Rebel 4X, no, 6X! Feel the power!

[Oznog]

I may have used a high number for the copper. Elsewhere I saw 355 W/(m*K) in the PCB context. The condition and temp range of the copper does matter and maybe I didn't have the most accurate figure.

But I wanted to figure FR4 conduction too. So the figure I have for that is 0.25 W/(m*K) from a TI app note. I've seen .36 and .4 mentioned elsewhere but anyhow I'm looking at TI's number.

So I get... 4000C/W per sq mm on 1mm board. Well 40 C/W per sq cm. Ya hardly good conduction but it does have an effect.

 

[KeithInAsia]

Oznog, your logic is weak on a number of points. I'm not going to have enough time right now to hit them all...

I'll try to hit the a few of them.....

You say the Gold plating is a waste of money. I disagree. Coper must be coated to prevent oxidation. Gold is a great conductor also, it's better than tin. It provides a very smooth surface for good contact (Tin does not). It's nicely reflective which will help inside some light fixtures. This idea that gold plating is throwing off my thermal reading is highly dubious. What I read thermally is what I feel by touch as well. The measurements make sense. Looking into the front of the unit, you will get an average of the sensor area. That is to be expected. By inspection, the unit is not much hotter up front than it is in back or on the sides. That is an effective solution. You just can't knock a common sense observation like that. I don't care how many numbers you calculate.

Let consider a very simple idea..... Look at that bottom of a Luxeon Rebel. Just how big is the surface areas of that thermal pad? It's very small. It looks to be about 2mm x 1.5mm. Just how much heat is going to travel through that point? All I have to do is match and slightly exceed that thermal area. That's it. I don't need a dozen vias to effectively match that thermal path on the LED.

Coppy is very effective -- it just is.

For my circle product (a sample seen earlier in the thread) where the screw holes bond to the top and bottom plates -- I'll bet you that I don't even need visas on that board for it to work well.

Fan speed on this test below means almost nothing related to the argument. If the heat sinking solution is good, that heat sink is going to get hot. 54 Rebels at 466mA pushes some heat -- it most certainly does.... The whole heat sinking design does get work the whole unit would over-heat. In this case whole unit warms up very evently -- that suggests that the boards are effective.

You said you had a sample -- correct? You tore one open right? Do you still have a good one to use in a real test?

If not, I'll send you a small handful if you promise to post the results here. How do I contact you for an address?

Will you run a real test?

Throw your thermal couple on there as you like. Let's see the real results and not a long laundry list of conjecture.

 

[Gomer]

If not, I'll send you a small handful if you promise to post the results here. How do I contact you for an address?

Will you run a real test?

Throw your thermal couple on there as you like. Let's see the real results and not a long laundry list of conjecture.

If only Tullio read this and took a hint lol (OT from another LED+reef light situaiton)

 

[Oznog]

Copper does not need to be protected against oxidation where it's soldered onto of course. The rest can be protected by mask or silkscreening or tinning. Actually mask IS much better than gold plating for protecting against the environment, but that's mainly gonna come into play if it's exposed to moisture which really shouldn't happen anyways.

Look at that bottom of a Luxeon Rebel. Just how big is the surface areas of that thermal pad? It's very small. It looks to be about 2mm x 1.5mm. Just how much heat is going to travel through that point? All I have to do is match and slightly exceed that thermal area. That's it. I don't need a dozen vias to effectively match that thermal path on the LED.

Who told you this? Seriously, who?? The board is an insulator. The Rebel makes a fair amount of heat yet cannot be allowed to reach high temps. We really really need low thermal resistance.

The mfg never intended those Rebels to be mounted on a single-sided board without vias for example, not at full power. That wasn't their idea. Of couse even a single-sided board with a wide copper land is vastly improved than running the Rebel bare (which would burn it up in a second or two)- but no one would ever run a Rebel bare that's not relevant. Doesn't mean it's adequate. I mean doubling the area of the pad isn't twice the minimum cooling it needs. It's still many times less than the absolute minimum to run at even 350mA.

