100 Lumen Rebel technical evaluation

[evan9162]

Good morning.

I received my 100 lumen Rebels yesterday, having only ordered them on Tuesday. That was with standard ground shipping even.

The bin I received is NVND. The new binning scheme for Rebels puts these parts as such:

N = 100-120 lumens
VN = Up and to the right of the W0 color bin (see binning chart)
D = 3.03-3.27V Vf

These binning values are at If=350mA.

Not bad - color probably could have been better, but that color bin probably also lends it self to a slightly warmer and more natural color output vs. something very cold, like a Y0 or XP binned emitter would.

The lumileds description for the new high output rebels mentions new manufacturing techniques - new thin film technology that eliminates the sapphire substrate, and roughening of the emission surface to improve extraction efficiency. Given the mention of new manufacturing techniqes, we'd expect some visible differences between the new rebel and the old, and indeed, there are.

Here is the old rebel on the left (50 lumen white) vs the new rebel on the right (100 lumen white)

The new rebel appears to be using a different phosphor mix and application technique. Whereas the entire surface of the package of the old rebel is covered in phosphor, only the die and metal circuit strips of the new rebel are coated in phosphor. I verified this by shining a blue LED at both. The entirety of the old rebel fluoresces, while only the die and electrical traces on the new rebel fluoresce.

In addition, the die structure of the new rebel is clearly visible through the phosphor, compared to the old rebel. This indicates a new phosphor mix. If the old phosphor mix were applied this thinly, the output color would have a tremendous blue tint to it.

The thinner phosphor and removal of the sapphire substrate in the construction of the die should also have an effect on the apparent height of the die as well. While difficult to see, this shot shows it best - the die of the new rebel is definitely shorter by a small amount:

This should affect the beam profile slightly, causing the beam of the new rebel to project more forward relative to the old rebel.

I attached the new rebel to my angular measurement rig, and created a beam profile. Here is the new rebel beam profile, compared to the old:

Indeed, the new rebel does project slightly more forward (slightly more narrow beam) than the old rebel does.


So the big question is: Do the new rebels perform as claimed? Indeed they do. I ran the new rebel through its paces at various currents, calculating the luminous output at each.

Here is the raw data for the new rebel's performance:

Current (mA)	0.1	30	130	310	670	980	1260	1570	1930
--------------------------------------------------------------------------------------
Vf		2.55	2.72	2.96	3.11	3.26	3.34	3.39	3.44	3.51

Watts		n/a	0.08	0.38	0.96	2.18	3.27	4.27	5.40	6.77

Lux		n/a	n/a	11.6	25.0	45.8	59.7	70.6	79.2	85.6

Lumens		n/a	n/a	45	98	179	234	276	310	335

Lumens/W	n/a	n/a	118	101	82	71	65	57	49

Here is the new rebel, compared to its competition: The Seoul P4, Cree XR-E, Luxeon K2, The older Rebel, a Luxeon V, and even a lowly Luxeon III:

First, Vf:

The new rebel has the lowest Vf of all the competing products. You can see the difference between the new rebel, and the old Luxeon III - the new rebel stays much flatter, reducing power dissipation at high currents.

Because of its serial-parallel arrangement, the Lux V has a Vf twice that of all the other parts. I've omitted it from this graph since it would change the vertical axis, reducing the resolution for the remaining plots.


Next, output:

Not too long ago, W-bin Lux Vs were highly sought, and demanded a healthy premium (upwards of $40). Today, a $6 rebel bests a W-bin Lux V in terms of output at a given current (and beats it hands down in terms of efficiency). The rebel easily breaks 300 lumens at 1.5A. This is important. It means Lumileds can slap the same die/phosphor combination in a K2 right now (which is rated for 1.5A), and market a 300 lumen, single-die LED.

Output really drops off at the near-2A mark. With output of less than 350 lumens, this tells us that there's still a ways to go before we hit the 500 lumens at 2A mark that Lumileds has promised. We'll have to wait and see how long it is until that part is released from Lumileds. After what happened with the K2, there are still many who are (rightfully) gun-shy about Lumiled's announcements. Hopefully this new rebel isn't the Q3 product release that Lumileds has alluded to in their press release about the 500-lumen parts.


