clad aluminum? For the base it can certainly help reduce the spreading resistance, but you will only see a benefit if spreading in the base is a significant part of the overall thermal resistance, so it depends on the application. How well does CFD predict performance? Our experience is that for a well-constructed model, the temperature rise is to within 10% of measured values. Generally during design there are far bigger uncertainties about the product, such as material properties, and boundary conditions, e.g. component powers.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

What is the best approximation for calculating the base plate thickness of a plate fin heatsink? The optimum base thickness depends on how much the heat needs to spread. If the heatsink is the same size as a package, having say an exposed metal slug over the entire top surface, then the heatsink base does not need to spread the heat. How is the screw pressure measured so that it doesnt damage the PCB? I dont believe its measured in the application. The screws have a spring to control the pressure, and a long shank (unthreaded part) which controls how far the screw goes into the board. The design controls how much the spring is compressed, and its the spring that applies the pressure. One thing that we used to do with heatsinks was to make the center of the heatsink in the long direction (flow) depressed to try to keep the flow on the heatsink. Do you recommend this? This isnt something Ive tried, but its somewhat counter-intuitive. It would be more logical to have fins close to the heat source that are taller with larger gaps to help direct flow over these hotter surfaces. That said, how the heatsink interacts with the airflow is very dependent on the local environment in the application. Are there any standard materials databases for thermal conductivity? How do highly anisotropic materials like graphene, carbon nanotubes and thin flims like diamond have their inputs? This relates more to thermal interface materials, where theres a lot of research going on. Choosing the best material for a particular situation is quite a challenge. Some caution is needed as vendors measure their data in idealized environments and with quite high pressure, so the actual thermal resistance in the application can be quite different, and usually higher. For a long heatsink, does it help to split the length of the fins half way along to create air turbulence? In general, the idea of staggering fins is to break up the boundary layers that form, so that air that exhausts from the center of the channel between two fins passes over the next staggered fin. The idea is not to increase turbulence, which would increase the pressure drop and so reduce flow through the heatsink. Splitting the fins mid length would have a minimal effect. It would disrupt the boundary layers, at the expense of some of the heat transfer area, but these would quickly reform. Is there much difference in cooling performance between clear anodizing and black anodizing? Both are better than leaving the surface untreated. Black anodized has an emissivity of around 0.8, whereas clear is closer to 0.75, so theres not too much in it. The following is a good source of information: http://www.electro-optical.com/eoi_page.asp?h=Emissivity of Materials Where is Theta-JB obtained? This is a JEDEC standard metric for package thermal performance, measured in a standardized environment a ring coldplate. This data should be available from the package vendor. JEDEC standards are available online. The Theta-JB metric is covered by JESD51-8, published in 1999. How do you normally apply Theta-JC in a CFD model? We build a 2-Resistor model for the package, comprising the Theta-JB and Theta-JC resistances, according to the JEDEC standard JESD15-3. To make it easy to use these and the more advanced DELPHI models, we have a Network Assembly SmartPart that allows an arbitrary network to be represented as a geometric shape that matches the

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

package body and stand-off. FloTHERM PACK generates these models for many package styles. Q. A. Do you know why anodization improves the emissivity so much? Material properties have changed but I did not think they changed that much. Its because it changes the wavelength range that the surface can absorb and radiate (emissivity is wavelength dependent). As its entirely a surface property, only a very thin coating is needed to change the behavior. Most materials (other than metals) have fairly decent emissivities. Snow is interesting, as it reflects visible light very well so doesnt absorb heat from the sun, yet it has an emissivity for thermal radiation close to 1. How do you decide whether forced air flow should be in the same direction as natural convection, or opposite to it? As a general rule, you would always try to force the air flow upwards, so that it augments any natural convection. Anodizing heatsinks are cost drivers, how important is emissivity in forced convection? Radiation is more important for large heatsinks, and when there is a significant gradient across the base, as radiation between the fins helps to spread the heat, but it is a minor effect. Id be inclined to decide based on the analysis how important it is, just by changing the emissivity attached to the heatsink in the CFD model. Ultimately analysis is used to drive down the cost of a product, and the cost and time spent in design, including physical prototyping. A material with a k that is 8 x Al and with the same weight seems to overwhelm all the secondary effects you presented with similar costs? Correct? A material thats the same cost as aluminum, same weight, but has a thermal conductivity of 1600W/mK would be an absolute winner in terms of reducing the thermal resistance within the heatsink. Thats IF the thermal conductivity is isotropic If its orthotropic, as it is in HOPG for example, there are issues getting the heat into the base and then turn from spreading in the base to travel up the fins. What remains is the airside resistance, increasing the importance of getting the airflow design right. How do we calculate the pressure drop in the flat plate fin heatsink. For very simple situations there are correlations you can use, as mentioned in our Heatsink 101 webinar. In more complex situations involving bypass, then there are methods that solve a network of flow resistances to get the overall flow and pressure drop through a heatsink. For everything else CFD is the only option. How to find out h(x,y) on fin surface? This is something thats calculated locally as part of a CFD simulation, and so can be accessed as part of the result post-processing. This level of detail is only available from a 3D CFD simulation. For identical aspect-ratio fins in axial flow applications, is there any benefit (pressure/h) between plate fin and interrupted elliptical fins (not pin fin, but hybrid forged elliptical fin) Generally, what Ive seen is that elliptical fins reduce the pressure drop, but also give a higher thermal resistance than both plate and pin fin heatsinks. This may explain why they dont appear to be that popular. How do heat pipes works? A heat pipe is an evacuated tube, filled with a small amount of liquid like water or ethanol. As its in a vacuum, applying heat to one end causes the liquid to boil, and condense where the heat pipe is cooler, so they are very effective at transporting heat. Typically they are

