Cree XR-E Thermal Concerns

[45/70]

In the original post,

Not a very efficient solution in my opinion.

refers to the MCPCB's as possibly being problematic. There have been suggestions of several solutions for their redesign, and later, some ways to redesign the emitter package as well. Reinventing the wheel and the cart it goes on, so to speak. Fascinating reading, for sure!

However, my question is, what can be done to the existing emitter/MCPCB setup? I don't have my stars yet, but feel one of the most important problems that might be faced is a poor reflow job, as I suspect they are done "in house". A few posts back, I mentioned a possible way to tell. In the first post, fifth picture of this thread, ArsMachina posted a very good picture that shows what I was referring to. What I see is solder around the + and - pads and only flux around the thermal pad. Would this not indicate that the reflow was incomplete on the thermal pad?

Dave

[Amonra]

has anyone tried these: http://theledguy.chainreactionweb.co...roducts_id=649 instead of the ETG star ?
Would they be any good ?

[McGizmo]

Anybody who has handled these LED's or soldered to them realizes that the ceramic package of the XR-E doe a magnificent job of distributing theheat well and throughout. I believe the FLIR images also show this. SO, the perfect thing to bond a XR-E to would obviously be another XR-E!





If there were room on the Z axis, this is a perfect application for Newbie's heat pipes! I used a diamond bur to remove the trace on the top corners of the LED facing you to open the "loop" for inputs. The thermal transfer to the brass pedestal seems to be quite good! ( A little more info)

(Doug S, not quite what you had in mind but I believe illustrative of the possibility?!)

[NewBie]

I've been working with these CREE parts for over two years now and find the ceramic package most definitely does not do that great of a job for heat spreading.

In the photo below, you can very clearly see a 7 degree Celcius delta, just across the ceramic, next to the metal center, then out to the end of the ceramic pad. It is even hotter inside, under the dome- especially in the center at the die, where you cannot see (glass does not transmit IR in the 7-14um wavelength range).




Sorry, I am most definitely not going to buy into that idea, and if I did, I'd have to be quite blind and ignorant to say the least...




Then you take the same LED, and mount it to a copper slab, such that you can pull the heat out of the center, and look how the pad temperature looks much uniform:



It is the copper that is helping tremendously.


The ceramic does a decent job of transferring heat, but not excellent.

I have an old image of a CREE with the dome and ring remove, just the die on the substrate, and the poor thermal spreading becomes quite clear. I'll need to do this to an XR-E, and put pictures up. This is why it is so important to pull the heat out from the center of the die well- but the area of the whole CREE does help in theory.


Isaac-

I took some more shots of the squares in the thermal pad of the CREE, see the ceramic under them?




The square hole in the thermal pad copper:



The electrical via on the electrical connection pad:

[McGizmo]

Newbie,
In the two FILR shots where you can see the LED and its surrounds, I assumed I was seeing a rectangle that represented the whole package. It also seems that the temperature is consistent across this rectangle. In the close up you just provided, there seems to be a greater contrast in color change and the color representing 78C which before seemed outside the package and on the MCPCB is now seen to be on the package? Is this a close up of the same image or a different image?

[NewBie]

Maybe this would help to see what I've been trying to show. Something that is obvious to me, isn't always evident to others. I've tried to "ghost" the image of the visible picture over the IR picture of the same item. Scaling isn't perfect:

[IsaacHayes]

Jar, great pics! yeah the squares aren't really holes in the ceramic but "holes" in the thin metal. Who know why they are there?? Cool pic of the via for the electrical connection!

Don: yes it's hard to solder to these things vs a luxeon, as luxeon is just a thin lead. This thing you gotta almost heat the whole package up as the solderpad is part of the led. I feel your pain!! I resoldered mine today and thought I was doomed! I almost torched the heatsink so it would help hold the heat.

[Doug S]

Maybe this would help to see what I've been trying to show. Something that is obvious to me, isn't always evident to others.

