LED life... how did they come up with 100,000 ?

[Handlobraesing]

The big three (GE, Osram Sylvania and Philips), as far as I know conducts a real lab test using statistically significant number of lamps to calculate life time on incandescent, HIDs and fluorescent to get the average lifetime, which is defined as the hours at which 1/2 of the large sample of lamps fail.


Most of LEDs claim the standard lifetime of "100,000 hours(11.42 years) " but we all know the newer LEDs haven't even been around for 10 years, so how do they know?

oops wrong section.. mods, could you move it to the LED forum?

 

[nerdgineer]

I think the early red LEDs had very long lifetimes which may have approached 100,000 hours running at spec. White LEDs don't last nearly as long. I read some test (can't find it now, but it's referenced in CPF every year or so) for 5 mm white LEDs which showed degradation to 50% output in about 6000 hours when running at 20 mA spec with reasonable heatsinking.

Many white flashlight LEDs, running at higher than spec and without good heatsinking, will go to 50% in a few hundreds or even tens of hours, manufacturer's claims notwithstanding. Still, your eye won't see the first 25% or 30% degradation hardly at all, and a few hundred hours is a LOT of use for most flashlights - excepting a few crazy CPF members; so a few hundred hours isn't that bad.

In answer to your other question, testers can project expected lifetimes of parts by testing big batches olf them and seeing when a much smaller percentage degrade to 50% or fail or whatever. Then they have historical curves they can use to estimate what the expected lifetime of the group average would be. Test engineers will go on and on (and on...) about this.

EDIT: Apologies to any test engineers out there. You all are right of course, and test consists of a lot more than statistical analysis. And yes, you guys catch all the cr** left over by lazy system engineers and design engineers and HW/SW engineers and the rest...

 

[MrAl]

Hi there,

What they do with transistors for one test is subject them to higher than
normal operating conditions and then count the number that fail.
The number that fail when they are operating at conditions that are beyond
the data sheet are then used to determine the failure rate of transistors
that are operated normally. I dont remember the formula but it's probably
on the web somewhere.

For testing an LED you could probably subject a bunch to high current and measure
the mean failure hours and extrapolate down to LEDs that run at normal current.
You would have to get the extrapolation formula from the LED manufacturers.

 

[EngrPaul]

http://en.wikipedia.org/wiki/MTBF

 

[sysadmn]

Dang! EngrPaul beat me to it. Which is kind of funny, since I are also a engineer named Paul.


Here's LumiLeds report: http://www.lumileds.com/pdfs/RD01.PDF

You do not typically test at greater than spec. You test at the max conditions you recommend, using long test periods and lots of devices. Then you work backwards and say "If the MTBF was X, we would have seen Y failures. Since we saw fewer, MTBF is greater than X". If you're really interested, you could read Military Standard 690, or hire my wife, the statisician.

 

[VidPro]

we have been running 1W luxes non-stop since they came out, they might indeed last the claimed 100,000, but i think they reached 50% long ago.
this light has in it, old 1w, newer 1Ws , and newest 3Ws all being driven about the same, and the visual appearance of the old ones is worth replacing.

another light has 3W ones in it, run slower most of the time, has run since the 3Ws were available, and its still looking ok, time will tell.

Failure and usefull light output are 2 different worlds.
that chart where they tested this stuff, shows they should NOT be saying 100,000 in any advertising, especially for 5mm leds. to me its a lie to indicate that the light has X ammount of output AND will last X ammount of time, it will not do both.

 

[Handlobraesing]

I don't know this too well, but I'm seeing a limitation of extrapolation.

Let's say you observe the death rate amont 10,000 newborns. Do you think that would give a reasonable value of average life expectancy of a human being? I think not.

 

[2xTrinity]

Many white flashlight LEDs, running at higher than spec and without good heatsinking, will go to 50% in a few hundreds or even tens of hours, manufacturer's claims notwithstanding. Still, your eye won't see the first 25% or 30% degradation hardly at all, and a few hundred hours is a LOT of use for most flashlights - excepting a few crazy CPF members; so a few hundred hours isn't that bad.

I can certainly detect a 25% difference in light output. Especially if I'm trying to either light up a big area, or throw light far away. Changes in brightness affect throw more than they affect how things look up close.

