Measuring led die temperature- Mtbr.com

# Thread: Measuring led die temperature

1. ## Measuring led die temperature

Hi all,

Wouldn't it be nice to measure the temperature of the internal led semiconductor junction, or die, as this is the temperature that really matters. To be precise, the maximum junction temperature for the Cree XRE and MCE is 150 DegC.

Actually, it should be relatively easy to measure junction temperature, given the published forward voltage temperature coefficient of -4.0 mV/DegC. In other words, for a constant led current, the forward voltage drops by 4.0mV for every degree of temperature rise. Therefore, by measuring the fall in led voltage, from the instant the led is turned on, we can easily calculate the rise in junction temperature.

Firstly, a quick check to see if the temperature coefficient is sufficiently large to produce an easily measurable change in voltage. Assume the led starts out an ambient of 30 DegC, and finishes with a junctiuon temperature of 130 DegC, so the junction temperature rise for this example is 100 DegC. The fall in forward voltage is therefore 100 x 4mV = 400mV, which is more than enough to accurately measure.

A minor practical problem is that it will be difficult to measure the forward voltage at exactly the time the led is switched on, because the thermal time constant of the junction itself will be quite short (probably well less than 1 second) so by the time we take the voltage measurement with our digital voltmeter, the led die temperature will already have risen above ambient. There are several ways around this problem. Ideally, use a computer based data acquisition system that can take thousands of voltage readings every second. I have that available, but just at the moment don't own any leds!

However, if you don't have a fast data acquisition card, the initial 'fast" rise in junction temperature can be estimated from the published thermal resistance from junction to package heatsink solder point, and is 8DegC/W for an XRE, or 3DegC/W for an MCE. Taking the MCE as an example. At maximum current of 700 mA, the power dissipation is typically 0.7x3.4x4 = 9.5 Watts. Thus the fast initial junction temperature rise is 9.5x3 = 28 DegC. For the XRE, the result is 3.7x8 = 29.6 DegC, which for practical purposes is the same. Fortunately this does not need to be known with great accuracy, as the overall temperature rise of around 100DegC or more is considerably greater and will dominate the result.

In summary, a simple procedure for measuring junction temperature rise is as follows. Firstly, you preferably need a precision constant-current lab power supply, though the better led drivers, eg Taskled, are probably good enough. Connect the voltmeter across the led, switch on the current, and as fast as possible note the initial led voltage, and then record the voltage versus time as the housing heats up. Let V0 be the initial measured voltage, and V be the voltage at time t. The junction temperature rise at time t is given by :

Temp Rise = (V0-V)/0.004 + 28

I'm ordering a couple of MCEs today from Cutter, and will certainly perform this measurement when I get them. Anyone out there care to try?? This method beats the heck out of guesswork, and lets you know whether your favourite housing is keeping the led(s) within thermal ratings, or not. To look at it another way, you can ascertain fairly well how hot your housing needs to be before the led(s) reach maximum junction temperature. It may be that most of are in fact nowhere near the max led temperature, or maybe we are, but at present we are guessing, and I don't guess when I can measure ....

Anyone else out there find this kind of stuff interesting?

Colin

2. I've noticed the effect and the cheapo DX drivers work well enough to measure the effect. I want to use it to keep my light from frying when there is no air. I'm still figuring on using a thermistor. What if the light gets hot somehow before you turn it on. You have to consider that.

Don't forget that some of the watts are coming out as light and it's not all turning into heat. I'm not sure of the conversion efficiency, but it's got to be 10% or so.

3. This is a great idea, with a PIC based driver, the voltage could be monitored and used in place of a temp sensor, or to back one up. I have just been winging it with a temp sensor attached to the heat sink and it seems to work well, but monitoring voltage seems much simpler hardware wise.

4. Originally Posted by HuffyPuffy
This is a great idea, with a PIC based driver, the voltage could be monitored and used in place of a temp sensor, or to back one up. I have just been winging it with a temp sensor attached to the heat sink and it seems to work well, but monitoring voltage seems much simpler hardware wise.
Hi,

Actually, I wasn't intending to suggest that the method could be used when riding the bike, because the forward voltage also depends on the current, which will likely be dimmed to a number of levels.

I envisaged this as a 'one off" laboratory test, essentially to establish the relationship between led junction temperature, and heatsink or case temperature. We can measure case or heatsink temperature when riding the bike, or even estimate case temperature "by touch" when riding, but what we do not presently know is how these easily measured temperature relate to the all-important junction temperature. This experiment tells us that.

Colin

5. I suppose one could figure it in real time based on the specific levels in a mode driver since those should be easy to determine, though there are probably other variables involved which I am not able to think of. Seems like it would be do-able though I probably won't be the one to

6. The question is how variable is the tempreture coefficent? There was a very good thread on this on CPF take a look at this. At least from the tests there the tempreture coefficent is not constant but depends upon the driver current and the actual tempreture. There is no indication about the variability between leds which it would be interesting to know about.

Personaly for looking at the die tempreture I would prefer working off the thermal ratings for the package and measuring a tempreture as close to this as possible which in my case is would be on the Star. You can then estimate back to the die tempreture.

In practice what I want to know is how quickly will my housing get too hot to comfortably hold (about 60 degrees C) as that is as hot is I want it to ever get. My die should be OK with this so long as I do not have too many thermal boundaries. In still air so long as I have two or three minutes for getting this hot then I am happy as in moving air I should not have any trouble.

