Future development of the maximum luminance of LEDs

That would be very interesting, but I don’t think it makes much sense here. These are automotive LEDs. The requirements for such components are very strict and changing specifications would be bad business unless denoted with the product code.

Well yes, but this only pertains to homemade devices.
There are already car headlights and projectors using this technology which of course have all the needed safety features.

Yuji phosphors are nice, but not very efficient at high power density (the more even the spectrum i.e. higher cri, the more inefficient a light source is).

[quote=EasyB]

[quote=The_Driver]

[quote=EasyB]

All die sizes are measured by either sma or koef3. They need to do this rather precisely because otherwise the luminance values would not be of much use. They do this by taking a macro shot of the LEDs at very low power from head on (as straight as possible) and then counting the pixels and using the package size as a reference length.

[quote=The_Driver]

[quote=EasyB]

[quote=The_Driver]

The dies very well could have areas different by 10, but I wouldn’t be surprised if the uncertainty in the area measurement is greater than 10. For example how do you deal with the brightness gradient near the edge of the die? The point I’m trying to make is that for some LEDs this method of measuring luminance doesn’t really work. In-flashlight measurements do tell us the actual die luminance, but they introduce other sources of uncertainty like reflector quality and focus.

No, sma measured the luminance like he always does, from head on. Using an aspheric lens should allow for the utilization of these high values, because it uses the light from head-on. Of course we don’t know by how much the luminance is reduced at what angle. It would be great if someone were to measure this.

I think the XP-G3, XP-L2 and XHP70.2 are special cases that we need to ignore in this discussion. It’s a step in the wrong direction in this regard (improving lumen density).

So you think that the quality of the InGaN active materials is a major factor? Interesting. Did you ever find any info regarding this?

This is a good point! It concerns all the measurements we discuss here regularily and especially so in this thread. Most people here (except maybe djozz and sma) never talk about what level of error is possible with their measurements and how it effects their analysis of those values.

In-flashlight measurements are of course a great way to check the validitiy of the luminance values. My experience has been that for the older, “classic style” Cree LEDs the luminance measurements are spot on when all the details are considered.

A 10% difference would be easily discernable though on a macro photo, even without counting any pixels.

I think you are misunderstanding. The luminance measured with this method is simply incorrect. With this method some of the measured light is coming from the area to the side of the die and not from the die itself, but the measured intensity is divided by just the die area, thereby inflating the calculated luminance. The actual die luminance is lower. It is not a matter of angular dependence.

Agreed.

Forgive me for not finding references right now. From what I have read the internal quantum efficiency of LEDs can be quite high at low currents. But so-called efficiency droop happens at higher currents, so there is room for improvement here I think. I don’t think the cause of the droop is definitively known, but one possible cause is the inherent electrical polarization present in the common GaN crystal structure (wurtzite). So one direction research is taking is trying to grow the InGaN with a different structure (zincblende) without the inherent polarization.

Anyways, I don’t know the current state of knowledge in this area, but that is one possible way to improve performance at high current densities where we need it most.

How is light coming from the side of the die going to hit the sensor of the lux meter during a correctly done luminance measurement?
Correctly done means that only direct light coming straight from the front of the die will actually hit the sensor. Thus dividing the measured Candela by the the visible die size will result in the actual luminance that the lux meter is “seeing”.

Here in the first post you can see a picture of sma’s test setup for measuring the luminance. He has implemented measures to prevent any stray light from hitting the sensor and thus inflating the measured values.

Using an aspheric with a high f-number should make it possible to make use of this high “front luminance”.

The light is coming from the area to the side of the die, so it is exposed to the luxmeter. I’m not talking about light going to the side.

Imagine looking down on the sliced XPG3 from above. The phosphor covering the package to the side of the die is lit and gives off light which the luxmeter detects. I showed with my measurements that the light from the side phosphor is significant in inflating the luminance calculated using this method.

Ok, I get you now. Partly the reason or this is the excess phosphor which is around the die, but not covering it. For a better measurement this should either be removed or one could use a pin-hole type cover with the exact size of the die (or 1mm^2 circle).

