Measuring LED efficiency

[TMaxElectronics]

For a while now I have been playing around with some ultra high power (>100W) LED modules. Obviously they need heatsinks, but how large?
I often read the following: "To get the power your heatsink will need to dissipate, take the LED voltage and multiply it with the current", but is this accurate?

Not really if you think about it; A large part of this is turned into light, exactly how much though, or with other words: how much power is transmitted optically that the heatsink doesn't need to worry about?
I did some digging but was not able to find any formula to calculate this from the very common lm/w (probably because of the wide spectrum of the light), so we need to get the efficiency from somewhere else, and I decided to start measuring a few of the LEDs that I have, starting with a 250W one.

There are two main ways to measure the optical power output: One is to use an optical calorimeter, that heats up from the light absorbed by it, the other is similar, but using the thermal power output instead of the optical (this is probably the better way, as there are no losses from light reflection). You can then calculate the energy received from the temperature increase and the mass:

Code: [Select]

E = c x m x dT with c being the specific heat capacity
From this you can then calculate the optical power, if you know how long the LED was on for:

Code: [Select]

P = E/tI chose the optical approach, because I honestly didn't think of the direct one in time... 

Using this method on the CXM-32-50-80 LED I measured an optical power output ranging from 107W-101W, depending on the integration time (probably caused by heating of the LED chips and the phosphor), with an overall power input of 260W.
So around 40% overall efficiency (this is probably lower than the actual value though, due to the losses from reflection on the calorimeter surface).

Does this really matter though?
I do think so, as long as you are using an ultra high power LED:
with 100W less going into the heatsink than the conventional method would tell you, it can be A LOT smaller, cheaper, etc.
But when using a 100W LED, the 40W less are not significant enough to care about, unless you are designing your own heatsink (which you probably won't if you are reading this post ), especially since a CPU cooler (the most common choice for something like this) can easily handle >100W.

Measured LED efficiencies so far:

• CXM-32-50-80: 40% at 260W, more when run cooler

I will update this list as I test more COBs (that will take some more time though. Exams coming up blah blah blah )

Hope this helps someone

If you want some more info on high power LEDs, I have a post on my webpage: https://tmax-electronics.de/projects/tungsten-led/
And if you want to see a cringy video of me trying to explain this, saying "ehmmm" and "so yeah" a lot you can do that on youtube:

[davekni]

Some LED products have at least a bit of information about real thermal load.  Luxeon spec's "electrical thermal resistance" vs. "actual thermal resistance" for their packages.  You can take that ratio to extrapolate the real thermal power being dissipated.

Somewhere I saw an LED specification that specifically listed the ratio of thermal power to input power.  Can't find that at the moment.

[Twospoons]

You'd think it would be a standard spec on a datasheet, given it is critical to safe operation of the LED. I guess the manufacturers just expect everyone to use oversize heatsinks.

[klugesmith]

Good work there, measuring the partition of I * V power between radiant power and conducted heat.

Could also work it out from specified or measured lumens and wavelength.
https://en.wikipedia.org/wiki/Candela
One lumen of green light at 540 THz (about 555 nm) is exactly 1/683 watt of radiant power.

Have read that one lumen of "white light" is very roughly 1/280 watt.
That is roughly consistent with report of 40% efficiency, at between 100 and 140 lumens per I * V watt.

I have measured a complementary phenomenon:
the cooling effect on a photovoltaic panel in sunlight, when electrical power is drawn from it.

[johnf]

LED junction temperature is what matters there is a max that should not be gone past
IMO you should stay well away from this point as light output decreases quite quickly when you run @tj max
so junction to case and case to heatsink now matter.
I designed the Hella LED truck lights circa 1996
The Hella spec for them was -50 to plus 80 degrees Celsius ambient and be fully operational.
The +80 end required careful circuit board design to heatsink the Pirahna LEDS correctly.
Extended temperature testing to +120 degrees saw the LEDS permanently loose brightness ie damaged junctions

[TMaxElectronics]

Have read that one lumen of "white light" is very roughly 1/280 watt.

