LED Lighting - Thermal Management

Main Index

Lamps & Energy Index

• Introduction
• 1 - Electronic Component Weaknesses
• 2 - Heatsinks
• 3 - Thermal Interfaces
• 4 - Temperature Rise
• 5 - Temperature Measurement
• Conclusion
  
  
• Design Of Heatsinks - Separate Article

LED lighting is now mainstream.  In less than 10 years it's gone from being a curiosity, with mainly relatively low-power LEDs that were far too blue for anything useful, and now keeps thousands of engineers worldwide fully occupied.  All LED lighting products are a marriage of very different technologies, none of which are traditional in the industry.  Of these, thermal management is one that determines the life of the end product.  Get this part wrong, and the fitting is destined for failure.

It doesn't matter how well a product works or how good it looks, if the electronics get too hot it will fail, long before anyone intended or hoped.  Most LED lighting products are intended for long life ... at least 30,000 hours.  Even a small error during the design phase or a misguided purchasing officer can reduce that to 10,000 hours or less.  Customers aren't happy, distributors likewise, and a small factory can easily go out of business as a result of the returns and warranty claims.

There are several potential failure mechanisms, and it is very important that everyone understands the limitations that must be considered once LEDs and electronic power supplies become involved in a design.  The design processes for each type are completely different, for a variety of reasons.

Unlike a traditional incandescent GLS lamp, heat cannot be tolerated.  When designing a fitting for GLS lamps, it is only important to ensure that wiring, sockets and other parts of the lamp can withstand the heat.  Everyone knows that the lamps will get hot, and everyone also knows that labelling a fitting as being suitable for (maximum) 60W lamps doesn't mean that 100W lamps won't be used.  The fitting might partially melt, but as long as it remains electrically safe it doesn't matter.  The customer won't get a replacement under warranty because it's obvious that the lamp was bigger than specified.

Now consider a LED (or CFL) lamp.  You can write a novel on the packaging, explaining to the purchaser that the lamp can't be used with dimmers, must have proper ventilation, etc., etc.  The customer won't read it! This is almost guaranteed.  There will always be curious or pedantic customers who read everything, but they are very much in the minority.  People (including professionals) simply aren't used to reading instructions for lamps, whether individual globes/ bulbs or complete fittings.  Lights and fittings are considered to be 'simple' devices that just need to be connected to the appropriate supply voltage and mounted in the ceiling or on a wall.

This was (more-or-less) true before, but no longer.  Once there are electronic parts involved (power supplies/ ballasts, LEDs, etc.) the product is no longer simple.  Heat is the natural enemy of all electronics, and failure to ensure that proper cooling is available will lead to reduced life for even the best engineered products.  Everything you thought you know about lights has changed.

All electronic parts are specified for a maximum temperature.  At lower temperatures, there are curves to show the maximum power the device can handle for a range of temperatures, with maximum possible device power often only being available at a case temperature of 25°C.  These derating curves usually indicate that the maximum possible junction temperature for most semiconductors is around 150°C, at which they cannot dissipate any power at all! For LEDs, the maximum temperature is usually somewhat lower, and there is a 'golden rule of thumb' for all electronic parts ...

Device life is doubled for every 10°C temperature reduction - and conversely ...
Device life is halved for every 10°C temperature increase.

In general, the lower the temperature (but ideally remaining above 0°C for many components) the longer parts will survive.  Some parts have a seemingly impossibly low rated life, in particular electrolytic capacitors.  Most are rated for only around 2,000 hours at full rated voltage and temperature.  If such a part is included in the power supply of a long-life lamp, then the only way to ensure long life is to keep the temperature as low as possible.  Electrolytic capacitors are one of the weakest links in the power supply chain, and it can be a real challenge to ensure that they last for the full life of the lamp.

Unfortunately there is often no easy way to eliminate electrolytic capacitors (electros), because they provide large capacitance in a relatively small volume.  The highest commonly available temperature grade is 105°C, so if electros can't be eliminated from a circuit it's imperative that the maximum temperature is always less than the maximum for the part.  To achieve 50,000 hours life at rated voltage, the temperature has to be maintained at no more than 55°C.  This is very limiting!