Spreading the heat only dissipates it through the board or off the top surface itself- the top surface's thermal resistance to air is way too high and it's totally ineffective if it's got a cover over it.

If the screws are stainless then the problem is that the stainless is a terrible thermal conductor. Like bad. 20x-30x less than copper. And unless there's thermal grease between the screw and board there's a significant boundary layer. I guess you could put grease in there but... well, stainless is just not an effective thermal conductor on this scale.

So if you discarded the vias you're limited to board conduction- which is 36C/W per sq cm on 0.9mm FR4. The round doesn't have anything like that. It's got under 1/2 a sq cm per device. So over 72C/W conduction through the circle board if it didn't have vias and we assumed the copper was so thick that the spreading is "perfect". That's terrible.

Again, we need 10C/W to be "good". That is, if you want to run it at full power, maybe a bit more, with a good heatsink behind it.

As we talked about- it is nearly impossible to actually measure the pad temp with any accuracy. I tried the delta-Vf method once but was somewhat disappointed, it's very small and hard to read. With photometric equipment, yeah, you could see the degradation in output as die temp increases, that'll be fairly clear on AlInGaP types. Perhaps you could engage someone here with that equipment?

It's not the board cost, but packing 6x Rebels on there is a fairly significant cost. And this is kinda silly because you're talking "common sense" based on feel- which doesn't mean much- and bogus ideas of measurement like heatsink temps and IR thermometers. Of COURSE gold plating reflects IR, plus the area is far too large. The measurement method is nonsense.

Vias are everything. Add vias. Many, many vias. For the most part the thicker you can place them the lower the thermal resistance, but eventually it may mechanically undermine the board's strength or the mfg's Design Rules Check will kick out your submission for not following the rules.

So what are you claiming these are? If a copper board is 5C/W or 10C/W, then are you saying it's less? What power level are you claiming they can sustain by design? 1A? 750mA? 350mA? Because I feel the way this is going to go is simply going to be you revising it to "well, obviously they work fine at >100C die temp because they're in use that way".

 

[KeithInAsia]

"Copper does not need to be protected against oxidation where it's soldered onto of course. The rest can be protected by mask or silkscreening or tinning."

I disagree -- The entire bottom of the board is touching the heat sink. You don't want a layer of Tin down there. Tin is a poorer conductor of heat when compare to copper and gold. Without the gold you will get oxidation on the copper -- that is negative. You also need a very smooth surface on the bottom. Tinning does not render that.

Quote:
Originally Posted by KeithInAsia
Look at that bottom of a Luxeon Rebel. Just how big is the surface areas of that thermal pad? It's very small. It looks to be about 2mm x 1.5mm. Just how much heat is going to travel through that point? All I have to do is match and slightly exceed that thermal area. That's it. I don't need a dozen vias to effectively match that thermal path on the LED.

"Who told you this? Seriously, who?? The board is an insulator. The Rebel makes a fair amount of heat yet cannot be allowed to reach high temps. We really really need low thermal resistance."

Are you following me at all? I said -- there is a very small thermal pad on the bottom of the LED that is mainly responsible for removing the heat from the die. All I have to do is match that thermal pathway.
As small as it is -- the challenge is not that great. That is what I'm saying..... You disagree in principle?

"If the screws are stainless then the problem is that the stainless is a terrible thermal conductor."

The screw hold the gold plated copper to the heat sink. This screw is the correct choice (although titanium might be a better choice). I'm extremely dissapointed with those who product the "Star" for providing such a terrible mounting method. How do those stars sit flat with nice even pressure when the screw holes are so hard like up with the product? I have never been able to get it just right.

"So what are you claiming these are? If a copper board is 5C/W or 10C/W, then are you saying it's less? What power level are you claiming they can sustain by design? 1A? 750mA? 350mA? Because I feel the way this is going to go is simply going to be you revising it to "well, obviously they work fine at >100C die temp because they're in use that way". "

Are you stuck on numbers? Thermal resistance has not been caclulated. I make the claim that the board works. I don't know the actual numbers.
But, if you're the expert on doing these kind of measurments - I'll let you give it shot.