I think I remember that too, but that depends heavily on the spectrum of the LED. As a cold white lamp outputs more higher energy blue photons than a warm white one would.
Maybe a factor can be calculated if the spectrum is shown in the datasheet, by integrating it to get the average wavelength per photon, it might be possible to use the equation for a single wavelength source.
I might try this later...

LED junction temperature is what matters there is a max that should not be gone past. IMO you should stay well away from this point...

In deed, but I try to get a closer approximation of the power the heatsink will need to deal with here. If the LED puts out 100W optically, the heatsink can be much smaller than one for the full power of the LED, while still keeping well clear of Tj,max

The efficiency decreased by  ~5% as the phosphor and the LEDs heated up from room temperature to the operating point. That would explain why the decrease was only really noticeable between the 10sec and the 30sec test.

You'd think it would be a standard spec on a datasheet

That's what I was thinking too, and I'm still not 100% sure if its just me not looking in the right place...

[klugesmith]

>>Have read that one lumen of "white light" is very roughly 1/280 watt.
I think I remember that too, but that depends heavily on the spectrum of the LED. As a cold white lamp outputs more higher energy blue photons than a warm white one would.

Yup, but beware of counting photons 'cause a watt of blue light is half as many photons as a watt of red light.
A lumen of blue or red light is a lot more power than a lumen of green light.
Charts below have nothing to do with electrical watts, and show conversion between radiant watts and lumens.

[Twospoons]

Wow, thats going to be really messy to calculate.

On a side note, it is possible to use an LED to measure its own die temperature directly.
You need a test current source which is a small fraction of the normal working current ( say 1/1000), a fast switch to change between the test current and working current, and a 'scope.  The use of a  very small test current avoids self-heating and  lead resistance effects.
Do a two point temperature calibration of the LED, measuring Vf using the test current (ice water and boiling water make good refs), then calculate the temperature coefficient of the forward voltage of the LED. Vf is pretty linear with temperature.
To measure the die temp at its working current you set up a scope across the LED leads, run at working current then switch over to the test current. You can read Vf off the 'scope just after switchover, then calculate the die temperature using the temperature coefficient from the calibration.  All this can be done without direct access to the die, via the leads.  Neat eh?

I did this at work with some tiny IR LEDs we were pushing way past their datasheet ratings in a fast pulse regime.  We ended up adding a small heatsink.

[TMaxElectronics]

Wow, that's going to be really messy to calculate.

Yeah I forgot that lm was not based on the photon energy but candela, and thus uses the luminosity function...
One would need to somehow get the Wopt/lm, which might be doable by combining the luminosity function with the spectrum and then doing... something

On a side note, it is possible to use an LED to measure its own die temperature directly.

Cool idea! I did a similar thing with transistors in an electronic load of mine by measuring base-emitter voltage, but using the measured emitter, collector (and thus also base) current and a measured thermal coefficient as calibration factors. It was pretty accurate afaik, at least compared to using the thermal resistance from the datasheet together with the power output to estimate it.
In my case I could do some pulsed current measurements of the forward voltage (basically measure the resistance in series with the ideal diodes without heating up the diode) and then use that with the set current (which is actually fairly accurate) to remove the current coefficient and get the forward voltage of the ideal diode even when at full current.
Might be an interesting idea for a video if I'm honest... My controller has got a PWM pin and can get to the set current in <10 switching periods, with adequate cooling that should be good enough to get the measurement.

[davekni]

I've also used the method as Twospoons described for FET die temperature.   One significant caution with interpreting these results, at least for larger dies:  The die-to-case thermal resistance we measured for TO220 FETs regularly came out about half of the datasheet spec.  After much discussion with ST and Infineon, we learned that such was normal.  The data sheet thermal resistance specification includes margin for voids in the die attach solder.  Pulsed forward-drop temperature measures (weighted) average die temperature, not hot-spot temperature.  At high power dissipation, hot-spot temperature can be way above average die temperature.  Well-written specifications keep hot-spot temperature under 175C or whatever the technology can handle.