It might be possible to use a much higher voltage part than needed, so (for example) a 450V, 105°C cap operated at 250V can be raised to 65°C and still provide over 50,000 hours life [1] - at least in theory.  While this is a very nice theory, it's much harder to achieve than it appears, and some capacitors are simply unreliable (high voltage, low capacitance types especially so).  Even if you manage to locate a supplier of ultra-reliable electrolytic capacitors, should the purchasing manager find a cheaper alternative, you can guarantee that it will be used without question.  Then there will be a cascade of failures, along with very unhappy distributors and customers (I have seen this and the results, and I can offer a hint - denial doesn't help!).

There are few alternatives to electrolytic caps, but in some cases a circuit modification may allow the designer to use a film capacitor instead of a low value electro.  Where a large amount of capacitance is needed, there is no choice - no other capacitor is suitable.  It might be possible to reduce the specification, for example to allow a high lamp flicker at 100/ 120Hz.  Most people won't notice, but some will, and some purchasers may also demand low flicker for a variety of reasons.  Fortunately, high value electrolytics (100uF and above) seem to be a lot tougher than their specification might infer.  Even when one would imagine that failure seems inevitable, they manage to survive against all odds.

Semiconductors are also heat-sensitive, and if they aren't properly rated failures are again inevitable.  Proper heatsinking and keeping other parts (such as electrolytic capacitors) away from anything else that generates heat are vitally important if a power supply is to give the expected life.  Ultimately, every part used in the power supply needs to be examined carefully to make certain that it will operate well within its ratings at all times.  From the smallest resistor, every capacitor and through to all the semiconductors, every part must be double checked, and alternative parts used if those initially selected are operating at close to their maximum ratings.  Despite warnings, end users will install lamps/ luminaires in unsuitable locations, and many failures will occur well within the warranty period if anything is overlooked.  I know this from personal experience, from examining failures in commercial products.

The single most important limitation of LEDs is their operating temperature.  The light emitting junction should remain below 85°C, although various manufacturers claim that full power can be applied at up to 100°C (and sometimes more) junction temperature for some of their products.  However, regardless of claims, the lower the temperature the better.  Light output falls with increasing temperature, and most of the quoted figures are for a junction temperature of 25°C.  Output can be expected to be around 90% of that claimed if the junction temperature is at about 60°C, falling further as temperature increases.

Maintaining the lowest possible junction temperature not only maximises light output, but also the expected life.

First and foremost, it must be emphasised that a heatsink without airflow is not a heatsink! A heatsink isn't some magical object that can dispose of heat - all it can do is act as an interface between hot semiconductors and the air.  If the air can't move around the heatsink so that hot air is replaced by cool(er) air, then everything just gets hotter until thermal equilibrium is finally reached.  Unfortunately, this temperature is almost always far in excess of that which is tolerable for semiconductors and other electronic components.

Thermal transfer between two materials involves conduction, convection and radiation.  Conduction applies between the semiconductor die and its substrate, and between all materials that are in direct contact with each other.  Thermal interface materials (thermally conductive grease/ epoxy, graphite sheet, silicone thermal pads, etc.) vary widely in performance.  Silicone materials should normally be avoided because they are usually rather poor thermal conductors - despite claims to the contrary.  Convection and radiation are the methods that the heatsink uses to transfer heat to the air.

Natural convection (no fan or other device to move air) requires that there is unimpeded airflow around the heatsink fins, so that cooler air can replace air that has been heated due to contact with the heatsink.  Radiation is another way that heat migrates from the heatsink to the air, but that typically only amounts to 10% (or less) of the total heat loss.  Convection is always the dominant mode of heat transfer.  Radiation efficiency can be improved by black anodising on aluminium heatsinks, or a thin coat of matte black enamel or similar.

A heatsink is an interface between the hot junctions (e.g of LEDs intended to provide light) and the ambient air.  Between the junctions and the heatsink there are many other interfaces, such as between the LED die and its substrate, substrate to carrier/case (often thermally conductive ceramic), carrier to aluminium backed printed circuit board, and finally from the PCB to the heatsink itself.  Getting a worthwhile thermal resistance from the heatsink to the air depends on the heatsink's thermal conductivity, surface area and the surface coating.  Black heatsinks work best, but only if the black coating is thermally conductive.  The temperature of the surrounding air must be low enough to ensure that heat will actually flow from the heatsink's surface into the air, generally by a combination of convection and radiation.  Forced air cooling improves the passive process of convection dramatically, but is not always suitable.