7. Would it be possible to measure the die temperature with an infrared thermometer, would this be accurate enough?

We have a handheld device at work that has a laser pointer, so you can see exactly the spot that you are recording temperature at.

I am working odd shifts this week, will try this method out if I get time!

What do you think? Is there a really obvious flaw to this method??

8. Hello,
IR method need a large enought surface, depending of the thermometer specs. The LED active surface is very small, and the mesured value will may be given by the surroundings materials.
Good luck.

9. Doh! I will check the spec sheet of the thermometer before building the test jig. I guess even an MCE will be too small.

Never mind, back to the drawing board

10. Originally Posted by ifor
The question is how variable is the tempreture coefficent? There was a very good thread on this on CPF take a look at this. At least from the tests there the tempreture coefficent is not constant but depends upon the driver current and the actual tempreture. There is no indication about the variability between leds which it would be interesting to know about.

Personaly for looking at the die tempreture I would prefer working off the thermal ratings for the package and measuring a tempreture as close to this as possible which in my case is would be on the Star. You can then estimate back to the die tempreture.

In practice what I want to know is how quickly will my housing get too hot to comfortably hold (about 60 degrees C) as that is as hot is I want it to ever get. My die should be OK with this so long as I do not have too many thermal boundaries. In still air so long as I have two or three minutes for getting this hot then I am happy as in moving air I should not have any trouble.
You are exactly right, as usual. The candlepower measurments show convincingly that the Cree spec of -4.0 mV/DegC is far to nominal and rubbery to be used in the way I had hoped.

I agree that the most practical approach is to use a tiny thermocouple to measure the temperature at a point as close as possible to the led heatsink surface on the back of the led.

Bumphumper suggested an infrared thermometer for this purpose, but of course this is impossible because the hot surface in question is not visible or physically accessible, because the back surfaceof the led is soldered to an MCPCB.

I have done some more reading on this topic, including Crees thermal management notes, and the recommended heatsink solder pad on the MCPCB has a small extension extending beyond the led, for the specific purpose of temperature measurement with a thermocouple. I suspect that the temperature measured at this point will be less than the true temperature at the back of the led, but is probably as close as it gets ....

As we both agree, calculating the additional temperature rise from back of led to semiconductor junction is easy and robust, because we have manufacturer specs for this thermal resistance. As I showed previously, at max rated current, this temperature difference is about 28 DegC, for an MCE or an XRE.

One question I was really curious about is the thermal resistance of the MCPCB. The way these things are made is very poor from a thermal viewpoint. The led is NOT soldered directly to the aluminium metal substrate. The led is soldered to a thin sheet of copper, and there is typically a 60 micron electrically insulating film between the copper and the thick aluminium substrate. Needless to say, this electrically insulating film also provides unwanted thermal resistance. Modern leds such as the MCE have an electrically isolated thermal path within the led, so it is really silly to have another thermal resistance in the MCPCB. A much better construction would be to make the metal substrate (slug) from copper, and solder the back of the led directly to this copper substrate, and I may end up making my own MCPCBs in this way. Note that the thermal resistance from substrate to heatsink or case is negligible, because the surface area is over an oder of magnitude higher, and also becasue this is a metal-to-metal interface.

It took me a while to find specs for the thermal resistance of a typical MCPCB for an MCE, but it appearsto around 0.8 to1.0 K/W. At full MCE led power of around 10W, this thermal resistance therefore represents an increased junction temperature of 8 to 10 DegC, which apparently the industry regards as OK, but bugs the heck outof me because it is totally unecessary!!

Some DIY constructors attempt to get around this by using AA epoxy to fix the led directly to the heatsink/case, but his has a couple of drawbacks. Firstly, it is inconveniently permanent (!!) and secondly, even the best thermally conductive epoxies have very high thermal resistance compared to metals, and the thermal resistance could be significant unless the adhesive thickness is VERY thin - I'll do the calcs on that if anyone wants.

I've bought 2 MCEs on 20mm star MCPCBs, and another bare so I can experiment soldering it directly to my own 20x20mm square copper substrate. In this case I'll mill the copper so there is a slightly raised area in the middle for soldering to the back of the led. The electrical connection legs will therefore be dangling without touching the copper substrate, and I'll solder the wires directly to them. Not difficult, really.

Sorry people, but I'm a perfectionist, and I just cannot stand having my led junction running 10 degrees hotter than it needs to ....

Now time to stop raving, while Ifor tells me where my reasoning has gone astray.

11. If you have ever tried to use a diode as a temperature sensor, that CPF thread is just about what is expected. A LED is even less "ideal" than a regular diode. If the goal is just to keep the LED from frying it can be good enough. The main issue that I see is calibrating the measurement.

Using a negative thermal coefficient (NTC) thermistor is a lot easier than using a thermocouple. The signal from a thermocouple is only a few millivolts and it takes more signal conditioning. Not too much of an issue on a lab bench, but it would require more circuitry for use on the trail.

12. cdcdcd ...will be very interesting what you find in correlating Fv to Substrate T for the MCE. Please report back!!

FYI, Currently I've used the small thermo couple next to the die, and a conservative thermal restance to work back to a substrate Temp. It leaves me searching for more as I'm close, but not exactly there yet. Worth it though, as by far T LED is the most important consideration in LED light design.

Thermal FEA (finite element) is where i'm heading myself. Couple weeks till. Correlate that w/ testing and it's solid. The over engineering way ...

~C

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•