Tell us you're more than just a "armchair quarterback" type. Tell us that you'll take the challenge and I'll send you enough samples of the board to run a fair and complete test. Then all this guess work will be over....

Or -- are you only firm and convinced on what you "think" is correct?

Come on now -- I'm putting the challenge in front of you. Do you have the time and resources to find out how well the "rubber meets the road" on this little board?

 

[TigerhawkT3]

...[snip]
...there is a very small thermal pad on the bottom of the LED that is mainly responsible for removing the heat from the die. All I have to do is match that thermal pathway.
[snip]...

It's true that you can't do much to increase the contact between the die and the heatsink, but the heatsink does need to have some combination of 1) enough mass to sink the heat from the die and 2) a method (fins, fan, heatpipes, Peltier, whatever) to remove that heat from the heatsink fast enough to keep the whole thing from getting too hot for the LED(s). I'm sure that you knew all that and meant something different, but I probably misunderstood it. Could you clarify what you meant?

...[snip]
Are you stuck on numbers?
[snip]...

I should hope so. I'd much prefer some (accurate) figures over "seems to work okay" or "seems not to work okay."

Oh, BTW, does your IR thermometer have adjustable emissivity? If you know the emissivity of your boards' surface coating, you could get pretty good readings with it. If you let the device run until the temperature stabilizes, you could get a reasonable estimate (just a tad low) of the temperature of the LEDs, AFAIK.

 

[KeithInAsia]

"It's true that you can't do much to increase the contact between the die and the heatsink, but the heatsink does need to have some combination of 1) enough mass to sink the heat from the die and 2) a method (fins, fan, heatpipes, Peltier, whatever) to remove that heat from the heatsink fast enough to keep the whole thing from getting too hot for the LED(s). I'm sure that you knew all that and meant something different, but I probably misunderstood it. Could you clarify what you meant?"

This LED board is not designed to be any kind of full heat sink. That task is solidly placed on the heat sink for which it is mounted on. So, the primary objective is to act as a conduit to pass heat from the LED to the heat sink below. Just how much mass and density would you say it takes to make that thermal conduit?


I am not stuck on numbers..... It would nice to have some of those numbers, but even numbers are just a arbritray measurments. Once you have those "numbers", then you have to work them over for stastical accuracy and experimental error based on calibration of your measuring equipment and the inherriant affects of the measurement equipment has on the system that it is measuring, etc. Getting numbers that everyone really respects is fairly difficult.

Remember all the old claims of Hi-fi equipment sellers back in the 70's who boasted about how many watts they had at the speakers and how low the distortion was in db's? Who was really right and who was just fudging "the numbers".

I'm telling you -- you first look for a common sense observation to point you in the right direction.... and my guess that this board is looking very good ise being suppported by the fact that this board is not hot to the touch anywhere on it. And I'm touching as close as I can to the emitter.

Now, I had a few lenses fall off - and then unknowingly I touched the emitter -- that burned me. That was painful -- so certainly that area is above 100C. But no were else did I feel any discomfort.

My inferred thermal measuring device reads perfectly when I aim it as my skin or at boiling water ... there is no mysterious "reflection" error going on there.....

In fact, if you look at the picture below, any reflection would only artificially increase the reading -- would it not? That 48 degree reading is influented directly by what is being seen right inside the die - is it not? (and just for the record I moved my sensor around to achieve the highest reading possible)

So 48 degrees is a pretty good number generally.

We all have been heavily influence and brainwashed to think that metallic sandwich boards are superior because they were first on the scene for LEDs. That was directly influence by LEDs which all had an electically active base.

Many if not most of the newer LEDs are coming out with electrically isolated bases, so why not modify our thinking about the type of board that it should sit on?