If the air temperature is the same as the surface temperature of the heatsink, there is zero heat transfer.  The efficiency of any heatsink is determined in part by the temperature difference between the heatsink and the air - the greater the difference the better the heat transfer.  This is why electronic lighting must never be installed in completely sealed fittings.  Because the hot air can't be replaced, the temperature climbs until the temperature gradient between the outer (sealed) case and the surrounding air is sufficient to cause heat loss.  Unfortunately, by the time that happens, everything is way too hot on and near the heatsink, and failure is just a matter of time.

Thermal management is not a glamour industry, so relatively few engineers are drawn to the exciting world of efficient heat transfer.  Mostly, it's just another task that has to be done, but without the deep understanding that's needed to ensure it's done properly.  I have pointed out that there is no such thing as a heatsink that's too big, but if looked at from an economic viewpoint there's probably no such thing as a heatsink that's too small - being smaller, it's also lighter and most importantly, cheaper.  However, it won't work very well.

For those who want to delve deeper into this topic, see the ESP article on Heatsinks.  There's no point repeating everything there, and the only real difference is that we are now trying to get rid of the heat from LEDs, rather than power transistors.  The underlying principles are exactly the same, except that with LEDs used for lighting the power is continuous rather than variable in an audio amplifier (for example).

Figure 1 - Heat Flow From LED Junction To Ambient

Of the many thermal interfaces that need to be addressed, the designer only has direct access to those between the LED base (be it ceramic, metal or a combination), from case to heatsink and from heatsink to the surrounding air.  The LED base is a heat spreader, and has to be able to collect and distribute the heat from the die itself, which may only be 1mm².  The spreader used must have very low thermal resistance to be able to collect the heat from such a concentrated source, and spread it over a large enough area to make it feasible to provide a good thermal interface between the base and heatsink.  Every additional thermal interface makes the problem of heat removal worse, and introduces extra thermal resistance.  The capacitances shown represent the thermal inertia of each section, but this is only meaningful for intermittent use! Only the thermal inertia of the PCB (if metal cored) and the heatsink are significant, and may require a long warm-up time to allow steady state conditions to stabilise.

Surprisingly, there is a way to reduce the thermal resistance from the LED die to the base - operate the LED at reduced power.  Rather than using a single 10W LED, you can use 2 x 10W LEDs, with each operating at 5W.  This approach is expensive though, and is not common.  Most LED lighting manufacturers want to get as much light as possible from each device - some will even push the boundaries and operate LEDs at greater than rated power.  Needless to say this results in premature failure, regardless of the heatsink's efficiency.

As an example, a 10W LED might have a thermal resistance (junction to case) of 1°C/W, with a further 1°C/W from case to heatsink.  That means that even with an infinite heatsink the LED die will be 20°C hotter than the heatsink (thermal resistances are in series and simply add together).  If one were to use two 10W LEDs, each operating at 5W, then the temperature rise for each LED junction will be reduced to 10°C.  The effective thermal resistance from junction to heatsink has been halved, but at the expense of using two LED arrays instead of one.  If you are struggling to get LEDs to operate at an acceptable temperature, this is one way that you might be able to win.

Where an intermediate layer is involved (such as an aluminium based PCB (metal core PCB - MCPCB) as shown above), there are now two extra thermal interfaces - case to PCB and PCB to heatsink.  Both must be able to transfer the heat well enough to maintain an acceptable die temperature.  Despite the claims made by various manufacturers of aluminium PCB and thermal interface materials, every interface makes the problem of heat transfer worse! When commercial imperatives are involved, the solution must also be cheap, both in terms of material and labour costs.  From many of the products I've seen, production cost is the overriding consideration and often comes before all else.

The commercial constraints make proper thermal management a real challenge, and the more power you need to remove the harder it gets.  The die temperature directly affects the amount of light, as shown in the following graph [2].  Cooler operation gives more light output and greater life.