 

[VidPro]

here is a few charts
http://www.candlepowerforums.com/vb/showthread.php?t=126857

so what is the MTBF of a human

 

[SemiMan]

we have been running 1W luxes non-stop since they came out, they might indeed last the claimed 100,000, but i think they reached 50% long ago.
this light has in it, old 1w, newer 1Ws , and newest 3Ws all being driven about the same, and the visual appearance of the old ones is worth replacing.

another light has 3W ones in it, run slower most of the time, has run since the 3Ws were available, and its still looking ok, time will tell.

Failure and usefull light output are 2 different worlds.
that chart where they tested this stuff, shows they should NOT be saying 100,000 in any advertising, especially for 5mm leds. to me its a lie to indicate that the light has X ammount of output AND will last X ammount of time, it will not do both.

You can not compare old Lux-1s to new 1s and threes. When the Lux-1 first came out, it was not unusual to get <20 lumens parts. Now, it seems rare to get them under 40. The first ones also has a different phosphor which did not make them look as bright as a spot source.

 

[VidPro]

You can not compare old Lux-1s to new 1s and threes. When the Lux-1 first came out, it was not unusual to get <20 lumens parts. Now, it seems rare to get them under 40. The first ones also has a different phosphor which did not make them look as bright as a spot source.

when i put the newer ones in they didnt look that far off from eachother, if they had looked THAT far off, i would have chucked them when i put the newer ones in.

 

[thehappyman]

They came up with 100,000 hours after a ratings tester had his light's LED fail after 11.4 Years !!!!!!!!

 

[sysadmn]

I don't know this too well, but I'm seeing a limitation of extrapolation.

Let's say you observe the death rate amont 10,000 newborns. Do you think that would give a reasonable value of average life expectancy of a human being? I think not.

No, but based on past experience, you say, "we think the mortality curve has this distribution, with this mean, and these parameters". Then you use observations to confirm or deny that hypothesis.

 

[Luminescent]

No, but based on past experience, you say, "we think the mortality curve has this distribution, with this mean, and these parameters". Then you use observations to confirm or deny that hypothesis.

This is essentially correct, but we are confusing MTBF which involves outright failures with "Lumen Maintenance" which involves the gradual loss of output.

Both are tested by subjecting a large lot of samples to what is called "accelerated life testing", where we run the samples under conditions calculated to speed up both the degradation [Lumen Maintenance] and failure mechanism [MTBF]. MTBF can be something like 1 outright device failure per 100,000 to 500,000 hours of operation at full power, and Lumen Maintenance on the order of 70% [30% loss] after 50,000 hours at full output. As a few folks have already noted, these numbers are HUGELY influenced by drive levels and temperature.

Understanding HOW the degradation and failure curves are influenced by drive and temperature is what make this "accelerated life testing" work.

Earlier comments that this is done by only running at MAX spec are simply incorrect, "accelerated life" type tests always involve running the device outside it's normally specified range in a way calculated to accelerate the failures in a statistically useful way that will give meaningful data about the long term performance of the part.

In order for this to work, we must first take the time to develop a model (based on both theoretical observations AND real world measurements) about how raising the drive and temperature should effect the "MTBF" and "Lumen Maintenance" of the LED.

10 years ago the models had to be mostly on the theoretical side, but as large numbers of the devices built back then have been subject to real world conditions and are reaching 'end of life' we can fill in the blanks, and greatly improve the accuracy of the model.

Here is the current Phillips Lumiled model for the K2's Lumen Maintenance from their web page:

This chart plots the lower bounds for both the B50 [50% outright failure] and L70 [70 % Lumen Maintenance] for the K2 and Rebels. Notice that to keep the Lifetime resonable at higher currents, the junction temperature must be kept lower. Unfortunatly this is the opposite of what happens in real life situations, where junction temperature goes UP at higher current levels. This means that to safely run higher current, without greatly reducing the life, you need a MUCH larger heatsink (perhaps four times larger to double the current). This image should be cause for concern for those who plan to run the little Rebels at 1 to 1.5 amps of current (but then again, some still like incandescent lights that use bulbs with ratings in the TENS of hours).

