Electronics Weekly’s guide to LED heatsinking
In the early days of power LEDs, many lighting firms rushed out solid-state luminaries only to find them failing in service.
The problem was heatsinking. Manufacturers were used to filament bulbs that run hot and are cooled by radiation, and fluorescent tubes that have a huge surface area from which heat convects.
Conductive cooling, the mainstay of electronics, was a black art.
That was a decade ago now, but the danger of unreliable product is still here, particularly as cost has to be low and power LEDs are being squeezed into smaller and more demanding applications such as MR16 down-lighters.
What are the basics of LED cooling? “It is thermal management, you can do calculations from thermal resistance. We do those on a regular basis,” Lumileds application engineer Michiel Kruger tells Electronics Weekly.
Thermal resistance (R?), measured in kelvin per watt (K/W), is the traditional metric used to determine how hot the junction of a semiconductor will get.
You simply add the junction-to-mounting-face R? of the LED package to the R? of the heatsink mounting face to the ambient air. If there is a PCB in between, you add that R? as well.
From the total thermal resistance, you can back-calculate the junction temperature from the ambient air temperature at a given LED dissipation.
But even at this basic approximation level, there are complications.
“The real question is how to take temperature into account as this affects the performance of the LED, so it is not a straight-forward call,” says Kruger. “LED forward voltage [Vf] goes up with power, and Vf goes down when the die gets hot. With just thinking the LED dissipates 1W, you miss a number of factors. If you don’t do it correctly, you may over-design.”
So at the minimum, the simple approach becomes iterative as junction temperature affects light output and the input power has to be changed to correct this.
And this is with a heatsink of known characteristic, in a plentiful supply of ambient air.
So what happens when custom heatsinks and restricted airflows are stirred into the mix?
“The devil is in the detail,” says Kruger. “It is pretty straight forward to do back-of-the-envelope calculations if you assume a certain resistance for the PCB and a certain resistance for heatsink. It is a good starting point. You probably have to do 3D simulation to get further.”
Mentor Graphics has a 3D computational fluid dynamics (CFD) tool called FloTherm that calculates heat conduction, as well as radiation, convection and air flow. “When it is an easy situation, the thermal resistance approach works,” Mentor Graphics marketing manager John Parry told Electronics Weekly.
He introduces one of the things that spoils the apparent elegance of the thermal resistance approach: “When heat is getting near the maximum level, spreading resistance is one of the challenges to include between the LED and heatsink.”
R? in standard heatsinks is measured by introducing heat evenly over its entire mounting face. If the hot device is any smaller than the whole mounting face, heat has to move sideways before it can reach the dissipation area. This additional flow resistance is the spreading resistance.
“The idea of adding up thermal resistance doesn’t work very well. Thermal resistance only applies in the exact situation the heatsink was measured in,” says Parry. “Spreading resistance gets worse the closer a heat source gets to a point source. It looks very simple, and can easily get you into trouble.”
With experience, spreading resistance – which is a function of area ratio and material thickness – can be accommodated to a first order, and there are graphs to help, but already things are getting complicated.
This is where 3D modelling can remove guess-work as well as allow for restricted air movement and custom heatsinks.
CFD input data is the physical geometry of each component in the system, plus the thermal characteristics of each material and their interfaces. The flow and thermal characteristics of air are included, and anything that restricts that flow also needs to be drawn in.
To simplify data input, FlowTherm plugs into several mechanical CAD packages, from where it can get most of this data automatically.
There are other CFD choices: Lumileds, together with distributor Future Lighting Solutions, has a thermal design package for LEDs called QLED.
To illustrate the advantages of CFD, Parry takes the example of a circular heatsink with radial fins – the arrangement that is becoming standard in LED downlighters.
“The simple view of heatsinks is ‘the more fins the better’, which is true up to a point, because it increases surface area,” he says, “but the problem with more fins in a fixed volume is decreased air gap and decreased air flow.”
So there will be an optimum fin thickness and spacing for each heatsink material in each application.
According to Parry, the only practical way for someone not steeped in analytical thermal calculations to find that optimum is to do a ‘parametric study’.
A parametric study involves modelling multiple situations with different numbers of fins and fin thicknesses – say 30 – and plotting the results on a ‘response surface’ – a 3D graph where the two horizontal axes are fin number and fin thickness, and the height is junction temperature.
If the response surface has a dip, that is the optimum number/thickness combination. If the dip is a valley trending down in a particular direction, more simulations are needed in that direction. If there is a dip, but not much of one, the number of fins and thickness is not that critical.
The computation is not trivial. “For an LED light fitting in a ceiling, or lamp on a desk, it takes something like an hour for one simulation. You can set a batch running on multiple machines overnight for a resp onse surface,” says Parry.
On the other hand, you could make a batch of different mechanical prototypes, which is increasingly difficult with the smooth curving geometry and complex shape of modern designs, but made easier by rapid prototyping techniques.
“Batches of runs are generally the argument for computation,” says Parry, pointing out that time optimising heat spreading and conduction in the ‘stack’ is seldom wasted: “If you can take a few cents out of an LED light bulb, its really does warrant putting that effort into design.”
In designing a stack, “the more you can spread heat before it gets to the heatsink the better, but you need a stack-up that meets your cost objective. You have to think about this in 3D”, he adds.
As an example of ‘Rolls-Royce’ heat spreading, Parry cites some telecoms laser diodes where the die sits on a synthetic diamond heat-spreader, on a copper second-level spreader, on the aluminium package base.
In LED lamps, as most power LEDs are surface-mounted, some sort interface is needed before the heatsink.
This can be a metal-cored PCB (MCPCB), a ceramic substrate, or even an FR4 PCB riddled with thermal vias – the latter of which can hit a creditable 3-4K/W if done well, according to Lumileds’ Kruger.
All have different thermal conductivity and different coefficients of expansion, and picking the wrong expansion coefficient can ruin reliability in products that undergo a lot of thermal cycling.
For limited runs, custom heatsinks may not be necessary. “There are off-the-shelf circular LED heatsinks and heatsinks for strips of LEDs,” says Paul Ward, opto-products manager at distributor Farnell.
And for large runs, there is an alternative to aluminium heatsinks.
“There’s a new thermally-conductive plastic that can be moulded. You can form the body and heatsink from the same material,” says Ward. “Philips has done this in one of its MR16s.”
He holds out some hope for people who do all the design optimisation and still find there is not enough natural convection to cool their designs: “Nuventix has a very clever way of generating turbulent flow using magnets and a diaphragm.”
Called SynJet, they are small DC-powered capsules that puff air out in a way that pulls in ambient air. Nuventix claims they will not clog with dust and can run silently. A range of circular finned heatsinks are available from the same company that are matched to specific puffers and specific LEDs.
Fans are used inside LED car headlights. Are these any use in luminaries for buildings?
“I have not found anybody using fans,” says Ward.
One last piece of advice:
“People always focus on the thermals of LEDs,” said Lumileds’ Kruger. “Quite often you will have the driver inside heatsink, and sometimes it will have electrolytics – whose reliability goes down very rapidly above 100°C. A system is only as strong as its weakest link.”