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

used to connect a high-power component to a remote heatsink, where its easier to remove the heat. Q. A. How does CFD assume the length element in calculating the transition from laminar to turbulent flow? Generally this is controlled by the turbulence model and the treatment used for wall friction. There are various ways to do it. One is to base it on the distance from the nearest wall, which generally works well for electronics applications. A more sophisticated approach is to solve for variables that represent the kinetic energy of the turbulence and its dissipation rate. However, the results on a coarse mesh can be worse. Both approaches are available in our software. How would you use a model in FloTHERM to obtain CFM through the heatsink? The easiest way to get the flow through a heatsink in FloTHERM is to co-locate a volume region with the finned region of the heatsink. FloTHERM will report flows through each of the faces of the volume so you can see how much flow goes in and out of each face. I want to know if FloEFD can be used for heatsinks? FloEFD can certainly be used for heatsink design and optimization. Being CAD-embedded we use the capabilities of the CAD system to support design variant analysis. Alex showed some examples of heatsink design with FloEFD in our Heatsink 101 webinar. Are rough fins a benefit in natural convection? Id say no. Roughness does increase heat transfer at the start of the fins, but the boundary layers quickly grow so that the roughness has no effect. Its then contained within the laminar sub-layer, so the surface is hydraulically smooth. It will have more of an effect in a highly forced convection situation where pressure drop is not a concern, but then Id argue that youre better off adding more fins rather than making the fins rough. What is the approximate emissivity of a copper heatsink? If its bright copper thats not surface treated its very low, at around 0.03. Oxidized black this increases to around 0.8. What is your opinion on compressed-stacked fin heatsinks for high power IGBTs? Stacked fins work like folded fins, with the interlocking helping keep the fin spacing, plus they can be made quite long. If you have a very high power application and need a long heatsink, then this might be the best fabrication choice. If a CFD model is not available for an existing product and cost optimizations are being sought, would an experimental method be appropriate or would CFD be the next step? Even though the product already exists, and presumably works satisfactorily since youre looking for cost improvements, Id be inclined to create a CFD model. The main reason for that is the insight CFD gives into the flow and heat transfer inside the system. Ultimately cooling is a system-level issue, and using CFD has been likened by one of our longstanding customers at Simclar as like having X-ray vision. You may find ways to save cost that you dont expect! Most of the heatsink cost is driven by manufacturing process (extruded, bonded fin, swaged fin, etc). Can CFD modeling provide a guide to which heatsink manufacturing process to use? Not exactly. What you could do though is include parameters in the DoE and optimization that ensure the heatsink is manufacturable by a particular method. Extrusion is typically the cheapest, so you could ensure that the fin height is say no more than 10 times the fin width.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Vapor chambers vs heat pipes; is there a tipping point where one makes better sense than the other? Again, this will depend very much on the application. The advantage of a vapour chamber is that it will help spread the heat in 2D whereas heat pipes tend to be more 1D in their operation, although less so when encased in aluminum in the base of a heatsink. We use a thermal pad between a DSP and a housing. There is a clamp load applied to the pad, but thickness tolerance on the pad is large. We can predict pressure with FEA. How much pressure is generally safe to apply to an IC device on a PCB? Well, you could go back to the package vendor and find out if they have any data, but I doubt it would get you very far. Id sidestep the problem and underfill the package with epoxy as part of the PCB assembly process. Being a leaded package there are two advantages. First, it will improve the heat transfer to the PCB, reducing the temperature rise of the package relative to the PCB, and second it will mean you dont have to worry about the pressure you apply to the gap pad damaging the assembly Is Indium modelled differently than thermal grease? Id generally aim to model an Indium TIM material as having a physical thickness in the model, so theres at least one grid cell in the material which will allow the analysis tool to capture any heat spreading in the TIM itself, which may be important (depends on the application) due to its high thermal conductivity. For grease I tend to model that as thin by collapsing the object as it acts as a 1D resistance. However, Im old school. Our software tools are capable of modeling lateral conduction in 2D objects, and with our meshing technology theres little benefit in modeling physically thin objects as 2D anymore. What would be the better approach for precise temp control of discrete components mounted on a base? Rely on the high thermal inertia of the base or isolate them and control the temperature with temp controlled air flow? If I understand correctly, in your situation you have a number of discrete components all mounted onto a single large heatsink and you want to know how to keep them all at the same temperature. A solution to this Ive seen in the past is to taper the length of the heatsink fins from the start of the heatsink to the end, so the fins get progressively longer in the flow direction. You addressed roughness of fins on a plate fin type heatsink, what about "augmentation" features such as cuts/interruptions in the fins thru cuts or other raised features to disrupt boundary flow? Rule of thumb on when they become worse than the original problem? I dont think I can provide a rule of thumb. In cross-flow, pin fin heatsinks offer little or no advantage over plate fin heatsinks, but it does depend on the details of the designs of the heatsinks being compared. Anything that aims to increase heat transfer by disrupting the flow comes at the expense of increased pressure drop. Is it possible to use a single heatsink to cool multiple components on a PCB? If so, can CFD be used to simulate this? Absolutely, this was done very successfully quite a few years back by Sequent Computer Systems in one of their high-performance symmetric multiprocessing systems, ensuring the processors were at the same temperature and therefore ran at the same speed. In some applications, we will "heat sink" a component directly to the PCB copper. Can CFD be used to determine if this is possible or if we need to add a heatsink? Yes, CFD will allow you to see the split of heat down into the PCB and up into the heatsink. We have sophisticated PCB trace modeling in our software that accurately predicts how the heat enters and spreads in the PCB.

