Cree's Binning 50+ LED's tested

[cmacclel]

Post removed.


Testing was not valid as Cree CR-E LED's cannot be comparitively measured at an angle. When the LED's where re-measured with the light meter directly over the LED the variation was only 10%.


Mac

 

[skalomax]

Thanks Mac.
The rebels are pretty impressive.

 

[Mash]

Wow, thats some variance there.
No wonder people are complaining that their upgrades sometimes are dimmer than the original!
How was the color tint of this largish sample?
+1 on the rebels! Bright and binned properly.

 

[johnny13oi]

Man that is a huge amount of emitters, wish I had that many to play around with.

 

[matrixshaman]

Thanks for the all the work and the heads up on what I suspected - although not too quite that degree. oo:

 

[Alan B]

Great test and interesting data.

Is measuring the intensity at one point an accurate indication of total flux?:thinking:

-- Alan

 

[VidPro]

:thumbsup: thanks for the data

 

[johnny13oi]

Wouldn't say placing a white box over it and then measuring the reflected spill a better indication of outputted light? These emitters could be putting out light unevenly causing the huge differences.

 

[chris_m]

None of the previous test results I've ever seen for XR-Es with proper integrating light boxes has ever shown results anywhere as far away from the specified bin as your standard deviation (let alone your outliers) would suggest you statistically should get. Applying Occam's Razor, the most likely explanation is a flaw with this test procedure.

 

[cmacclel]

Wouldn't say placing a white box over it and then measuring the reflected spill a better indication of outputted light? These emitters could be putting out light unevenly causing the huge differences.

Why would they be outputting uneven light? This was just one of those "Let me check something" idea's and is no way a scientific study

The Difference in brightness from emitter to emitter was very noticable.


Mac

 

[VidPro]

Why would they be outputting uneven light? Mac

i can think of one possible reason for uneveness, that emitter texture, and the emitter grid pattern, which shows up like crasy with a aspheric projection, or under a scan or micro view of it.
the old Luxeon things only had "bond wires" as something that gets in the way, but all this new stuff with the light releacing microhole phosphors has a pattern to the ouput. then the cree has a connection grid besides.

 

[cmacclel]

None of the previous test results I've ever seen for XR-Es with proper integrating light boxes has ever shown results anywhere as far away from the specified bin as your standard deviation (let alone your outliers) would suggest you statistically should get. Applying Occam's Razor, the most likely explanation is a flaw with this test procedure.

Do you have a link to these previous tests? I have built a good amount of Lights using the Cree XR-E LED's and have noticed huge differences in light output from emitters in the same reel. This test of mine just confirmed by findings.

I never had a problem with Lumiled's binning. You KNEW what you where buying. With Cree's binning it's a crapshoot. The Colors, Forward Voltages, and Brightness vary greatly within the same bin off the same reel.

Mac

 

[chris_m]

Just various test results on here, and elsewhere. Not seen somebody test as many as you have, but enough different test results that it would be statistically very unlikely for one of them not to have thrown up a binning error if the LEDs they were testing had the same variation your results do.

Crees are supposed to be binned for colour and brightness. The variation you're seeing suggests there's something wrong in the binning QC - maybe you've just been unlucky with your particular source.

 

[cmacclel]

Just various test results on here, and elsewhere. Not seen somebody test as many as you have, but enough different test results that it would be statistically very unlikely for one of them not to have thrown up a binning error if the LEDs they were testing had the same variation your results do.

Crees are supposed to be binned for colour and brightness. The variation you're seeing suggests there's something wrong in the binning QC - maybe you've just been unlucky with your particular source.

The LED's I tested where from the following sources

P4's Authorized Cree distributor
Q5's 2 Separate group buys here on CPF
Q2's Authorized Cree distributor
Q2 Cutter

So I doubt there is a source problem as the LED's I have on hand have come from various sources.

As for Cree's binning QC I can't believe they have a QC department for binning with the variations I have seen. It can go from 6000K tint to a 3000k tint in the same reel

Mac

 

[EntropyQ3]

The LED's I tested where from the following sources

P4's Authorized Cree distributor
Q5's 2 Separate group buys here on CPF
Q2's Authorized Cree distributor
Q2 Cutter

So I doubt there is a source problem as the LED's I have on hand have come from various sources.