Finally, here's a unique way to look at efficiency: "Droopyness". I'm including this, because one of the big announcements from the early year press releases was that of Lumileds claim of fixing "droop", which is a dropoff in efficiency (lumens/W) at high currents.

If the new rebel has this anti-droop technology, then we should see its efficiency stay relatively higher at high currents compared to other devices.

The droop factor for a given current is the efficiency at a test current, divided by the efficiency at 130mA, or:

Df = E(test)/E(130mA).

So if the new rebel does indeed have anti-droop technology, then its efficiency at high current shouldn't drop as much from the efficiency at 130mA, compared to other products. Let's find out:

In fact, the new rebel suffers from "droop" the exact same amount as other products in its class. The K2, Seoul, and Cree all exhibit the same level of "droop" as the new rebel. While the old rebel appears to droop more, this is due to less-than ideal heatsinking for the old testing, as the droop graph for the old rebel matches all the others until high current (1A and greater).

So we can conclude that either A) the new rebel does not have the "anti-droop" technology, or B) such claims are bunk. Who knows? We won't really know until Lumileds directly claims that a product does contains such technology, and we can test it ourselves.

So, there we have the technical bits. As far as a conclusion?

Well, is it worth swapping out your Seoul or Cree parts for a new rebel? Probably not.

Is it worth using a rebel instead of a Seoul or Cree part for a new build/mod? That really depends on you - the rebel can be a real PITA to deal with, not only because it has a different mounting method than we're used to (power and thermal pads are on the bottom), but also because the thing is so damn small! So the decision to use the rebel depends on if you're willing to work through the unique mounting requirements, and the possibilty of losing a $6 LED if you happen to sneeze

However, it does finally make the rebel line a more compelling part to use, now that it directly competes with the brightest power LEDs. The diminutive size is either a curse or a blessing, depending on what you're trying to do. If you're trying to cram a high brightness part into a tiny space, you now have another option.

Just like with my last evaluation of rebel parts, I'm likely to not use these very much. However, I hope this means that Lumileds will soon bring the K2 line up to the performance level of the new rebel.

 

[McGizmo]

Well done and clear comparative work! Thanks!! :thumbsup:

 

[kiely23+]

thanks evan... :twothumbs

 

[LukeA]

You can solder to the top of the Rebel, you just have to remove some silicone with a pin.

 

[OhMyGosh]

evan9162: Thanks a lot for all the info! WOW!

It is very interesting to see that the lumen/watt for different drive levels is similar to current LED's. Really amazing you did all this in one day.

Did you get these from Future? I have 10 of them but can't find any info on what bin they are. I am still busy destroying cool40's learning to solder them. - oops just saw the NVND is on the invoice but not the package.

LukeA: that tip is just what I needed! thanks

 

[IsaacHayes]

Great work! Was the Cree you used in your test the old 3 bond wire or the new 4 bond wire ones?

Makes me want to get one to experiment with. Heh.

 

[McGizmo]

I would like to add some preliminary optic observations in regards to the Rebel. Although the package itself is quite small, the image size and location is such that it is more comperable to the larger LED's than one might initially expect. To bring the Rebel into focal allignment with most of the reflectors we are familiar with, one needs to bring the whole LED into the reflector; package included. This means that in many cases the heat sink itself also needs to come inside the parabolic chamber.

Now one can design a reflector with a very short focal length and small enough that it seats down on the package and comes close to the dome of the LED. I designed one such reflector that has a 10 mm OD and fits down as one would expect a reflector to mate to this LED. I have production pieces now and they work nicely with the 80 lumen samples I received a while ago (I was slated for some of these 100 lumen parts but a rep brainfart and or loss in the mail have left me still in waiting). So, neat little reflector and neat little package.

BUT.

The beam distribution I see from the Rebel coupled with this McR-10R reflector is very much akin to the KL4 with its 5W LED and 20 mm reflector. Spot and spill angles are very similar as is the transition from spot to spill in terms of intensity. I said BUT because I believe many expect that the tiny Rebel will lend itself to a tiny "throw monster". I do believe that one can make a tiny light that competes with the KL4 in terms of flux and lux and for those who appreciate the KL4 there is clean potential for shrinking the overall package.