Q. A.

Q. A.

Q. A.

Q.

A.

Q. A.

Q. A.

Q. A.

What kind of factor needs to be considered as you mentioned that the two contact surfaces won't be perfectly flat. Is using TIM data efficient enough? TIM vendors measure the thermal resistance of their products using ASTM D5470, which uses very flat, polished surfaces either side of the TIM and a far higher pressure than is found with package-mounted heatsinks. In the application the thermal resistance is usually somewhat higher as the surfaces tend to be rougher and the pressure a lot less. Values can be several time that quoted. Can you estimate degradation over time of a TIM by general material types or applications? No, sorry. However, greases with fillers can dry out over time due to thermal cycling, and then crack. Sheet materials that have some elastic properties are probably a better bet, but its a field that is experiencing considerable innovation, so youd be best advised to go directly to a vendor for advice. Electronics Cooling magazines Buyers Guide is a good place to start. Can you simulate free convection through a fin heatsink with FloTHERM? Absolutely. FloTHERM performs very well in natural convection and we typically find that agreement against experiment is very good for natural convection cases. What about a ceramic PCB? In Europe we see these from suppliers for LED applications. Ceramic and metal core PCBs do tend to be used for high-brightness, or power LEDs and play an important role in spreading the heat before it enters the heatsink. From a purely thermal perspective, metal core boards are generally better. Can FloTHERM handle non-rectilinear components and enclosures? Yes, up to a point, as FloTHERM is enabled with a cut-cell approach for handling angled geometry and has an alternative voxelisation method available via the FloMCAD Bridge Module. If youre interested in handling general CAD geometry, then FloEFD offers the highest productivity of any CFD tool in the market as it operates within the mechanical design environment. The software also has an optional module that includes some of the electronics cooling features that are available with FloTHERM. Any webinars with respect to acoustics/noise in forced convection systems? Not that Im aware of. In electronics, flow generated noise through the system is only part of the issue. Fan noise is very difficult to predict unless you have a great deal of detail about the design of the fan. One very practical step you can take is to make sure that the fan is operating within the recommended part of the fan curve during design, and then rely on the manufacturers values for fan noise dB. If this is too high for your application, then you need to use a larger fan and de-rate the voltage to so it runs at a slower speed. There are fan laws for de-rating fan curves that also include the impact on the fan noise. I need to design a free convection heatsink that will be under solar loading (desert), what finish do you recommend? Do you have thermal data on paint or anodize? Im assuming that you want to minimize solar absorbtion. What you want is something that has low solar absorbtion and high thermal emissivity. I expect that something like a clear lacquer or white paint would be the best option, but youd need to research what would withstand the harsh environment of the desert conditions. Data is available at http://www.solarmirror.com/fom/fom-serve/cache/43.html for both solar absorptivity and emissivity. When using heatpipe assisted heatsinks how critical is soldering the fins to the heatpipes vs. a press fit of the fins over the heatpipes? That would depend on how much interference there is between the fins and the heatpipe. If theres a good contact area between the fins and the heatpipe and compression can be