Here is one last point.... Will my design be affected by the heat sink that is under it? If I mount on aluminum (as seen in the previous pictures) and then I mount a group of the same LEDs on a copper heat sink -- which group will function with great cooling? All thing being equal - a copper heat sink will make my board work better..... So now what? Oh -- the board is ok now? No -- the board just the same as it was.

What if I use a more superior heat sinking compound between this board and the heat sink? Will the board be better? It's still the same board.

So, I think we need to look at the entire picture.

Having a guideline C/w measurement for this board would be interesting, but it not the only factor in the over-all LED heat sinking system. So, tell me again why we can't consider anything else without those C/w numbers?

 

[VanIsleDSM]

Well.. a die temperature over 100C, and a heatsink temperature of 48C. What does that tell you about well your boards are conducting heat?

I agree MCPCB aren't the best, that's why I built these boards:

Solid copper for the heat path. If you reduce the thickness of your FR-4 and add many more tiny vias.. then you'd had a really nice product I think.

But as it is.. without using numbers, and just my common sense as you like to do.. I think the board, as it is, will not conduct heat near well enough for the kind of performance I am looking for.

 

[Oznog]

But what would I be testing? Suppose this shows up at 20C/W. You're going to say it "feels cool to me" and "the LEDs are blinding bright" and "it's in use and works fine so whatever number it is it's GREAT cooling!" These claims are nonspecific. Do you want to claim it's superior to MCPCB at least? Better than the 5C/W-10C/W range? How much better do you think it is? What power level do you claim these can be run at continuously or peak?

I did think about it... if I score the top and bottom copper into 6 pie slices, I can isolate the thermal performance of each device pad. That would accurately show the thermal performance without loading the board up with 6 devices. I don't have 6 to spare.

The IR therm wouldn't work even if you could adjust the emissivity unless you also had one which could focus to a fine point. The temp at the pad is the only relevant temp and an IR therm 1" away has a measuring zone at least 100x too large to give any information on pad temp or thermal resistance. It's reading temp across the board and LED face which doesn't tell you anything (and the emissivity question wasn't resolved anyways).

Yes C/W is everything as far as thermal performance goes. Cheap and easy to mount and high component densities can of course be good features, but thermal performance is all about C/W to the heatsink.

Actually the HiFi sellers are a good parallel- sure people can give false numbers but anyways guys here won't buy it. The alternative was numberless descriptions of the results of how wonderful $1000 8-ga silver-plated oxygen-free Litz-wire speaker cables "sound" when electrical and acoustical analysis shows no difference whatsoever. And you're kind of in that class unfortunately, going on what you "feel" and see as "common sense" theories about thermal conduction.

I said -- there is a very small thermal pad on the bottom of the LED that is mainly responsible for removing the heat from the die. All I have to do is match that thermal pathway.
As small as it is -- the challenge is not that great. That is what I'm saying..... You disagree in principle?

Yes I disagree. The pad on the Rebel is already a restriction, not too bad though. However, any additional restrictions will increase die temp and reduce the power level that can be sustained as well as the efficiency and life. For example, the junction-to-pad resistance on the Rebel InGaN is 10C/W. The MCPCB is maybe 8C/W, a moderately large sink 5C/W. That's a total of 23C/W. My point is that even though the MCPCB and heatsink were a lower resistance than the pad, they contribute heavily to the bottom line total for thermal performance. Thermal transfer performance is a cumulative number, the performance is not simply equal to the weakest link in the chain.

I'm extremely dissapointed with those who product the "Star" for providing such a terrible mounting method. How do those stars sit flat with nice even pressure when the screw holes are so hard like up with the product? I have never been able to get it just right.

I don't understand. The Star mounts just fine! It has holes, the AS board has holes, and either one needs alignment of PCB holes over the heatsink holes. It actually had 6 and doesn't necessarily need all of them you can just screw down 4 in most cases- or 2 or 3 probably. Could you describe this problem?

 

[TigerhawkT3]

Well, I typed up a huge reply (epic, really), but the board ate it. "You have logged in since loading this page," or something. Never seen that one before.

I'll try to make a condensed version.