If you want to come up with a light that is on par in terms of beam angle and level of divergence comperable to the Seoul P4, Luxeon or XR-E with the Rebel, you will be working with the same sized optics.

The Seoul 1/2 watt with its tiny die is the candidate for a tiny "throw monster".

I am no expert by any means and what I have said is based on a limited understanding and some empirical investigation.

 

[jtr1962]

Even though it appears that the Rebel doesn't use anti-droop technology, the efficiency of 71 lm/W near the maximum rated current is very impressive. This basically means you can use these for home interior lighting, and even including driver losses of 10% of so they will still be better than most CFLs when giving the maximum amount of light per LED. No need to underdrive them at all in the name of efficiency, although if you do you could approach T8 linear tube efficiency. 234 lumens is enough to more than replace a 25 watt small base incandescent (those are horribly inefficient) while using maybe 1/7 of the power.

I can't wait to get my hands on a few of these so I can run my own testing.

 

[evan9162]

Great work! Was the Cree you used in your test the old 3 bond wire or the new 4 bond wire ones?

Makes me want to get one to experiment with. Heh.

The 3 bond wire version. My XR-E isn't the cream of the crop, just a P4 bin.

 

[filibuster]

No wonder Cree has been marching up the flux bin chart so quickly lately with new bins coming out before they are even announced as available.

 

[Gryloc]

Evan,

Thank you for the thorough review of the newer Rebels! It is nice seeing some real data every now and then! It is also great seeing a dedicated truth-seeker. Your review was so detailed and it shows how technology is really moving along.

I was wondering about a few things first. Even though it is a PITA to solder, but once you have just that one heat-sinking thermal pad soldered to a chunk of copper (or whatever), did you find the thermal transfer to be adequate when driving it up to specs (700mA)? Is there a way to find how hot the emitter gets, or does it get too hot to touch? I was trying to think of ways to increase thermal transfer, like having a little thermal epoxy between the thermal pad and the two +/- pads. How did you set it up?

Do you think that one could file off the thermal pad and two contacts (for a perfectly flat surface) and just use a super-thin layer of thermal epoxy? You would then connect your power wires to the top of the LED where the traces are. With the pads still in place, and with the use of epoxy, I figured that the surface would be too rough and there would be too many medians for the heat to travel through (heat -> ceramic substrate, to adhesives for the traces, to copper pad trace, then to epoxy, and so on). It sounds stupid, but it may allow more beginner modders to embrace this Rebel. I suppose this technique is like how you would epoxy a Cree XR-E. I was just wondering -hoping that there may be an improved method of mounting these. If only they made that copper thermal pad a little bigger.

I noticed that you have a decent testing rig, but how do you measure lumen output? Do you have access to an integrating sphere, or did you make your own? I was always wondering about that.

How did the Rebel like the 1000mA to 1500mA, or even the 2000mA? Did things still stay cool at these levels? Were you at all concerned about heat at any specific current? I know that the LED die can handle the current (you proved it -and LED dies are tough), but I was hoping that the package could handle the transfer heat. Did the tint shift any? If so, at which current level? I was hoping that the new phosphors used would hold up to intensity-degrading heat. Do you think that the Rebel uses a phosphor similar to what Seoul uses on their P4?

Finally, is it possible if I can use your data to compare it to other LEDs? I am making a spreadsheet that compares data with what other testers found, as well as the original datasheets. There will be several graphs comparing luminous flux, intensity at the different currents, efficiency (lm/W), and forward voltage. I would like it to be very complete, while being as accurate as possible. With this preadsheet, sometimes you are able to see variations in the heatsinks used, bins of LEDs, etc. It would be very useful. When I finish, I plan to post the spreadsheet on CPF for all to use. That would be very nice of you. I am still waiting on jtr1962 to post his results of the Rebel 0100's that he is getting.