Jar, the pictures are great! I am, however, in the camp that is having a bit of trouble interpreting them. I have a couple of questions whose answers might be buried in this thread but I failed to see them.
1. What is the drive level for the MCPCB thermal images?
2. For those MCPCB thermal images is the MBPCB in turn mounted on a heatsink or is it just in free space? It would seem that the former configuration would be the one most revealing of the MCPCB performance relative to having the package mounted directly on a massive heatsink.
3. In the thermal images is the cursor location showing the peak temperature located over a portion of the package where the coating which equalized thermal emissivity *has not* been applied?
4. You seem to have an almost endless supply of great Cree emitter pics! One I'd love to see if you already have it is a XR or XR-E sectioned in a vertical plane.

Thanks for sharing your great work!

[glire]

Just wondering, what is the wattage that the XR-E can dissipate without any heatsink under an environment of 40°C ?
I guess about 300mW.

[chimo]

Doug, while waiting for Newbie's response, perhaps I can hazard a guess at some of the questions:

1, 2: See post #64 above. I believe the MCPCB is the light engine mounted with a 825mA drive current.

4. Same here.

Cheers,

[MillerMods]

Newbie, is it possible to guesstimate what the thermal conductivity of the ceramic is?

Also, what would the average delta (die to center on the back of the ceramic) be @ 350mA?

Thanks for your time.

[Anglepoise]

Newbie,
In your opinion, would it be fair to presume that a round post directly under the die, soldered or glued ( AA) to the XR-E would offer the same or better heat movement that a rectangle post that covered the whole base ( not including the electrical connections pad)

My thinking here is that it is easier to machine a round post than mill a rectangle post.

I am trying to use the data ^^^^ to formulate a strategy that will move heat, allow electrical connections, and not interfer with reflectors.

[ViReN]

NewBie..... since it's a LED's Thermal Image, shouldn't the die itself be hotter than the surrounding? looks like the die portion is cooler than the Max Temperature Point.... wondering why....

[mudman cj]

MillerMods - The ceramic is almost certainly Al2O3, which has a thermal conductivity of about 25-30W/m*K. Compare that with copper which has a conductivity of 385W/m*K and you can see why NewBie is not impressed.

Thanks for the great data, pics, and explanations NewBie. I too have worked with FLIR cameras and I understand what you are accomplishing. You may find it interesting to know that I work where the cryocoolers are made that go into those cameras.

[McGizmo]

If I understand correctly (dubious since I am likely one of the blind and ignorant), the heat from the die is passing through the ceramic to get to the metal sink pad and this resistance is inherent in the package design. In the pursuit of optimal, why don't we filed down the whole package until we approach the die itself?

Since one can easily break the LED lead connections from the botom solder pads, I don't see an advantage in putting the LED on a pedestal when you can mount it with full bottom surface contact; either with reflow to your large hunk of copper or via thermal epoxy to Al? If less surface contact is better then take it to the limit of no surface contact?!? Or is it the case that we just need to match the surface foot print of the die under the LED and any surface contact beyond this is less important? It goes against my intuition to think that reducing contact with a material that is thermally conductive makes sense in terms of thermal relief, even if the thermal conduction of the material is not fantastic.

I was under the impression that the XR-E has reached an improvement in terms of thermal conduction over previous 7090 packages. Anyone soldering to copper trace on these LED's has experienced the effect of the ceramic wicking away the heat applied!

In real world applications of flashlights and at drive levels at or below spec, how does the various means of installing the LED in the light effect the projected flux output? If one method has an advantage of say 5C/ watt over another, what is the result in flux difference?

Well this blind and ignorant person will leave this thread because of said conditions and further, I really don't have any "Cree XR-E thermal concerns".

[IsaacHayes]

In discussion with Newbie, he told me that there are better ceramic types (I think he listed some in this thread) that conduct heat better than what they are using. And that they likely will change over to them to make the led better once there is demand and more competition out there. They are still holding some cards, and close to their chest for when the time is right as he said.

I can't wait for when they decide to go ahead and release a package with all the best stuff and bells and whistles without holding back. But then I'll have to re-mod all my lights!

[Anglepoise]

Since one can easily break the LED lead connections from the botom solder pads, I don't see an advantage in putting the LED on a pedestal when you can mount it with full bottom surface contact;

My experimenting with the pedestal is purely to see if I can solder reliably to the bottom connections, so that the reflector and a very thin insulating washer does not have wire, solder, spatter etc to get in the way. Re flowing without the special equipment and my lack of knowledge bothers me a bit.