...

Okay, I read the preview of the condensed post, and it came off really argumentative. I'll keep this short and sweet. Here's draft 3:



People do care about measurements (just look at any datasheet, product manual, etc.). We also care about the methodology used to obtain those measurements. "Fudging the numbers" is not entirely dependent on there being numbers attached. Someone could say, "these speakers sound great" and be lying. The point is that we hope that people don't lie.

A finger is not a very good thermometer.

A more reflective surface (lower emissivity) makes for a low readout in a fixed-emissivity IR thermometer. The material will be hotter than the readout shows, as Oznog's experiment with the aluminum and the stove demonstrates.

If an emitter is hot to the touch, so should be the board and possibly the heatsink (depending on distance to the emitter, size of the heatsink, etc.). If an emitter burns you but the surrounding area feels fine, that means the heat is staying in the emitter.

Different heatsink compound, different heatsink material, etc. may help the performance of the system, but they don't affect the performance of your board.

C/W seems like one of the few relevant quantifiable properties of such a product as these mounting boards. I still can't understand why you don't care about it.

 

[frenzee]

Here's my 2 cents:

Your finger is definitely not a good heat sensor. A rebel or Cree running at 1 amp will easily burn your finger if you put it right on the dome, no matter what the jT. There is almost 1W of energy coming out of a area the size of a match head.

An LED is probably not the best candidate for testing a board's C/W. Why not use a metal film resistor where you can accurately measure how many watts of heat is being generated? Of course it'll have to be properly thermal epoxied or soldered to one side of the board and insulated from the surroundings air to mitigate the effects of emissivity and convective cooling, and you'll need two thermocouples, one on each side of the board.

Infrared thermometers typically have a 15- or so degree viewing angle and basically take an average reading of whatever comes in within that angle. So unless you are looking at a surface with a uniform temperature, you can't really tell what the peaks and valleys are. The only way to get an accurate reading, in the context of an LED, would be to use a digital thermograph and those things are pretty expensive.

 

[Oznog]

Here's an example of thermal vias for a Cree, that's a 1W nominal device just like the Rebel:

 

[Frobe22]

Bare metal acts like a mirror when you use IR thermal measurement. (You may even measure your own skin temperature if you have 90° angle).

When I use a thermal camera (superiour to a simple IR thermometer) I have to add coating or a thermal pad to metal surfaces, otherwise they just show up at room temperature or reflect any hot surroundings. The pad/tape/paint shows up in the display as a bright beacon compared to the dark bare metall. I know it is possible to adjust for emissivity, but most metals don't have a specific number, so the easiest solution is to "patch it" to get correct temperature.

Glass is a barrier to IR Thermal measurements, and will show up at room temperature. Cree LED lenses are glass, but I think Rebel use some type of silicone?

Optimal junction temperatures are below +80°C, and maximum are about +150°C.

 

[KeithInAsia]

Bare metal acts like a mirror when you use IR thermal measurement. (You may even measure your own skin temperature if you have 90° angle).

When I use a thermal camera (superiour to a simple IR thermometer) I have to add coating or a thermal pad to metal surfaces, otherwise they just show up at room temperature or reflect any hot surroundings. The pad/tape/paint shows up in the display as a bright beacon compared to the dark bare metall. I know it is possible to adjust for emissivity, but most metals don't have a specific number, so the easiest solution is to "patch it" to get correct temperature.

Glass is a barrier to IR Thermal measurements, and will show up at room temperature. Cree LED lenses are glass, but I think Rebel use some type of silicone?

Optimal junction temperatures are below +80°C, and maximum are about +150°C.

Bare with me people, I have a few posts to make....

I hear your point, but I also had my hands on it. The reading made some sense. Could they be a few degrees off? Yes, but you have to look at the general results and factor in the experiemental error.

The results are that the front of the unit close to the LEDs were not that much hotter than the back of the unit.... Conclusion -- heat sink and mounting PCB are working in a satisfactory way.

There is no trick photography here.