Thank you. There is still many things to learn about these new beasts. I need to order a few for myself to try out. I wish I had a light meter and better equipment to show my own results. I would like to use them up to 1500mA (or more depending how they performed). It would be very nice to use the awesome efficiencies of this thing at over 1A (70lm/W at 1A down to 45lm/W at 2A). It looks as though I may have to mod my quad M*g again and up the current more.... :duck:


-Tony

 

[Gryloc]

McGizmo,

Sorry this is not the correct thread, but can you give some more information about this Rebel reflector in your "McGizmo" zone of CPF? I had some questions about it and how it compared to your old McR8. It sounds great. Finally an optic/reflector that takes advantage of the Rebel's small size! I also had some technical questions about the reflector for the Rebel. Thank you.

-Tony

 

[evan9162]

I was wondering about a few things first. Even though it is a PITA to solder, but once you have just that one heat-sinking thermal pad soldered to a chunk of copper (or whatever), did you find the thermal transfer to be adequate when driving it up to specs (700mA)? Is there a way to find how hot the emitter gets, or does it get too hot to touch? I was trying to think of ways to increase thermal transfer, like having a little thermal epoxy between the thermal pad and the two +/- pads. How did you set it up?

Here's how I mounted the rebel to my testing heat sink:

I used some thermal epoxy and epoxied the rebel, hanging halfway off a strip of 1/16" aluminum. I made sure there was a minium of thermal epoxy on the rebel, then pushed down hard on the corners of the rebel to ensure as thin as possible of a layer. I also ground a couple of round notches for the screws. Finally, I hand soldered the wires to the underside of the rebel. If I were to use the rebel in a real application, I think I would consider this method first.

On my previous rebel testing with the 50 lumen part, I was concerned about damaging the part, and didn't press down as hard. This left a much thicker layer of epoxy between the rebel and aluminium. As you can see from the output dropoff, and "droop" graph, this had a pretty big impact on the high current scaling. So keeping as thin of a layer of epoxy as possible is a must for the rebel, due to the miniscule contact area. You can get away with a thicker layer of epoxy on other parts, like the Seoul, K2, and Cree, because the contact area is so much larger (probably up to 10x as large).

Do you think that one could file off the thermal pad and two contacts (for a perfectly flat surface) and just use a super-thin layer of thermal epoxy? You would then connect your power wires to the top of the LED where the traces are. With the pads still in place, and with the use of epoxy, I figured that the surface would be too rough and there would be too many medians for the heat to travel through (heat -> ceramic substrate, to adhesives for the traces, to copper pad trace, then to epoxy, and so on). It sounds stupid, but it may allow more beginner modders to embrace this Rebel. I suppose this technique is like how you would epoxy a Cree XR-E. I was just wondering -hoping that there may be an improved method of mounting these. If only they made that copper thermal pad a little bigger.

I think the key with the rebel is to keep it as simple as possible - these parts are small...damn small. I can't imagine trying to grind or cut anything on the parts, so my plan is to leave them as-is, with no modifications. So for me, the hanging-over-the-edge mounting technique is what I plan to persue.

I noticed that you have a decent testing rig, but how do you measure lumen output? Do you have access to an integrating sphere, or did you make your own? I was always wondering about that.

I'm using a lumen integration technique, where I use an angular measurement rig to rotate the part under test relative to the light meter. This is how I generate a beam profile of the bare part. I then use some geometry to find out the area of a set of concentric spherical caps at the measured angles, given the area of a partial sphere of a given angle. I then use the definition of lux (1 lux = 1 lumen / square meter), and sum up all the parts. In essence, what I do is to create a "lumen factor", where I can measure a bare part on-axis, then multiply by a constant to get luminous output.

I used this method to measure the output in lumens of the CPF test lights from the light meter testing thread. We later sent those lights off to a test lab to be measured on a real, calibrated, integrating sphere. All of my calculations for lumens were within 5% of what the lab produced. So I'm quite confident in my measuring methodology.

How did the Rebel like the 1000mA to 1500mA, or even the 2000mA? Did things still stay cool at these levels? Were you at all concerned about heat at any specific current? I know that the LED die can handle the current (you proved it -and LED dies are tough), but I was hoping that the package could handle the transfer heat. Did the tint shift any? If so, at which current level? I was hoping that the new phosphors used would hold up to intensity-degrading heat. Do you think that the Rebel uses a phosphor similar to what Seoul uses on their P4?