As for Cree's binning QC I can't believe they have a QC department for binning with the variations I have seen. I can go from 6000K tint to a 3000k tint in the same reel

Mac

First off, thanks for making your data available.
It is incredibly useful for people who would otherwise only deal with small numbers or even individual samples, to get an idea of the normal spread of a given product. Helps immensely with troubleshooting.

As others have said, your data would be easier to interpret if you had made some kind of "total light output" estimation, never mind that it wouldn't be up to the official standards. Lots of folks get reasonably repeatable results from pretty simplistic setups. As long as you don't have some such device however, you will find that people would rather believe that your methodology is flawed, than to accept that your data reflect luminous output. Ungrateful *******s. You say that the differences in output is obvious to the eye as well as the lux meter. That implies that variations in output patterns is not the sole explanation (and it would be a problem in itself if it were).

Looking at your data, I realize that I may have been far too trusting in accurate binning, and too inclined to look for problems in places where, perhaps, there were none. I admit the spread looks extreme, far beyond what even rather lax deviation standards should allow.

If you could make some integration sphere hack, it would be great.

 

[cmacclel]

I'm not going to take this testing any further. Is my testing flawed? Thats is quite possible *if* the light emitting from the XR-E lens is uneven as some people have mentioned but then that means there would be huge variations in the Manufacturing process which I find hard to believe. Each LED was placed on the heatsink in the same exact location and orientation.

Also look at the 3 rebels I tested with the same test setup. They are within 2% of each other.

I have 40 of the 0100 Rebels at home maybe I'll test a few more.

Mac

 

[chris_m]

As long as you don't have some such device however, you will find that people would rather believe that your methodology is flawed, than to accept that your data reflect luminous output. Ungrateful *******s.

I'm struggling to see where his flaw is - it's just that as I said before Occam's Razor suggests that it's more likely there is one than that everybody else is wrong. If I hadn't seen several other consistent testing results I'd be more prepared to believe that Cree are dishonest and don't actually bother binning their LEDs.

I don't believe the flaw (assuming there is one) is to do with variation in light output across the beam - I'd expect the beam pattern to be more consistent than the light output. About the best I can come up with is a fault in your power supply, given correlation between your light meter and Mk 1 eyeball.

 

[cmacclel]

I had a calibarated Fluke True RMS DVM measuring current at 347ma I had my variable supply set to 4.2v with the current limiting set to 350ma. When I first saw the huge variances between LED's I also questioned my setup.

Mac

 

[AvPD]

What's the process when the LEDS are manufactured? Are they similar to computer chips which have "yield" rates? Do some have less defects than others and the "bin" categories represent the ones that work from best to worst? Or is the machinery retooled every few months and they all come out the same?

 

[EntropyQ3]

What's the process when the LEDS are manufactured? Are they similar to computer chips which have "yield" rates? Do some have less defects than others and the "bin" categories represent the ones that work from best to worst? Or is the machinery retooled every few months and they all come out the same?

I'm not sure anyone not working on Cree knows the details. The LEDs are probably similar to computer chips in so far as them being produced in batches, but there still being some individual variability.

If we assume that the data cmacclel has contributed is proportional to luminous output, then that could imply for example that Cree made batches of a few thousand, tested a sampling of these, and labeled that particular batch according to the outcome of their samples. It could also be that all batches where their samples failed to comply with their minimum criteria for a given upper batch are labeled "P4" or whatever the lowest batch is called at the time, leading to extreme variability in output in samples from the lowest batch.

I plugged his data into Excel, and:
30 P4 samples yielded
Average output:341
Standard deviation:58

12 Q5 samples yielded
Average output:423
Standard deviation:72

Note that comparing averages, his Q5 samples averaged 423/341=1.24 or 24% higher output than the P4s, whereas Cree bin data would suggest 110/84=1.31 or 31% higher output for the Q5. This is pretty close, particularly considering that the Q5 bin is the top drawer, and would thus tend towards the lower part of the bin, and the P4 is the current bottom drawer.

So his average results are, bearing in mind the smallish sample on the Q5s, pretty much in line with what we would expect, it's just the spread that is shocking. Indicating perhaps, that the bins actually just indicate where the average luminous output of that particular batch ended up, rather than having individually binned chips.

This is what the data indicate, and lacking better information, I'm inclined to believe this is the case.

Of course, it makes the value of paying for a higher bin that much more of a lottery.