And granted, I do not have a method to measure the junction tempertures. Not many people do.

 

[KeithInAsia]

Here's an example of thermal vias for a Cree, that's a 1W nominal device just like the Rebel:

This generally is a decent design, but I would have done it differently.

I think putting the screws far away from the die is a mistake.

Also, the thickness on this copper looks rather thin. You can see thick copper by the change in light reflection next to the traces. LED Dynamics favored a more standard copper thickness and compensated with a larger number of vias.

Here is a big mistake. They should have never interrupted the flow of heat away from the LED by putting traces around it. That boxed in their design. Traces should have exited the design left and right and heat plane should have extened out to the screw holes. They were more concerned with having + and - outlets on both sides of the board over having a better heat sink.

I also would not have insulated the thermal plane with a solder mask. I favor gold plating so that heat has a chance to radiate from that area.

You may disagree with my points. These are just the things that I see.

I would reduced the number of visas and increased the copper. In fact, I in about 3 weeks I will have a 22mm square board that will mount this LED. And I will have 22mm designs to accept XP and MC-E units as well. Each one with the same approach with heavy copper, mounting holes closer to the die, minumum solder masking, and fewer vias.

 

[KeithInAsia]

Here's my 2 cents:

Your finger is definitely not a good heat sensor. A rebel or Cree running at 1 amp will easily burn your finger if you put it right on the dome, no matter what the jT. There is almost 1W of energy coming out of a area the size of a match head.

An LED is probably not the best candidate for testing a board's C/W. Why not use a metal film resistor where you can accurately measure how many watts of heat is being generated? Of course it'll have to be properly thermal epoxied or soldered to one side of the board and insulated from the surroundings air to mitigate the effects of emissivity and convective cooling, and you'll need two thermocouples, one on each side of the board.

Infrared thermometers typically have a 15- or so degree viewing angle and basically take an average reading of whatever comes in within that angle. So unless you are looking at a surface with a uniform temperature, you can't really tell what the peaks and valleys are. The only way to get an accurate reading, in the context of an LED, would be to use a digital thermograph and those things are pretty expensive.

Well, granted my elbow would be a better heat sensor... (you guys know this don't you? This is the reason you see guys squirting baby milk on their elbows in old movies. We have better thermal nerve endings down there.)

I think my fingers do just fine for a raw inspection....

And NO - I don't feel any over-heating when I touch a lens dome of a Rebel. Not at all. There had been a couple of domes that I knocked off and then touched the actual die. That was painful. I don't then any mounting technic can prevent that.

The idea of testing with a metal film resistor may be interesting. What is the determination of watts? Do you measure the voltage drop and assume all waste energy is being turn into heat?

I agree with the digital thermograph. I'd love to have one those because you could see the full spread of heat dissipation. I agree fully.

 

[KeithInAsia]

I don't understand. The Star mounts just fine! It has holes, the AS board has holes, and either one needs alignment of PCB holes over the heatsink holes. It actually had 6 and doesn't necessarily need all of them you can just screw down 4 in most cases- or 2 or 3 probably. Could you describe this problem?

I never seem to get my mounting holes in the right place and the screw can easily be off center and hanging off the edge of the plate. I use a drill press and my holes are staight, but they are never on center and these open holes don't natrually allow you to center up your screw to any degree.

If there was a full encapsulated hole there, a hole drilled slightlyl off-center just a bit would kind of self correct.

So, that is just a gripe. I like fully encapsulated holes and further -- I would want a drilling template to nail those holes with some precision.

Correct mount of screw is a significant contribution of good surface contact and good cooling....

I provide drilling templates with my products that accept a 1mm hole so anyone can get the hole drilled correctly. They tap a start with a 1mm bit and then follow up with a 2.7mm bit to make the final holes. All screws are on center and perfect. That is thinking ahead and giving the customer the best chance to do a great job.

I fault the companies who make the "star" boards with lack of quality in this regard. The are all designing like zommies following each other down a bad path. Are the bean counters making them optimize the surface area out of these designs?