The rebel seemed to handle all of the current levels well, except for the 1970mA level. I think Lumileds rates the rebel at 1A because of package power dissipation constraints, rather than what the die can handle (since the thermal pad area is so darn small). Since the die is directly mounted to the package (no bond wires), you're definitely in no danger of fusing open bond wires - so you would have to do serious damage to the die in order to actually kill the LED.

You can tell if your LED is starting to seriously heat up in 2 ways:
1) Since the Vf of an LED drops when it heats up, then you can monitor the Vf after powering it up to a certian current level. If it drops rapidly, then the LED is heating up quite a lot, and you're probably driving it at a very high level compared to what it can handle thermally. By rapidly, I mean dropping 10mV every 2 or 3 seconds, and dropping 100mV in the first 10-20 seconds. If it only drops 10 or 20mV in the first minute, then you're probably doing okay.

2) The light output also drops with increasing temperature, so monitor brightness with a light meter. My guesstimate is that dropping 1% every 10 seconds right after powerup indiciates pretty high drive level, and higher temperatures.

Finally, is it possible if I can use your data to compare it to other LEDs? I am making a spreadsheet that compares data with what other testers found, as well as the original datasheets. There will be several graphs comparing luminous flux, intensity at the different currents, efficiency (lm/W), and forward voltage. I would like it to be very complete, while being as accurate as possible. With this preadsheet, sometimes you are able to see variations in the heatsinks used, bins of LEDs, etc. It would be very useful. When I finish, I plan to post the spreadsheet on CPF for all to use. That would be very nice of you. I am still waiting on jtr1962 to post his results of the Rebel 0100's that he is getting.

Definitely, please do. I post this kind of information for the benefit of everyone on CPF.

 

[Anglepoise]

Thank you very much for the info you have given us. Much appreciated.

 

[Gryloc]

The rebel seemed to handle all of the current levels well, except for the 1970mA level.

So, you were saying that over 1620mA, the Rebel wasn't very "happy". What was going on at that 1970mA level? Were you talking about heat? Anything else wrong? I wanted to know if I would be able to thermally attach the Rebel to a good heatsink well, could the LED handle the higher currents? I would love to hook this to a water-cooled heatsink and really crank up the amperage (2.5A? 3A?) until the efficiency drops to the 30lm/W that the old K2 would operate at. Would you be willing to do such a test as this? Sure the intensity levels off some, but why not go further? It seems as though heat is the only limit. That would be interesting.

I would have to work on "shaving" off those three bottom pads. Maybe an hobby knife would work. Filing would be tough, as would using a cutting wheel on a dremel (shooting Rebels lol). If you are using epoxy on a small amount of surface and it is doing well, then epoxy over the entire bottom surface may be a little better. The ceramic substrate is a good heatsink, so having as much of the surface potted to a heatsink as possible is key, otherwise it seems like a waste. It is like applying thermal epoxy to only the right half of a Luxeon emitter slug. Now I want to really try this out.

Did your power supply have fixed current levels? I was wondering why it had steps with odd current levels (odd to us high-power LED guys atleast). I knew there was a way to measure lumens with a light meter. Once I get mine (some day), I will check back with you on the exact process. Good job with your close readings compared to calibrated equipment.

When you measure the lumens, do you have to read the lux at 5 degree increments for the entire 90 degrees? Does that mean that you have to have the LED running almost constantly? Hopefully the Rebel doesn't get hot during that time. In that case, do you think that the lumen output could be just a little better at higher currents if the luminous flux reading was taken at one time? Well, now I think of it, this heat could be a good thing because you get more "real-life" results by having the LED on for longer to allow the temperature to stabilize.

