Log In   |   Sign up

New User Registration

Article / Abstract Submission
Register here
Register
Press Release Submission
Register here
Register
coolingZONE Supplier
Register here
Register

Existing User


            Forgot your password
December 2005
library  >  Application Notes  >  Ake Malhammar

Component or PCB Oriented Thermal Design


the physics of thermal design is sometimes difficult to understand, particularly for air-cooled pcbs. the complexity is one reason but a good understanding also requires combination of knowledge about both electronic hardware and heat transfer theory.


this article brings up some issues that, in the author’s experience, can cause a great deal of confusion if not properly understood.

 

fig_1_577.

figure 1. the component and the pcb oriented approaches.

 

two opposite approaches


an air cooled pcb offers two major flow paths for the generated heat. the heat can take the chip-component surface-air path and it can take the chip-component leads-pcb-air path. both these flow paths are important and they also interact in a complicated way. large variations in component size, heat generation, lay out and pcb design does not make the problem any simpler. it is apparent that a theoretical approach to the problem must be based on considerable simplifications.

 

two extreme simplifications are possible, the component approach and the pcb plate approach. the former looks at the components as an array of heated blocks from which all heat is dissipated. the latter looks at the pcb as a thermally conductive plate with heat sources, see figure 1. neither of these two lines of attack have the intention to fully map all heat flow paths but rather to create a set off from which further refinements can be made.

 

the strategy of the component approach is to map the component-to-air convection as accurately as possible and then introduce compensations for the fact that some heat also is dissipated from the pcb. a typical experimental set up for this line of work is heated aluminum blocks mounted on a balsa wood plate.

 

the pcb plate approach takes the opposite strategy. the tactic is to map the pcb-air convection as accurately as possible and then introduce compensations for the fact that some heat also is dissipated from the components. since convection from flat plates is a rather well known subject, this line of work focus on the disturbances caused by the propounding components. this approach has attracted little attention.

 

it is, therefore, somewhat difficult to perceive a typical experimental set up. aluminum blocks mounted on an isothermal plate have, however, been used.

the major factors that favor the component approach are large components and high air velocities. the major factor that favor the pcb plate approach is a high thermal conductivity in the pcb. one could, therefore, be tempted to believe that the development tendencies in modern hardware have favored both approaches approximately equally.

 

this is however not the case. the balance shifted radically towards the pcb plate approach when the copper content in the plates started to increase and this development commenced more than 20 years ago. the first major impact was the shift from 1 to 2 copper layers. the second major impact was the shift from 2 to multi-copper layers.

 

fig_2_606.

figure 2 - a small strip of free pcb space around a component can increase
the available heat transfer surface radically
.


the reason why the pcb plate is so important is that even a relatively small strip of free pcb space around a component can increase the available heat transfer surface radically, see figure 2. it can be objected that a pure surface comparison does not reflect the actual heat transfer conditions correctly.

 

compared with the pcb the component has a higher temperature, has a higher heat transfer coefficient since propounds from the surface and has a larger heat transfer surface than its projected surface. but, even if these impacts are given a factor 4, which is reasonably realistic, the pcb will still dissipate the major part of the heat.

 

fig_3_477

figure 3. air velocity has a much more radical impact on the fan power requirement
and the noise intensity than it has on the heat dissipation
.


the component approach and air velocity

 

a high air velocity favor the component approach. this raises a question. will the current development towards higher heat densities result in a velocity increase such that this approach becomes more realistic?

 

basic physics can help to clarify the subject, see figure 3. this diagram shows how the heat dissipation, the fan power and the noise intensity increase with the air velocity. a typical telecom application currently operates at 1.5 m/s air velocity and the fan power requirement is typically 5 – 10% of the heat dissipation.

 

as shown in figure 3 only a small air velocity increase is needed to radically change this ratio. for much larger velocity increases one can even imagine a situation where the fan power equals the heat dissipated. the factor that in most cases limits the air velocity is nevertheless the noise intensity. it increases extremely sharply with the air velocity.

 

to increase the air velocity is apparently only a realistic option in the low end of the scale. the step from natural convection into forced convection on the 1.5 m/s level was relatively easy. further moderate increases are probably possible but any radical changes can not be expected.

 

it can be argued that figure 3 only is valid if the size of the airflow channels are unchanged and that higher air velocities also should be accompanied by larger flow channels. this is a good argument but the community of electronic hardware designers have a collective “horror vacui neurosis”, (fear of empty spaces), which makes such changes difficult.