 

[KeithInAsia]

Thermal transfer performance is a cumulative number, the performance is not simply equal to the weakest link in the chain.

Agreed, every link in the chain has an effect, but the weakest link in the chain is the bottle neck to effective heat sinking.

I don't need a hunderd visas to keep up with the smaller pads on these Rebel parts. I believe there is a dimishing return after a certain point.....


People -- since none of you have actually seen this product in action, you are somewhat disadvantaged in your perspective.

Further, I see no takers to my offer of a few free boards to make your own real tests.

I think this is rather foolish and short sighted to tear a board part and comment about it's strengths and weakness when in fact you never put it through a real test.

Thats like pulling the wheels of a Ferarri, slicing a cross section of a tire and saying that it can't possibly accelerate from 0 to 60 in 4.3 second because the tires shouldn't work under those conditions.

You alll sound very close minded.

Here is a little reality check about the supposed unbeatable metal PCB.

At some point there will be a delamination of the layers. Its going to happen. Obvoiusly at this point it won't matter how good the C/w thermal conductance it.

What I think is interesting is that the diaelectric layer looks very much like fiberglass in nature. Hum....

I suppose this could be my fault since I soldered a fairly large wire onto the pads. This weight could have stressed the top layer. Of course there could have also been over factors of wind. Or moisture worked its way in between the layers, etc.

 

[KeithInAsia]

You built this for the MC-E right?

One the surface, a slab of copper of course is unbeatable for heat sinking. You will have some trouble heating this this enough to solder the LED down to it. That is going to be a bit challenging....The solder will be over-temp for long periods of time driving the flux out of the joint.

But I think you have another issue lurking here....

Pan head screws are going to short out your traces when you screw this unit down --- Or am I missing something? Is there a particular reason that you placed the traces right on the holes?

 

[snarfer]

It seems to me that an easy (and cheap) way to test out boards would be to use the relative forward voltage of the LEDs. Heat the entire assembly to various temperatures and measure forward voltage at a known current. You would have to pulse the current for very short time in order to avoid allowing the LED to heat beyond the measured temperature.

Then when you have various datapoints at different temperatures you can test forward voltage at actual operating conditions and have an accurate idea of what the die temperature is. A digital thermometer will give you heat sink temperature and the math to extrapolate to temp rise per watt is simple. Since temp rise per watt is not necessarily linear it would be useful to measure under various drive wattages and ambient temperatures.

Procedure is further explained here.

Keith I don't think anybody is really interested in sacrificing still hard to obtain (in small quantities) Luxeon Rebels for your test. Nevertheless I agree with your sentiment that the proof is in the pudding. I don't know if I'd go so far as to compare it with pulling the tires on a Ferrarri in order to calculate maximum acceleration of the vehicle. I would say that your product is more similar to the tires if anything, and people would really like to know whether they are racing slicks or retreaded economy tires before they get in the driver's seat.

As a manufacturer I think this is really some testing you should do in an accurate and repeatable way. If someone is going to incorporate your product into a luminaire of any sort they will need the information in order to determine the parameters of the thermal design. This is especially true for natural convection applications. At least if there is a fan you could turn it up or get a more powerful one, but if you are using natural convection you'd have to redesign the whole heatsink in case of excessive die temperatures.

If you want I can refer you to an engineer at an independent testing laboratory in Bangkok. He does exactly this sort of testing, with thermocouples, for major international OEMs, on a daily basis. Also his rates (for side jobs) are quite reasonable.

Finally I have to point out that the feedback you're getting here is really very valuable. Ignoring for the moment the merits of the thermal issues raised, what you can extrapolate from this thread is important information about what a certain number of your potential customers would purchase. If I were in your shoes I'd be thinking along the lines of redesigning my product to fit what the customers want, rather than spending too much energy trying to persuade them that the existing product is already adequate. If there is an unmet market need for a commercial product that will allow you to mount your LEDs to a heatsink through a piece of solid copper, well then you better believe someone is going to fill that need sooner or later.