Finally, if you would like to see more of the Rebel in all its glory (the older 0050 part, atleast), then you may want to check out this link:

http://www.candlepowerforums.com/vb/showthread.php?p=2065408#post2065408

It is a couple weeks old, but I guess not too many were interested in it at the time. He showed some pretty good close-up pictures of the naked (although phosphor stripped) Rebel die. Maybe McGizmo can benefit from this. The die on every LED is so small (the 1in X 1in), but the Lambertian dome really magnifies the die. If the Rebel had a truly 180 degree dome, would it collaminate better with a reflector? Thanks again for all the information. Keep up the great work! I hope to see more results of this Rebel.

-Tony

 

[SemiMan]

I think for good heat transfer it is essential to solder the rebel as opposed to epoxy. Solder is roughly 5-10 times more thermally conductive than thermally conductive epoxies. It is not easy to do by hand, but the heat plate method works and if it is a small piece, you can probably hand solder if you have a good iron and can heat the whole piece.

Semiman

 

[evan9162]

So, you were saying that over 1620mA, the Rebel wasn't very "happy". What was going on at that 1970mA level? Were you talking about heat? Anything else wrong? I wanted to know if I would be able to thermally attach the Rebel to a good heatsink well, could the LED handle the higher currents? I would love to hook this to a water-cooled heatsink and really crank up the amperage (2.5A? 3A?) until the efficiency drops to the 30lm/W that the old K2 would operate at. Would you be willing to do such a test as this? Sure the intensity levels off some, but why not go further? It seems as though heat is the only limit. That would be interesting.

Heat was certianly the limiting factor, but there's probably not much that can be done about it. At 1970mA, there's 6.77W being dissipated. The junction->case thermal resistance is 10C/W, but given the tiny size of the contact area, and the multiple thermal interfaces (Pad->Al strip-> Heat sink), thermal resistance was probably higher (13C/W), pushing the junction temp up to at least 120C or higher. I measured the top of the package at 1970mA, and it hit over 50C. So things are getting pretty darn hot at that current level. It's pretty difficult to get great thermal transfer with such a small area - again why I think lumileds specs the rebel to 1A.

I would have to work on "shaving" off those three bottom pads. Maybe an hobby knife would work. Filing would be tough, as would using a cutting wheel on a dremel (shooting Rebels lol). If you are using epoxy on a small amount of surface and it is doing well, then epoxy over the entire bottom surface may be a little better. The ceramic substrate is a good heatsink, so having as much of the surface potted to a heatsink as possible is key, otherwise it seems like a waste. It is like applying thermal epoxy to only the right half of a Luxeon emitter slug. Now I want to really try this out.

It would be interesting to try, but hard to determine how effective it is. I don't think the ceramic is very good at transferring heat linearly, which is probably why they didn't bother to make the thermal solder pad on the bottom much larger. The ceramic is only 0.5mm thick, so heat travelling from the die to the thermal pad only goes through 0.5mm of ceramic, but heat travelling to where the power pads are travels through 3-4mm of ceramic, 6-7 times the distance, making that part of the package much less effective at moving heat than the part right below the die.

Did your power supply have fixed current levels? I was wondering why it had steps with odd current levels (odd to us high-power LED guys atleast). I knew there was a way to measure lumens with a light meter. Once I get mine (some day), I will check back with you on the exact process. Good job with your close readings compared to calibrated equipment.

Yes, I'm using an LM317 in constant current mode, with a set of currents I can additavely switch in and out. There are 5 switches , 30, 130, 310, 670, and 1260mA. I can turn on multiple switches and the sum of the two switches is the current that flows - i.e. turn on the 310 and 670 switch to get 980mA.

When you measure the lumens, do you have to read the lux at 5 degree increments for the entire 90 degrees? Does that mean that you have to have the LED running almost constantly? Hopefully the Rebel doesn't get hot during that time. In that case, do you think that the lumen output could be just a little better at higher currents if the luminous flux reading was taken at one time? Well, now I think of it, this heat could be a good thing because you get more "real-life" results by having the LED on for longer to allow the temperature to stabilize.

When I'm doing the beam profile, I usually run at a lower current, like 310mA. At that current level, heating has little to no impact on the lux readings - they settle down in a few seconds after powerup and are stable, and I do a check at 0 degrees right at the end to double-check my readings to make sure heat hasn't caused the LED output to drop. Once I have the beam profile and I calculate the lumen factor for that kind of LED, I can apply that lumen factor across all measured current levels, and I can measure multiple LEDs of the same product type simply by measuring the brightness at 0-degrees.