 

it is therefore not likely that future cooling problems will be solved by a radical velocity increase. it is much more likely that the surface enhancements, which currently take the form of heat sinks on individual components, develop into large pcb scale heat sinks.

 

it can be concluded that the component oriented thermal design approach apparently is a blind alley. this does not mean that all research efforts on this theme are useless. many of the results can probably be quite valuable for other purposes.



figure 4. velocity profile for flow over a flat plate and the displacement boundary thickness definition.

 

the flat plate approximation

 

to bring up the issue of whether a pcb from the thermal point of view can be simulated as a flat plate or not, may come as a surprise to many. the component side of a pcb is after all a very coarse surface. for a heat transfer engineer it is nevertheless a natural question. the reason is that the heat transfer community, from its early days and forward, has tried to attack the opposite problem: how to enhance the heat transfer from a flat surface by introducing surface disturbances.

 

the results from this line of work are not encouraging when the air velocity is moderate. in most cases these efforts result in small gains, typically below 30%, even if larger gains can be achieved under special circumstances.

 

the impacts of propounding disturbances on a flat plate are of two kinds. they can increase the convection by stirring up the air and they can enhance the heat transfer surface. this will however only happen if the heights of the disturbances are above a certain critical limit. the microscopic roughness of the pcb plate is obviously below that limit but what about the components? apparently there is a need for some kind of critical roughness reference. a further discussion about this subject will require some insight into boundary layer theory.

 

when a flow of air hits a surface, the velocity at the surface quickly descends to zero, which provokes the velocity profile near the surface to change radically, see figure 4. as the flow progresses downstream the disturbed zone successively spreads outwards. this zone is called a boundary layer. it is somewhat difficult to define its thickness because there is no elbow in the velocity profile curve that could serve as a natural landmark. the fact that there are at least 3 different definitions is a reflection of this difficulty.

 

the most common boundary thickness definition is the displacement definition. it is defined as the possible virtual displacement of the surface that could be made if the velocity profile had been uniform and the passing flow rate unchanged, see figure 4. the thermal boundary layer definition takes another approach. it defines the thickness as the depth of a still layer of air that results in the same thermal resistance as the heat transfer coefficient.

 

the velocity at the top of a boundary layer is of some interest. for the displacement definition it is approximately 55% of the full speed velocity and for the thermal definition it is 70%. an alternative way to evaluate a boundary layer definition is to compare the average kinetic energy in the layer with that of the full speed velocity. these values are much lower, for the two definitions above, 11% and 20% respectively.


fig_5_482
figure 5. average displacement and thermal boundary layer
thickness for a 200 mm long plate.

 

figure 5 shows the average boundary layer thickness for the displacement and the thermal definition on a 200 mm long flat plate. typical components have a height of 2 – 4 mm. had the plate in figure 5 been a pcb one would therefore expect that the disturbances caused by the components are negligible at low air velocities and that they have some impact when the air velocity approaches 2 m/s.

 

this is indeed also what is indicated by the few measurements on isothermal plates with component-like metal block disturbances that have been made. the convection enhancement approximately increase linearly, from zero at low velocities up to about 20% at 2 m/s. since most pcbs have components only on one side, the typical total impact at 2 m/s should be of the order 10%.



figure 6. components that propound above the boundary layer can cause
a vertex street that enhances the convection.


one must however be aware that components which propound form the boundary layer can cause a vertex street that enhance the convection downstream considerably, see figure 6. this impact is predominately important near the air inlet where the boundary layer thickness is small.

 

it can be concluded that although a pcb visually appears as a very rough surface, this roughness only seems to have a moderate impact on the heat transfer coefficient. this is a very important conclusion because it opens up the possibility to make estimates without any detailed knowledge of the pcb lay out or the component sizes. such estimations are very important in the early phases of the design process.


the cfd approach

 

a new line of attack to the air-cooled pcb calculation problem appeared about 10 years ago, the cfd approach. the basic idea behind this method is to use finite element or finite difference theory to calculate the heat transfer coefficient on each spot of the pcb. it is an attractive methodology since none of the simplifications in the analytical approaches are needed.

 

the results that can be achieved with this method are very good. for well-controlled conditions the error of the chip temperature prediction is on the 10% level. considering all other uncertainties associated with thermal design this is a much better performance than is needed for the vast majority of applications.

 

only about 50% of the temperature difference between a chip and the incoming air to a pcb is however used for convection. if the error on the chip level is 10% it is therefore probably higher on the convection level. it can thus not be claimed that the cfd approach is radically better than the pcb oriented approach, at least not for applications where the inlet airflow has a reasonably uniform velocity profile.