[NewBie]

As far as soldering to the package, I see zero difference in how well they wick the heat away, as compared to the old XL7090 part.

Don,
The old XR7090 part also has a thermal resistance of 8 C/W as shown here:
http://www.cree.com/products/pdf/XLamp7090XR.pdf

The die itself is larger than the old die from the older XL7090. It is 1mm by 1mm, instead of 0.9mm by 0.9mm. This means the area of the die is increased by 1.24 times larger than the old die, which would be likely to reduce the thermal resistance. Possibly they increased the area of the SiC ESD diode that they mount the die upon, which would also help thermals.

When you solder things down, and if you only had a 5C/W difference, the temperature on the die would be 15C difference. This would work out to a 5% loss in light output. However, this die is not mounted on solid copper, and has a thermal spreading resistance. So, one could reduce the thermal spreading resistance in the system, by getting the area under the die closer coupled to the heatsink and utilizing something with a very high thermal conductivity to reduce the thermal spreading resistance.


As I presently understand it, the ceramic that is used is an Alumina based ceramic.


Here is a comparision chart of the thermal conductivities (higher is better) for a few metals, and the various ceramic fillers, that primarily comprise the ceramics or compressed ceramic fired forms:

[NewBie]

Jar, the pictures are great! I am, however, in the camp that is having a bit of trouble interpreting them. I have a couple of questions whose answers might be buried in this thread but I failed to see them.
1. What is the drive level for the MCPCB thermal images?

If you look back in the thread, you will see the A19 XR-E flashlight converter measured 800mA to the LED on the ETGTech MCPCB that is mounted in the A19 XR-E flashlight.

The direct copper plate soldered one is also at 800mA.

2. For those MCPCB thermal images is the MBPCB in turn mounted on a heatsink or is it just in free space? It would seem that the former configuration would be the one most revealing of the MCPCB performance relative to having the package mounted directly on a massive heatsink.

The A19 flashlight has an overlapping rim where the edge of the MCPCB is attached with thermal epoxy, and behind the board, they also filled the ecan with thermal epoxy. They staked MCPCB into the ecan by squishing the MCPCB Aluminum material edges up against the ecan wall at sixteen points.

I have further tests underway now to learn more. On the same exact copper plate, three LEDs are wired in series, so they will each get the same exact current.
-One, direct solder to copper plate.
-One, AA thermal epoxy'd to copper plate, cured under heavy weight.
-One, ETGTech MCPCB AA thermal epoxy'd to copper plate, cured under heavy weight.

3. In the thermal images is the cursor location showing the peak temperature located over a portion of the package where the coating which equalized thermal emissivity *has not* been applied?

The cursor is on the peak (85 degrees C). The peak is on a surface where +98% thermal emissivity material has been applied. Correction factors for the coated areas have been already applied. This peak temperature is on the ring, as well immediately around the copper metal lens holder at the base of the LED. As I did not want to get the emissivty material on the lens, only parts of the copper ring were coated on accident. The line across the photo is a connection I made, and a wire is blocking the IR "light".

See the ghosted image of the visible part overlayed upon the IR Image. Notice how abruptly the temperatures transistion near the ring, and then drop on out to the edge of the electrical pad area (this is a IR image of the part mounted on the ETGTech MCPCB):

4. You seem to have an almost endless supply of great Cree emitter pics! One I'd love to see if you already have it is a XR or XR-E sectioned in a vertical plane.

Thanks for sharing your great work!

I'll see if I can find the cross-sectioned device photos I have done. It was a pain, and I ended up having to use diamond to cross-section the hard ceramic.


Isaac,

As I have mentioned before, glass does not pass 7-14um wavelengths, and thus you will not see the die temperature, or an image of it. This is just one of the reasons, that attempting to use a low cost IR thermometer is nearly worthless for measuring die temperatures, and is a lesson in futility. However, you can measure the surface temperature of the dome, if you know the emissivity of the surface.

[Moat]

This is all some wonderful and interesting information, Newbie... even if I never get the opportunity to touch soldering iron to a Cree LED. Thankyou - this "nobody" lurker appreciates your efforts here, and in so many other threads over the years!!