I only need to measure the beam profile for one sample of a kind of LED. But I make sure to measure the beam profile on any parts that could have significant changes, such as the new rebel vs. the old. If/when Lumileds releases a K2 in this same output range, I will most certianly re-measure the beam profile since changes to the die and phosphor will definitely affect the beam profile.

If the Rebel had a truly 180 degree dome, would it collaminate better with a reflector?

If the beam were even across 180 degrees, then we would get a higher percentage of the emitted light hitting the reflector and in the focused beam, but really, the beam profile of the rebel is targeted at being a Lambertian profile, which has certian characteristics.

I did some calculations once and found out that using standard reflectors, like the 50mm reflector in C and D cell Mag Lites, only 40% of the light emitted is in the focused part of the beam, and the remaining 60% is emitted out the front as spill light. The forward-emitting nature (without optics) of most power LEDs lends themselves more to using an optic rather than a reflector.

So working well with a reflector can mean a lot of things - it can mean that more of the light hits the reflector and ends up in the focused beam, it can also mean that the beam focusses down to a smaller spot, allowing more throw. Focusing to a smaller spot depends on how large the light source appears to the reflector - so smaller dies will focus better. The optics of the package also make a difference. The side-emitting luxeons throw most of their light in the 80-120 degree range, perfect for a reflector. So the vast majority of the lumens are going into the reflector and not out the front with side-emitters. However, the beam distribution is such that the side emitter appears as a larger light source with multiple nodes, so the beam can't be snapped into a tight focus like when using a high-dome (lambertian) emitter (like the rebel/k2/etc). So even though there are more lumens in the focused part of the beam, it may not be any brighter, just larger.

Optics are something that people have been trying to tame for some time now. It doesn't help that in the last couple years, the brightest LEDs are now coming from 3 different manufacturers, in 3 different packages, with 3 different kinds of beam patterns (K2, Seoul P4, Cree XR-E). It was hard enough when we had 5 years where the optics remained the same (just luxeons).

 

[LukeA]

I think for good heat transfer it is essential to solder the rebel as opposed to epoxy. Solder is roughly 5-10 times more thermally conductive than thermally conductive epoxies. It is not easy to do by hand, but the heat plate method works and if it is a small piece, you can probably hand solder if you have a good iron and can heat the whole piece.

Semiman

I think sanding a corner of the Rebel so the heatsink solder pad is visible from the side, laying it into position, and wicking a little solder beneath it would work.

I would try this but I have no more (inexpensive) 50/95 Rebels to play around with.

 

[jtr1962]

I don't think the ceramic is very good at transferring heat linearly, which is probably why they didn't bother to make the thermal solder pad on the bottom much larger.

Not necessarily true. The ceramic faces of a thermoelectric module are usually made of alumina ceramic. My guess is that the Rebel and the Cree both use this. It has a thermal conductivity of 35.3 W-m/K. Berylia ceramic can be 230 W-m/K. For comparison aluminum is 204 W-m/K. You're probably right though about them not making the pad larger because the ceramic, even with a high thermal conductivity, won't transmit heat linearly that well when it's so thin. They could have made the ceramic thicker, which would transfer heat linearly better, but then you'll have worse performance transferring heat directly under the LED die. There is a optimal thickness for which the total thermal impedance is at a minimum. My guess is that this is the thickness Lumileds used, and then based the thermal pad size on that.

 

[pokkuhlag]

Thanks for your testing the rebel, but I still have some questions about your test.

Since light meters are more sensitive to yellow-green, doesn't this affect your lumen measurements, by using different bins of leds, for example a WO bin vs VN bin?

About the thermal epoxy layer being extra thin for the Rebel, a thin epoxy layer is always better. Wouldn't this be unfair for the other leds with thicker layer of epoxy? Especially the Cree XRE since it uses a ceramic substrate as well?

The die of your rebel is off center, how would that effect the beam if you put it behind a reflector? This is would be interesting to see, since the die vs lens ratio is larger than other leds.