Figure 2 - Light Output Vs.  Die Temperature

Note that anywhere that you see a reference to 'ambient' temperature, that refers to the ambient in the immediate vicinity of the lamp or luminaire, not the ambient temperature in the room where people are.  For lamps installed in ceilings (such as downlights), their own ambient can be a great deal higher than that in the room.  Electronics don't care how cool you might be, they are only interested in their immediate surroundings.  This point is often overlooked, and causes many products to run far hotter than expected.

While it might seem that if 1W of electrical energy is supplied to a LED you'll have to get rid of 1W of heat, this isn't actually the case.  With a modern high-efficiency LED, you can expect that ~25-30% of the energy supplied will be emitted as light, so you may only have ~700mW of heat to be disposed of.  With any lamp or luminaire you can still end up with a comparatively large amount of heat that has to be disposed of.  Since many lamps (in particular) have a small allowable space, that's where the challenges really add up.  The situation is greatly alleviated if the product is a complete luminaire, and full use should be made of the available space.  Ventilation is critical!

In the above table, emissivity refers to the material's ability to radiate heat.  This is not usually the primary way that heatsinks work, and most rely on convection as the main disposal method, with emissivity being a secondary consideration.  Black surfaces radiate far more effectively than polished surfaces, and fins increase the effective area dramatically.

The thermal resistance is vitally important, because if it's too high, heat will not be conducted away from the heat source into the body of the heatsink effectively.  This can cause the LEDs or other semiconductors to overheat.  Likewise, one must consider the thermal conductivity of the various interfaces ... LED - substrate, substrate - PCB and PCB - heatsink.  If any of these are inadequate, the die will overheat.

Forced air cooling using either fans or 'jet' mechanisms improve the ability of any heatsink to move heat from the electronics or LEDs into the surrounding air.  As noted above though, if the surrounding air is the same temperature as the heatsink there is zero heat transfer.  Forced air cooling is generally something to avoid with fixed products that don't lend themselves to being cleaned regularly, because mechanical failure and/or dust accumulation will block the airflow and again, the LEDs or other semiconductors will overheat.

Thermal interfaces cause everyone grief, no matter if they are building an audio amplifier, power supply or high power LED light.  Some of these interfaces are internal to the LED itself, and there's nothing that can be done to change them, other than buying a different brand or type of LED.  Of those you do have control over, the choice of thermal interface material (TIM) is critical to the long-term performance of the LEDs.

One technique that is sometimes used is to use a double-sided FR4 (fibreglass) PCB, and use multiple vias to transfer the heat from one side of the PCB to the other.  The effectiveness of this is somewhat doubtful, and this method can only be used when the LED power is fairly low - probably no more than 1W or so.  As noted below, this method produces two thermal interfaces (one solder based), and the underside of the PCB needs to be insulated from the heatsink, adding to the total thermal resistance.

Figure 3 - Thermal Interfaces With A LED And Heatsink

In the above, all thermal interfaces are shown in green.  The designer has access to those interfaces between the LED base, MCPCB and heatsink, and these need to be as good as they can be.  In some cases, the base of the LED may be electrically connected to one or the other LED connections.  This complicates the mounting, because the LEDs then need to be electrically isolated from each other (assuming more than one LED), but still in intimate thermal contact with the MCPCB and/or heatsink.  Many LEDs are designed for surface mounting, and the base is soldered directly to the MCPCB.  Solder is a rather poor thermal conductor, so thermally conductive epoxy or other material may be used in addition to the solder which then only needs to provide electrical connectivity.

When MCPCB materials are used, the insulating layers above and below the copper have to be as thin as possible to aid heat flow, but need to provide electrical isolation.  This is not easy to achieve.  When two thermally conductive materials are in contact, tiny pockets of air are always present, and this impedes heat flow.  So-called 'thermal grease' contains very fine particles of thermally conductive material, and when used to join a metal core PCB to a heatsink (for example) the air is excluded and thermal performance is improved greatly.  If the MCPCB and heatsink have to be electrically isolated this always makes the situation worse.