EDIT: I just noticed that you do state the current which you measured at, 2.8A. From this and other test results I will calculate the real maximum luminance.

EDIT2: ~157cd/mm^2 at 8.5A and 85°C.

EDIT3: maybe instead of back paint you could try something reflective on the sides of the shaved XP-G3, basically turning it into a “laser”.

BTW: I just calculated the real thermal resistance of the Luminus and Lumileds LEDs where the datasheets don’t have these values. I just assumed an efficiency of 20%. Based on this I also fixed the values in the column “min. temp at max. luminance and 25°C”. The LEDs are actually reaching their maximum at about 90-107°C.

Perhaps it turns out that a luminance value that can be defined as the total light flux divided by the light emitting surface area does not 1 to 1 predict the throw because it does not take the spatial distribution into account, and even then spectral shifts with changing angle can have an influence. In theory, an ideal spatial distribution for throw may be different for reflector lights and for lights with aspheric lenses.

If the aim is getting a value for any particular led that directly gives potential throw, maybe an empirical approach for measuring it must be devised. I have an abandoned project for testing reflectors (as always I found designing the method way more fun than doing series of measurements using the method) that when used the other way around, with a fixed reflector and a fixed aspheric lens (say, a C8 reflector and Brinyte B158 lens) and varying led, can be made into a standard method for measuring throw. Spot intensities can be measured with all leds at a standard current, and the current/output graph from the output test can then be used to calculate the maximum throw.

will heating ever go away?
I^2 * R losses?
it would be nice to design without having to consider heating and all its problems…

wle

Well it’s not based on lumens. It’s based on Candela. You only need a lux meter to measure the luminace, not a sphere. But still, you make a valid point regarding the spectral shifts although experience has shown that de-domed emitters have less spectral shift at high angles.

For lambertian emitters (perfectly flat) there is a somewhat accurate method though using lumens. Simply divide the lumens by the surface area times pi. It gives you a ballpark figure.

The used approach works for all LEDs except these new, “weird” ones from Cree (although the Luxeon V might also fit into this category). It’s based on physics. The luminance is a constant at all angles for a lambertian emitter. Otherwise it would never work with reflectors…

I ask me if it is about the influence of the yellow lateral light areas, which I have noted in the sketch and their relevance in practice in a Thrower.
Does this light amplify the intensity of the hotspot in a Thrower?

What about measuring the illuminance to calculate the luminance?
Is the light from the side surfaces still being detected by the luxmeter?

A note to everyone:

You can now click on the luminance values. It will lead you to the sources of where I got them (sometimes requiring additional calculation for the maximum value). Some of the linked tests are in German.

Also, there is a new king of luminance, just recently tested by Köf3, the Osram SYNIOS P2720 KW DMLQ31.SG with a 0.5mm2 die. It does up to 316cd/mm2.

Taking some data from here:

we can calculate thermal resistance of the solder layer under the thermal pad.
This assumes quite good thermally SNC solder, for many builds results will be worse.

75 μm / 58 (W/m°C) / x mm² ~= 1.293 °C / Wmm².
So thermal resistance of DMLN31.SG from MCPCB to junction is 20.5 °C/W and XHP 35 HI - 2.14 °C/W. At peak power this leads to the increase of temperature of 1.6 °C and 15.4 °C respectively. I guess it’s significant enough to include in the emitter temp calculations.

Below the solder level, what is the effect of thermal pad size? There must be some temperature gradient….I don’t know really.

And I can’t help but think about djozz tests that showed MOSX MCPCB as the performance-limiter of Blackie. Doing calculations like I do here, it looks that at peak power MOSX gives 31 °C temperature for this LED. It may be worse, led4power just specified thermal resistance to being equivalent of 250μm of solder without telling which one.

I guess that with some LEDs it may be worthwhile to think again about picking a better solder paste or go with thermal glues…Compared with SNC, SnAg would give a 4°C improvement for XHP35 HI. SnPb is another 2.4 °C worse. Though the difference is less at sweet spot.