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

applied, then a press fit may work. The risk is that this will work loose over time, so it also depends on how the product will be used. Q. A. Will you be discussing heat pipes and vapor chambers in the future? We are planning a Heatsink 301 webinar, but have not yet finalized the content, so we may well touch on these. Beyond that we have no immediate plans. When working on LED heatsinks, do I use the same techniques to design an active flat device of say 30 watts? Yes, the same CFD techniques work well, and we do a lot of work on LED cooling, using simulation tools like FloTHERM and FloEFD. We also have thermal and optical characterization hardware that is used by LED vendors and SSL producers to measure correct thermal resistance values using only the heating power, by subtracting the optical power from the supplied power. Are you really saving that much weight/cost with such a small difference in fin height? It really depends on the volume (units shipped). I know of a guy that does nothing but heatsink optimization for a large OEM. It also depends on how important weight is etc. in the application, plus heatsink weight contributes to reliability issues. Which type of fin do you recommend for high power electronics? For example, a power dissipation around 600W? If you can direct flow directly through the fins, then a plate fin heatsink may well prove the best option. Is there a formula that relates the temperature of the fins to the temperature of the air as it passes through the parallel fins of a heatsink? Yes, for simple 1D flow between what are in effect parallel plates, you can apply flat duct correlations for developing flow. In general the flow physics is much more complicated, unfortunately. Isnt emissivity an important property in forced convection? No, not really as the more forced convection you have the lower the radiative loss to the surroundings, since the heatsink temperature is lower. Is h a function of source power? Not if forced convection is strongly dominant. It is in natural convection. As the power is increased, the heatsink gets hotter but this also increases the natural convection over the fins, and hence increases the heat transfer coefficient. This is why for natural convection applications heatsink vendors quote heatsink thermal resistance vs. power. How is "h" used in a formula? Its used to calculate the convective heat loss from the heatsink, with Q=hAT but this can only be used if the effective value of h is either known or can be estimated with reasonable accuracy. The normal approach is to treat the system as a network of thermal resistances, with the convective resistance being 1/hA How much is a good safety margin? That depends on what stage youre at in the design process, and what techniques youre using for design. Assuming youre using CFD you might use as much as 30% early in design, when details of the product are still sketchy, so the model is not that representative of the actual product, reducing this down to 10% in late design just to be sure the product will be in spec.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A. Q. A.