If an insulating material is needed between two surfaces, it needs to have high dielectric strength to provide proper electrical isolation, yet still have high thermal conductivity.  With only a few exceptions these two requirements are at odds with each other.  Good electrical insulators are usually also good thermal insulators, so the material has to be extremely thin.  Silicone compounds are touted as being ideal, but most have rather poor thermal performance.

Note that with most of these materials, a coating of thermal grease is needed on each side to fill air gaps and obtain best results.  Silicone and graphite are the exceptions, but graphite is electrically conductive.  Beryllia (beryllium oxide) is one of the best of the affordable thermal interface materials, however the dust is toxic and it doesn't seem to be used very often.  It may be used where extremely good thermal performance is demanded, along with needing an electrical insulator, but I know of no LED assemblies that use it.

The mounting surfaces (e.g. LED and heatsink) must be as flat as it is reasonably possible to make them.  Any distortion of the surface will seriously affect heat transfer, and reduce the life of the LED(s).  While silicone pads are often used for bulk 'gap-fill' applications, they are only useful for low power because their thermal resistance is too high.  The thinner the material the better.  While this applies to all semiconductors used in the power supply as well, in general the continuous power levels of switching transistors and/or IC packages are quite low.

This is one of the biggest problems with LEDs - they have a high continuous dissipation, and that makes heat removal much more difficult.  The thermal inertia of the heatsink works wonders for an audio power amplifier, but it's of no use at all when the dissipation is continuous.  All that happens is that if tests are not run for long enough, you will never know just how hot everything will get.  Lights can be on 24/7, and there is heat being generated the whole time.  It's not uncommon to find that a LED array might need to be on for several hours before the temperature stabilises, at which point you can measure the temperature rise.

There is no point measuring the operating temperature of a LED or heatsink in isolation.  You must note the starting and ending temperatures, with the latter only taken after the temperature has reached a plateau.  This needs to be done for every possible mounting position for a lamp that offers options.  Once the temperature is stable - typically after an hour or more - measure the final temperature.  Subtract the starting temperature to find temperature rise.  For example ...

Start Temp. = 25°C
End Temp. = 48°C
Temp. Rise = 48 - 25 = 23°C

Should the lamp's heatsink be mounted in the ceiling space where the ambient temperature can reach 45° (could be as high as 60°C !) instead of 25°, the heatsink will be at a temperature of 65°C.  Depending on the thermal resistance and power dissipated, the LED die will be considerably hotter than the heatsink.  For example, if we have a 11W dissipation in a 15W LED (4W is emitted as light) and a total of 5°C/W thermal resistance between the die and the heatsink, the die temperature will be ...

LED Trise = 11W * 5°C/W = 55°C
HS Trise = 23°C
Total Trise = 78°C

Now, should the ambient rise to 45°C, the LED junction will be at a temperature of 123°C.  It won't last very long, and based on the above chart light output will only be about 90% of that specified at lower (more sensible) temperatures.  We did the measurements above that let us determine the heatsink's thermal resistance ...

Heatsink Rth = 23°C / 11W
= 2.1°C/W

That's a pretty decent sized heatsink, so in order to reduce the LED's temperature we need to reduce the thermal resistance between the die and the heatsink - the thermal interfaces must be improved.  To keep the die temperature at or below 85°C and allowing for a maximum ambient of 45°C, it's apparent that the total thermal resistance between die and ambient can be no more than ...

Junction-Ambient = 85 - 45 = 40°C
Power = 11W
40 / 11 = 3.6°C/W

Since the heatsink has a thermal resistance of 2.1°C/W, the maximum thermal resistance between the die (junction) and heatsink is 1.5°C/W.  If this cannot be achieved, it must be made to be as small as possible, and the heatsink must be made bigger (lower thermal resistance).  The only way to achieve the required thermal resistance may be to operate the LEDs at lower power, and use more of them to get the light output back.  The following shows a simplified diagram of what might be a typical mounting arrangement.  It is assumed that the contact area between the MCPCB and heatsink is fairly large, and uses a suitable interface material to achieve the low thermal resistance shown.  Note the parallel paths, based on using 2 LEDs at 5W each, rather than a single 10W LED that would overheat.