 

it may be surprising to some that the cfd approach, which seemingly does everything correct, not can produce perfect forced convection predictions. there are well known difficulties with these types of programs such as determining the impact of the element sizes and other user defined calculation parameters but this is not the profound reason. the profound reason is that laminar and turbulent flow often coexist in pcb applications.

 

everyone has observed that the jet which comes out of a water tap is crystal clear when the flow rate is low and that it abruptly changes into a milky appearance when the flow is increased. a near by conclusion would be that fluid flow either is laminar or turbulent and that the intermediate zone in between these two flow types is very limited. this way of looking at the issue is not totally wrong. it works very well for long and narrow channels but it does not work for flow between parallel plates of pcb size.


figure 7. flow between parallel plates is in an important velocity range characterized
 by both laminar and turbulent flow.


the flow between parallel plates of pcb size is a typical inlet flow. that is, the velocity profile undergoes quite large changes and is never fully developed when it exits. the flow in each cross section can nevertheless be either laminar or turbulent. turbulence is provoked when the product of the air velocity and the boundary layer thickness exceeds a certain critical value. the instabilities always start near the outlet and progresses upstream when the velocity is increased, see figure 7.

 

the flow between parallel plates is therefore both laminar and turbulent within an important velocity range, 1 – 3.5 m/s for typical pcbs. most textbooks would nevertheless categorize this type of flow as turbulent. this is why the term “weak turbulence” has been used in figure 7. the bulk of forced convection pcb applications is unfortunately found within the limits of this type of difficult flow.

 

the problem with cfd calculations is that the algorithms used are unable to determine whether the flow is laminar or turbulent. one has to manually choose between either one of them. there is in addition also a menu of several different turbulence models. calculations for the weak turbulence flow type will therefore always be a problem for this kind of software.


figure 8. the choice of the appropriate turbulent model for flow between
parallel plates can be difficult.


conclusions

 

the component approach, which assumes that the major part of the heat is dissipated from the components, has very little physical relevance for modern copper rich pcbs.

 

the pcb plate approach, which assumes that the pcb dissipates the major part of the heat, is much more realistic. although the surface of a pcb visibly may appear as rough, this roughness only has a moderate impact on the heat transfer coefficient when the air velocity is below 2m/s. the flat plate approximation is therefore reasonably good. this conclusion is important because it opens up the possibility to do estimations in the early phases of the design process, before many non-reversible decisions are taken.

 

the cfd approach, which strives to calculate the heat transfer coefficient in each spot of a pcb, results in temperature predictions that are much better than needed for most standard applications. simulations on the air flow types that typically are found between forced convection cooled pcbs are however problematic. chip temperature predictions with an error smaller than 10% are therefore difficult to achieve.
 


about ake malhammer




 

 

 

 

 

 

ake obtained his master of science degree in 1970 at kth, (royal school of technology), stockholm. he then continued his studies and financed them with various heat transfer-engineering activities such as deep freezing of hamburgers, nuclear power plant cooling and teaching. his ph.d. degree was awarded in 1986 with a thesis about frost growth on finned surfaces. since that year and until december 2000 he was employed at ericsson as a heat transfer expert. currently he is establishing himself as an independent consultant.

 

having one foot in the university world and the other in the industry, ake has dedicated himself to applying heat transfer theory to the requirements of the electronic industry. he has developed and considerably contributed to several front-end design methods, he holds several patents and he is regularly lecturing thermal design for electronics.



to read ake's web site for more thermal information and software tools he has developed, please visit http://akemalhammar.fr/ - see more at: https://www.coolingzone.com/library.php?read=534#sthash.y3rcxrow.dpuf

to read ake's website for more thermal information and software tools he has developed, visit http://akemalhammar.fr/.

Choose category and click GO to search for thermal solutions

 
 

Subscribe to Qpedia

a subscription to qpedia monthly thermal magazine from the media partner advanced thermal solutions, inc. (ats)  will give you the most comprehensive and up-to-date source of information about the thermal management of electronics

subscribe

Submit Article

if you have a technical article, and would like it to be published on coolingzone
please send your article in word format to [email protected] or upload it here

Subscribe to coolingZONE

Submit Press Release

if you have a press release and would like it to be published on coolingzone please upload your pr  here

Member Login

Supplier's Directory

Search coolingZONE's Supplier Directory
GO
become a coolingzone supplier

list your company in the coolingzone supplier directory

suppliers log in

Media Partner, Qpedia

qpedia_158_120






Heat Transfer Calculators