Is there an optimal fin thickness that you could recommend using or is it contingent upon the application? No, it depends on the application. Can you replicate degradation of the heatsink efficiency over time due to dirt/dust build up, in the case of forced air. People have done research into this. The difficulty is knowing how the dust builds up in the heatsink. For very narrow fin spacings the heatsink acts to sieve the dust out of the air, so it can actually block the entrance to the fins. It thats expected to be a problem, then filtering the air going into the system may be necessary. Dont forget to increase the filters loss coefficient to account for it getting clogged with dust. How can the thermal expansion of a heatsink affect the components that are in contact to the heatsink? More important probably for fixed mounted heatsinks. the main effect, as you surmise, is through mechanical stress caused by differential thermal expansion of the heatsink and the package. Large fixed mounted heatsinks are often designed for a specific package and the mounting force controlled by mounting springs. The aim of the mechanical attachment is to reduce the stress on the package interconnect. Directly mounting the heatsink onto the package can result is a worse situation from a reliability perspective, as the package interconnects experiences stress due to the weight of the heatsink and more importantly any shock or vibration. Clipping heatsinks directly to components is only recommended for small, light heatsinks. Will you be discussing thermoelectric devices in future? We dont have any immediate plans to. Their thermal behavior is quite simple and we have a very good representation of these within our software. There is not too much to say about how to represent these. That said, their operation depends on their temperature, which is predicted during a CFD simulation, otherwise it can be difficult to estimate this accurately. Is there a ratio of thickness of fins at the base to the top? You often see a taper on fins, but this is normally a consequence of the manufacturing method. Extruded profiles often have a taper top to bottom, as do cast heatsinks. A heatsink with forced air flow rate. Is it for example better to have more thin fins or less thicker fins? theres an optimum for each flow situation. The choice of fin thickness and spacing will differ depending on the heatsink length, bypass and the approach velocity. If all dimensions and features were equal, would a cast heatsink work better or worse than an extruded heatsink? For the same design of heatsink and the same aluminum alloy, I doubt theres any significant difference in the resulting thermal conductivity. Given the same geometry and conductivity, they should perform almost identically. We are using FloEFD. Would you recommend it for heatsink modeling or is there a better software solution? FloEFD is an excellent tool, especially if youre dealing with more complex mechanical designs. Its what was used for the second example in the Heatsink 101 webinar. What can I do in a case where the heat source is at a higher level than the venting area? Thats something to be avoided as the heat source will be sitting in stagnant air which cant escape from the system so it will get pretty hot. If it absolutely cant be avoided Id look at using copper or some other conductor to conduct heat down to the level of the vent and then use a fan to draw air through a heatsink attached to the copper. A heat pipe wont work very well as the evaporator (heat source) is above the condenser (heatsink) so the

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

wick structure will really struggle to carry the liquid back up to the heat source. Beyond that, youre looking at a pumped liquid system. Q. Accurate heatsink design using CFD analysis is in relationship to the data obtained for the electronic components within a given system CFD model. Have you had any issue obtaining accurate Theta-JB data for electronic components such as the FPGA components within your presentation? I find most component vendors do not provide Theta-JB within their data sheets. Have you experienced this as well? If so, what alternative approach or method do you apply to accurately model these electronic components? I dont think theres a particular reason that vendors are less likely to provide relevant thermal data for FPGAs, as theyre packaged into BGAs, etc which are thermally well understood. The Theta-JB measurement is covered by the JEDEC Standard JESD51-8, which was published in 1999, so you can reasonably take vendors to task over it. You may want to look at using FloTHERM PACK and use it to build your own models, from 2Resistor, increasing in complexity to DELPHI models and full detailed models. How do I use heatsink with a vapor chamber? Any suggestions? The advantage of the vapor chamber is that it is much more effective at spreading the heat than say copper, so its particularly useful for situations where a lot of heat spreading is needed in the base. They work best when the base is horizontal so its easier for the liquid to wick back to the evaporator. Many times we do not have the space for a mechanical device to hold the heatsink down. Is there a recommended epoxy or bonding agent you recommend? We do not use tape. When you say you dont have space I guess you mean you dont have, or havent reserved with a keepout during design, space on the PCB. You can use a clip that clips the heatsink to the package body in addition to an epoxy. Unfortunately glue layers tend to delaminate under thermal cycling. Id be inclined to use one with a low elastic modulus in combination with a clip. The lower modulus will make it more compliant, and the compression from the clip will help stop it delaminating. What type of heatsink would be best for natural convection? If the base of the heatsink is vertical, then a plate fin heatsink should work best. If the base of the heatsink is horizontal, you may well find that a pin fin heatsink will give slightly better performance, being able to take air in from all sides. If we are forced to use COTS heatsinks due to cost, how would you recommend following your optimization DOE? The Command Center has a spreadsheet-like user interface, so you could simply build the designs by hand, changing the input variables to match each heatsink and then simply pick the best one from the results. Alternative you could use the optimization to find the best design, then find the closest design to it from whats available. You may need to have an extruded heatsink cut down in terms of width or height for example. The optimization will tell you where youre trying to get to. If the design calls for heatsinks on components in close proximity would l be better off with a blanket heatsink or individual heatsinks for each component? It depends on the sizes of the components, heatsink weight etc. Using a blanket heatsink can make contact with the components, which will all be slightly different heights off the board, quite challenging. Have you had much success with "T-Wing" or copper tape type heatsinks? Yes, we modeled one of these a few years back as a cooling solution for a large PQFP, so a relatively low thermal performance package. It did a good job of getting heat into the air.

A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Q. A.

Would you expect to see a difference in performance of heatsinks with an increase in clamping force? Yes, as it will generally reduce the interfacial resistances for the thermal interface material, but you will need to be careful as the force will stress the package interconnect.