Figure 4 - Thermal Gradients Due To Thermal Interfaces

Please note that the above is representative only, and is not intended to represent a particular LED module.  Determining the optimum thermal circuit will often be an iterative process, and it might require several attempts to get it right.  Should the end user then ignore the maximum temperature rating then naturally the LED(s) and/or the power supply will not survive for the rated life of the product.  This process becomes far more difficult if the power supply must share the heatsink with the LEDs, and both the LEDs and power supply components will most likely run hotter than expected.

It is important that the designer errs on the safe side.  It's far better to have a LED or LED array that runs a bit cooler than expected than the reverse.  Lower temperature will never cause reduced life! Whatever the designer ends up creating will be challenged by the end-user or installer, and ceiling/roof space temperatures will nearly always be higher than expected.

Even on a reasonably warm (but not especially hot) day, it's not uncommon for the roof-space of a domestic dwelling to reach 50°C or more - it is a very hostile environment, and allowances have to be made if LED lighting products are to survive.  Ceiling insulation and the ceiling itself will reduce airflow, and this should have been considered in the design - the heatsink needs to be larger than expected.  Never allow insulation to cover the heatsink - it will overheat as a result, regardless of how good it might be.  A very rough idea of a heatsink's thermal resistance can be obtained from ...

Thermal Resistance = 50 / √A     Where A is the total surface area in cm²

Note that the above does not consider the thermal differential between the heatsink and the air.  Any heatsink has a thermal resistance that's inversely proportional to the temperature differential between it and the surrounding air.  So, the hotter it gets (with respect to the air around it) the better it performs ... but the semiconductors will die of heat exhaustion unless you get the balance right.  Also note that thermal resistances in parallel are calculated the same as electrical resistors in parallel, so 2 x 1°C/W thermal resistances in parallel gives an effective 0.5°C/W.

One thing that is becoming apparent is that the use of LEDs for lighting has created more research into heatsink materials and thermal interface materials than ever before.  Traditional heatsinks (aluminium, copper, etc.) may end up being replaced by various graphite (carbon) materials that are being developed.  New techniques and alloys promise greater thermal conductivity than we are used to [3].

Measuring the temperature of any LED junction is usually very difficult.  Some manufacturers specify the temperature at the 'solder point' - right at the lead where it enters the package.  This allows the test engineer to attach a miniature thermocouple or other sensor to the lead, and a good reading should not be too difficult.  Thermal imaging cameras are often used, but unless you know exactly how to use one and perform the setup so that the emissivity is set properly, the readings will be wrong.  This also applies to the more pedestrian infra-red thermometer.  These usually don't even have the provision for entering the emissivity of the target, and the readings obtained are generally not very useful other than for comparative readings.

It would be nice if it were possible to monitor the junction voltage and get a reading.  Unfortunately, the thermal coefficient of voltage for LED chips is not a fixed quantity.  The LED junction voltage coefficient varies between -1 to -4mV/°C, and while it is certainly possible to obtain a good reading for the voltage decrease with temperature, unless you know the exact tempco of the junction it does you no good.

A method that suggests itself is that one can use the LED manufacturer's data for light output versus temperature.  Figure 2 (above) shows the light output with temperature for a Cree XLamp, and if the light output can be measured accurately this will provide an indication of the junction temperature.  The light output at (say) 25°C can be obtained by using current pulses that are sufficiently short to ensure that the junction doesn't have enough time to get hot (see the thermal resistance and thermal inertia diagram in Figure 1).  As long as the current pulse is short enough and enough time is provided between pulses, the thermal inertia of the junction and substrate will ensure that the junction remains at (close to) ambient temperature.  Current pulses need to be the same magnitude as the normal operating current.

Much the same method probably can be used for junction voltage, provided it is characterised first.  You need a mounting plate that can be set for a specific temperature, and the LED junction is pulsed the same way as for measuring light output.  The pulse current must be the same as the normal operating current to minimise errors.  If the LED is tested at (say) 25°C and again at 50°C, it's quite simple to get the junction's thermal coefficient of voltage.  Once that and the 25°C voltage are known, the junction voltage at the design current and with a standard heatsink will reveal the LED chip temperature.  Fast instruments with excellent common-mode range and the ability to hold the reading are essential for tests that rely on pulsed current.