ADDED:
Note that the following assume no voids in solder. This assumption is totally unrealistic. I am unable to tell how much difference does it make though.

BTW Conductonaut on the thermal pad and solder on electrical ones may be interesting too.

Thanks a lot for you help!
I will revise the temperature column.
EDIT: done!

In the meantime: pure indum is the best TIM that I found so far. 86 W/m°C. Compared with SnAg it drops the temperature by another 1 °C. So 7.4 °C better than SnPb at peak output.
But as far as I can tell it is not available as a paste, only wire (it seems to be the case with SnAg too). How about this:

  1. Put MCPCB on the hotplate
  2. Put paste on electrical pads
  3. Cut a tiny bit of indum wire and put in on the thermal pad
  4. Place the LED on the hotplate next to the MCPCB
  5. Heat up until the solder melts
  6. Place the LED on the MCPCB. It should self-center
  7. Tap it
  8. Let it cool

This way should ensure:

  • no cold joints
  • not overheating the LED
  • not heating / cooling too fast

German Modder Vinz uses self-made diamond based thermal glue. It has 85% synthetic diamond content. The diamond has a thermal conductivity of 3300 W/(m·K). It theoretically has a better thermal conductivity than copper. He doesn’t use it for the central thermal pad though. He uses normal DTP copper pcbs and a solder with a high silver content, Mundorf MSolder Silver Gold. This solder is very expensive unfortunately (50€ for a 100g roll). MSolder Silver Gold Supreme has an even higher Silver content (9.6g instead of 3.8g), but it’s twice as expensive.

I would prefer we talk about more fundamental changes. Using better solder will not increase the luminance by visible amounts. How do we get from 300cd/mm2 to 500 or even 1000?

MSolder Silver Gold Supreme also has much higher melting temp.
ADDED: pure indium solder wire is even more expensive.
Indium scraps are more expensive too, but available in small packages:
https://www.aliexpress.com/item/10-grams-High-Purity-99-995-indium-in-Metal-Lumps-Vacuum-packing/32739500012.html?spm=2114.search0104.3.2.7c21aaafymebW0&ws_ab_test=searchweb0_0,searchweb201602_1_10152_10151_10065_10344_10068_10342_10343_10059_10340_10314_10341_10534_100031_10084_10604_10083_10103_10304_10307_10615_10301_10142,searchweb201603_25,ppcSwitch_2&algo_expid=c716d6d9-e58e-4ac0-8471-c9cddbe5f119-0&algo_pvid=c716d6d9-e58e-4ac0-8471-c9cddbe5f119&transAbTest=ae803_4&priceBeautifyAB=4
100g Indium blocks are cheaper than than MSolder wire:
https://www.aliexpress.com/item/100-grams-High-Purity-99-995-indium-in-Metal-Lumps-Glass-bottle-packing/32792125760.html?spm=2114.search0104.3.92.250dca01dBpe1y&ws_ab_test=searchweb0_0,searchweb201602_1_10152_10151_10065_10344_10068_10342_10343_10059_10340_10314_10341_10534_100031_10084_10604_10083_10103_10304_10307_10615_10301_10142,searchweb201603_25,ppcSwitch_2&algo_expid=1d849d3e-dd8b-4476-b01d-c53bb4b03806-14&algo_pvid=1d849d3e-dd8b-4476-b01d-c53bb4b03806&transAbTest=ae803_4&priceBeautifyAB=4
but who needs 100g?

I would expect Indium-Silver to have better conductivity than pure indium, but according to Indium Corp that’s not the case:

Too bad MSolder doesn’t list conductivity of their solder….though Indium Corp shows a similar 96.5Sn3.5Ag to have 33 W/mK. And this one at 78 W/nK :person_facepalming:

It says 60W/mK for Msolder SilverGold in your link…
Which is not a bad value, but SnPb 63/37 is almost as good (50W/mK ) and doesn’t cost much.