I have further tests underway now to learn more. On the same exact copper plate, three LEDs are wired in series, so they will each get the same exact current.
-One, direct solder to copper plate.
-One, AA thermal epoxy'd to copper plate, cured under heavy weight.
-One, ETGTech MCPCB AA thermal epoxy'd to copper plate, cured under heavy weight.

The results of this test will likely provide some very useful answers/insight for many folks here. Bravo... and looking forward to it!

[Anglepoise]

Newbie,
Look forward to see the 3 way test results.
Your hard work on this is appreciated

Below is a section of an XR-E.
What looks like a copper connection between top and bottom in the photo
is just the copper 'smearing'.Second photo is a little clearer.
There is no visible to me connection between the die and the bottom rectangular copper pad.

[AilSnail]

Quote:Originally Posted by McGizmo
If the MCPCB itself were thermally relieved in a better manner (I.E. bonded to a hunk of copper) then I believe the steady state temp would be lower than is seen in the example. With a lower steady state, would the delta be as great?

Actually, the delta would increase.

How come? If the led is colder, it is more efficient, producing less heat..

This peak temperature is on the ring, as well immediately around the copper metal lens holder at the base of the LED.

so, would it be effective to use the reflector to draw out heat from this metal ring, or is the thermal resistance from the die too low that it would matter?

That ceramic plate looks awfully thick, I wonder why? Must be a lot of C/W just through this millimeter(?) of useless(?) spacer?

[LightBright]

I believe Cree is trying to electrically isolate the LED substrate with that ceramic layer, where on the Luxeons, the copper slug is NOT electrically isolated.

The ceramic may be a manufacturing/engineering cost saving measure or a way of getting their LEDs to market in the shortest amount of time.

[ViReN]

Newbie,
Look forward to see the 3 way test results.
Your hard work on this is appreciated

Below is a section of an XR-E.
What looks like a copper connection between top and bottom in the photo
is just the copper 'smearing'.Second photo is a little clearer.
There is no visible to me connection between the die and the bottom rectangular copper pad.

Now we have one more way of heat sinking (atleast to some extent). i.e. through Al Reflector.

One more worry, it is learnt that Creamic is not 'that good' (compared to copper) conductor of heat, which kind of Ceramic is being used by Cree? any ideas?

it would have been great if they could have extended copper slug directly to the base, I am sure the thermal resistance will drop much further.

I will be seriously looking at getting multiple Cree's if they come up with a well designed product....

From My Point of View, it's True, it's bright, but's it's still a 'shabby' design (especially after looking at the C/S).. Yes I have already signed up for a GB for Q3, let's see how they perform when I get my hands over em

[ViReN]

<SnIp>

As I presently understand it, the ceramic that is used is an Alumina based ceramic.


Here is a comparision chart of the thermal conductivities (higher is better) for a few metals, and the various ceramic fillers, that primarily comprise the ceramics or compressed ceramic fired forms:

how about Epoxy + Boron Nitride... i think...many IC packagers use this for a better heat transfer (not sure how it compares to Ceramic)

it would be great if you could compare Cremic+Alumina with Epoxy+Boron Nitride

[NewBie]

Here is the setup with no emissivity coating, the temperatures are *HIGHLY INACCURATE*




Here is the same setup, with a high thermal emissivity coating on all parts, so we can see the actual temperatures accurately (NOTE, TEMPERATURE SCALE ON SIDE HAS CHANGED!):




The parts that were mounted with Arctic Alumina were bonded under weight for over 12 hours.

The copper sheet metal that the parts were mounted to was placed upon another plate to help get rid of the ~9W produced.

As you can see, the Arctic Alumina Thermal Epoxy to the copper plate nearly works as well as soldering directly to the plate. The ETGTech MCPCB doesn't work nearly as well as the other two options.

The AA Thermal epoxy XR-E has a dome temperture of 42 degrees C.
The direct solder XR-E has a dome temperature of 39 degrees C.
The ETGTech MCPCB has a dome temperature of 54 degrees C.

Notice the wire going from the MCPCB to the direct copper soldered one, and how the wire is conducting heat from the MCPCB mounted XR-E, and how it fades along the wire length.

[KWillets]

Can't you deduce the die temperature from the output spectrum? You can even calibrate if you heat the whole assembly to known temperatures and run at, eg, a 1% duty cycle.