Some LED driver ICs made by Texas Instruments/ National Semiconductor, Cypress, Allegro and several other manufacturers have the ability to monitor the LED temperature and reduce power should the temperature become unsafe.  These ICs rely on the designer to furnish the correct component values to make the temperature sensing works properly.  They are not able to monitor the LED die temperature directly, but it may come to pass that high power LEDs eventually incorporate their own thermal sensor, and the output from that can be passed to the driver electronics.  This would be an ideal solution, as the sensor can monitor the true die temperature.

Most modern LED drivers are current sources.  The voltage can vary over a fairly wide range to account for different LEDs, changing temperature, etc., but the current through the LEDs is fixed.  Typical drive currents are 300mA, 350mA, 700mA, 1,000mA (1A), 3.2A, etc.  Ideally, a power supply should have an adjustment so the current can be set to the desired value.  While this allows sensible people the opportunity to reduce the current and thus the heat, it also allows non-sensible people the option of setting the current too high so the LEDs will give a bit more light, but will have a dramatically reduced life.

Because the circuitry of the power supplies is real-world, there are limitations.  The voltage needs to be within a reasonable range, such as (perhaps) 27-42V, and this will be suitable for 10, 11 or 12 x 1W LEDs in series.  The nominal LED voltages will be from around ~32V (10 LEDs) up to ~38V (12 LEDs).  If the current is fixed at 330mA the total LED power will range from 10.6W for 10 LEDs up to 12.5W with 12 LEDs.  If the power supply current is adjustable, it could be reduced to 250mA, giving 8W and 9.5W respectively.

It is very likely that no-one would notice the slight reduction of light output, and by reducing the current the heat is also reduced, leading to lower temperature operation, greater efficacy and longer life.  The more I work with LED light fittings, the more I like the idea of operating them at a lower than normal (rated) current.  Very few fittings that I get to see could be accused of not being bright enough - quite the reverse, many are too bright.

Figure 5 - Various LEDs - 100W, 1W and 10W (2)

The LEDs shown above are arrays, except for the 1W LED which is a single die on a standard 'star' MCPCB.  The 100W LED uses a 10 x 10 array - 10 parallel strings of 10 LEDs in series.  The nominal voltage is around 32V, and current for full power is 3.125A (312.5mA for each parallel string).  10W LEDs are usually arranged in a 3 x 3 array, with a forward voltage of 9.6V at a current of about 1A.  The power supply's output voltage will typically be from around 8V to perhaps 15V or so.  The actual output voltage is fixed by the LED array being driven.  Note that the supply current must be no more than the LED is rated for.

For the following examples, the total input DC power is used, with no allowance for light radiation.  This is a simplified approach, and if you calculate it like this you automatically apply a safety margin.

If we assume a median thermal coefficient for LEDs of -2.5mV/°C, it is obvious that with a junction temperature of (say) 80°C, the voltage will be ~120mV less than at 25°C for a single LED.  For a 10W LED (3 x 3 array), the voltage across each series string will fall by 360mV, from a nominal voltage of around 9.6V to 9.24V.  At a current of 333mA for each paralleled string of 3 LEDs, the power will fall when the LED gets hot ...

Constant Current ...
Ptotal = Vf * 1A = 9.6 * 1A = 9.6W @ 25°C
Ptotal = Vf * 1A = 9.24 * 1A = 9.24W @ 80°C

There may be cases where constant current power supplies are not suitable, and a more traditional constant voltage supply is called for.  The LEDs still require a current limited source though.  In a few cases (low power only), it might be appropriate to use a series resistor or a linear current regulator, but these both dissipate significant power and generate heat.  Remember that the total efficacy for a LED fitting includes all losses in the power supply, and if too high the overall efficiency of the fitting will suffer badly.  If we use a 12V supply and a 2.2 ohm limiting resistor, we get the following ...