[McGizmo]

Newbie,
Thanks for your work and effort here. In this test set, what would you estimate the relative light output differences to be between the LED's? I can't find a graph on the XR-E in any of the data sheets I have showing the reduction in relative light output based on junction temperature. In a possibly related graph I have access to, there seems to be a 3% reduction in output per 10 degrees of C. Does that sound about right?

[NewBie]

Newbie,
Thanks for your work and effort here. In this test set, what would you estimate the relative light output differences to be between the LED's? I can't find a graph on the XR-E in any of the data sheets I have showing the reduction in relative light output based on junction temperature. In a possibly related graph I have access to, there seems to be a 3% reduction in output per 10 degrees of C. Does that sound about right?

Using my calipers on the graph off their website datasheet, I see ~4% per 10 degrees.

Keep in mind, the case temp is only relative to the die temperature. And we cannot see the case temperature directly underneath the die.

The best thing to take away from this is to note the differing case temperature from LED to LED, and note the changes in case temperature at different points on the LED.

On the photo with the thermal emissive coating (the only accurate one), the dome hot spot is 54C. The average temp of the ceramic base is roughly 44C.
The delta here is 10C. (don't get mislead on the LED case, due to the solder pads on the MCPCB).

On the direct solder one, the dome hotspot is about 39C. The ceramic base temp is about 34C. The delta here is 5C.


Odd...the power input is nearly the same, within 2%, accounting for differing Vf and equal currents.


If you think about it a bit more, if there was no spreading thermal resistance, the deltas should have been about the same, with the same power input. But they are not. This would lead me to consider the idea that the die is even more proportionally hotter on the MCPCB mounted one, than what the LED case temperatures are showing.


The copper sheet metal temperature is about 31.5C.


When I get a chance at some point in the future, and I have plenty of extra XR-Es, I have some 42 guage K-type thermocouples, and it might be interesting to place them directly on the dies.

With everything already mounted on the copper sheet metal (oops), soldering on the MCPCB one was immensely easier to solder on, as compared to the other two.


KWillets,
Your technique works great for amber, decent for red, okay for green, but blue die used in the white LEDs do not shift their wavelength that much with temperature. An example, for one of the Luxeon power type LEDs dies at ~15.5C the peak wavelength of the blue was 463nm, but at 82.2C it was 469nm. One would need a method of measuring rather fine shifts, as it works out to only 0.135nm per degree C. This is an entirely different matter as compared to Amber LEDs which shift all over the place (or even red LEDs in comparision). If you had a full optics lab at your disposal, one could probably accomplish this, or build up a diffraction grating/prism, some known precise spectal sources, and take advantage of distance across a room (or optics), in order to calibrate and then measure the shift. Now, if you characterized a particular lot of LEDs, you could do what Nandaren(sp) did, as the phosphor helps amplify this, when looking at the whole spectrum. One could use the CIE co-ordinates or Kelvin temperature(less accurate). You'd do a very narrow duty cycle, to minimize heating, measure that LED every 5C change in ambient temperature (after allowing the temperatures to stabilize), and build up a chart with the LED and a thermal chamber. Not all LEDs track the same though, so there would be some errors if you didn't use the same LED. One could also use the old HP method, where you look at the Vf change (they stated something like 2.8mV change per degree C, if memory serves correctly, for blue).

Each of these methods is a project by itself to do properly.

It isn't too hard to actually do, it just takes time and care. I'd love to read the write-up, look at the photos, and see the results- if you'd like to take that project on.

As it stands, we were just looking to see if the MCPCB works as well as direct soldering, and how well Arctic Alumina epoxy the emitter to a copper plate works, when compared to direct soldering.


Results are pretty simple, MCPCB is the worst, and direct soldering is the best, with AA epoxy being close to soldering. No surprises here.


Do not forget, that these measurements do not include heat generated by converters, heat generated by batteries, and the various thermal resistances typically found in an actual flashlight.


Anyhow, it is the next day here, and it is time to hit the sack now.

[chimo]

Great work Newbie! I was waiting for this one. It's nice to see the comparison between the three.

You've already answered a question I had re power to each LED. They seem to be fairly closely matched (2%). The heat flow on the interconnect wire is quite telling as well.

On the MCPCB, is there any thermal compound between it and the copper sheet?

BTW, thanks for doing this.

Paul