Constant Voltage (12V supply, 2.2Ω Resistor) ...
I = Vin - VLED / R = 12 - 9.6 / 2.2 = 1.09A (25°C)
Ptotal = Vf * 1A = 9.6 * 1.09A = 9.7W @ 25°C
Ptotal = Vf * 1A = 9.24 * 1.25A = 11.6W @ 80°C

Oh dear - when the LED gets hotter and the forward voltage falls, more current is delivered and the power increases.  This is undesirable in the extreme, and we haven't considered the power dissipated in the resistor yet.  This will be 2.6W at 25°C, rising to 3.4W at 80°C, and this is wasted because it doesn't contribute to light output.  Using a linear current regulator will give similar losses (dissipated in the regulator), but with any current regulator LED power is reduced when the LEDs get hot.  Dissipation in a linear regulator will not increase as much as with the resistor because the current is fixed - a linear regulator's dissipation will range from 2.4W to 2.76W at 1A.

Customers are expecting to see figures of 80 lumens/ Watt or better, and if power is lost anywhere this is impacted.  It also means that there's more heat to get rid of, so efficiency is the key to both efficiency and minimising heat - these parameters are inseparable.  If we assume 100lm/W for the LED, the total efficacy will be reduced to 67lm/W due to power dissipated in the resistor, and that's not including losses in the power supply.

In addition, the LED is receiving 16% more power than it's rated for at maximum temperature, so will run even hotter than expected and life will be reduced.  Resistors can be used, but normally only where the total power is fairly low (no more than ~3W or so), and where overall efficiency is not an issue.  For all general lighting applications, a switchmode current regulator is the only way to ensure that losses are kept to a minimum.  There is also a small amount of temperature compensation built-in, so if the LED(s) get too hot the power is automatically reduced.  Many of the latest ICs intended for LED lighting have temperature sensing, and will reduce the power even further if the temperature is too high.

Everything in a LED lamp or luminaire design has to be carefully worked out to ensure that nothing is stressed, both electrically and thermally.  Heat is the natural enemy of semiconductors and electrolytic capacitors, and a worst case design is always called for.  Heat generation is continuous as long as the light is on, and in many cases it will be 24/7.  Building temperature variations can be extreme, and the design of the LED lighting solution has to assume that the ambient can easily be as high as 55°C or even more in some cases.

To make matters worse, you can't always count on adequate ventilation, and in some cases you can assume there will be none.  Installation instructions must point out very clearly that ventilation is mandatory - the fitting simply won't survive without it.  This general warning isn't just for LED lighting - it also applies to CFLs and electronically ballasted lamps of all kinds, including fluorescent and induction types.

All electronic components are stressed by high temperatures, with LEDs, electrolytic capacitors, transistors (bipolar and MOSFET), diodes and ICs being the most vulnerable.  It is essential that the manufacturer of all electronic ballasts/ power supplies intended for lighting indicate the maximum allowable ambient temperature, and this should be realistic - provided by engineers, rather than determined by the marketing department.  While inflated claims might entice more customers, they will be unhappy customers if the claims are not based on engineering principles.

It's very important that customers are aware of these limitations.  There will always be installations that indicate that some other lighting type is needed, and sometimes the only alternative is to stay with traditional incandescent lights or perhaps magnetically ballasted fluorescent tubes.  Even with the latter, there is an upper temperature limit imposed by the plastic tombstones and other connectors, and the ballast itself.  As an electrical component, even that is vulnerable to high temperatures and it may fail prematurely if the temperature is too high.

Consider that an incandescent lamp runs hot, and everyone knows that this is the case.  You can get ceramic sockets fitted with high-temperature wiring, and these can operate at 200°C or more for many years without failure.  No electronic lamp can survive high temperatures, and it is completely unrealistic to expect the LEDs and/or other electronics to handle the same thermal abuses as incandescent lamps.  No electronic lighting system should ever be subjected to more than 50°C ambient temperature, remembering always that the ambient is the air around the lamp/ power supply itself - not the temperature in the room!

Attitudes must change, and customers (and sales people) must be aware of the limitations of all electronic lighting products to avoid disappointment and bad outcomes.

1. Capacitor Life Calculators - Illinois Capacitor
2. XLamp Thermal Management - Cree
3. Advances in LED Thermal Management - Digi-Key
4. ESP Article On Heatsinks

Other references that have assisted include Wikipedia and manufacturer data sheets from Cree, LumiLeds and various others.

Main Index

Lamps & Energy Index