For more accurate info on the new Rebel, you can grab the full PDF doc for the Rebel's reliability data:

http://www.luxeon.com/pdfs/RD07.pdf

From what I have seen MTBF data (outright failures) have been, if anything, on the conservative side right from the beginning for the high power Luxeons, but for the early "Lumen Maintenance" projections they were 'shooting in the dark' a little bit, by basing them on mostly theoretical models, but now, with more than 10 years of data and experience with high output white LED's, the industry is getting better at projecting long term performance.

 

[sysadmn]

This is essentially correct, but we are confusing MTBF which involves outright failures with "Lumen Maintenance" which involves the gradual loss of output.
...


Earlier comments that this is done by only running at MAX spec are simply incorrect, "accelerated life" type tests always involve running the device outside it's normally specified range in a way calculated to accelerate the failures in a statistically useful way that will give meaningful data about the long term performance of the part.

Thanks for an excellent follow-up!

However, as the person who wrote the statements you took issue with, I'd like to split hairs, and defend my reputation for pedantry.

You're right about the difference between MTBF and lumens maintenance, but the initial question seemed to be about MTBF. LM is often glossed over, since the MTBF figures given by manufacturers define a failure to include a specified level of lumen loss. Certainly Handlobraesing's comments about newborns concerned MTBF - my newborns emitted an astonishing variety of surprises, but sadly, photons were not one of them

The first paragraph of the report I quoted states "Lumileds tests parts at the absolute maximum rated conditions recommended for the device". I was responding to MrAl's observation that for transistors, "one test is subject them to higher than normal operating conditions and then count the number that fail". In the specific case I quoted, the data Lumileds reported was not obtained out of spec.

I agree the accelerated life testing you describe is done to define the life curves. Some, maybe most, manufacturers may also do ongoing acccelerated life testing to determine if the process is in control, and the curves are still valid.

PS - The same report defines failure as no light at all and max level % degradation >70%.

 

[Luminescent]

Thanks for an excellent follow-up!
PS - The same report defines failure as no light at all and max level % degradation >70%.

If you check the RD07.pdf document for the Rebel that I posted the link for, you will see that sometimes the distinction between the low (10%) and high (50%) failure rates (B10%, B50%) at a specific Lumen Maintenance(L70%) is quite useful in determining how confident we can be in the device continuing to meet specifications at various drive levels and temperatures. To provide this information there are multiple plots available, some at B50/L70 (50% failures @ 70% Lumen Maintenance), and some at B10/L70 (10% failures @ 70 Lumen Maintenance).

Looking at the difference between these plots can tell us a great deal about how the LED will perform if overdriven (which I know is near and dear to the heart of some here on the CPF).

For the InGaN type Rebel [blue/white], there is little difference between the [B10/L70] and the [B50/L70] plots indicating that a relatively small shift in junction temperature made the rate at which the devices failed to meet the 70% lumen spec, shift from 10% to 50%. For example at 1 Amp of drive to get 50,000 hours of projected life with 50% confidence we can run at about 134C junction temperature, but dropping the temperature by only 10C to 124C increases the confidence to 90% that we can get to 50,000 hours with 70% output. This says that the blue/white rebels are very forgiving up to a point (124C is quite a high junction temperature), but beyond that point the failure rate increases dramatially.

For the AlInGaP types [red/amber etc], it's a different story. The ultimate drive level is lower (they don't even show a curve for 1000mA), and the temperature curves are also lower, but there is a much larger temperature spread between the 10% mortality and the 50% points. The 700mA curve shifts quite dramatically in temperature between the B10 and B50 plots (10% failures, and 50% failures), which shows that we have to lower the junction temperature quite dramatically to operate these emitters at 700mA if we want a high confidence factor that they will still meet the 70% spec with a reasonable statistical probability of reaching their projected life expectancy, but if you push things a little junction temperature wise, the failure rate only increases by a little (so there is less of a 'brick wall' effect at a specific temperature).

I didn't mean to nit-pick the earlier discussions, but when I figured out why they bothered to do both types of plots, I realized that this was good information to have, so I thought I would pass it along.

Many thanks to evan9162 for comments and corrections, the errors that remain, if any, are my own.

 

[evan9162]

You're a bit mistaken in your explanation.

First, your definition of B50, L70 and B10, L70 are off. B50 L70 means that 50% have failed catastrophically OR lost 30% output due to lumen degridation. The 50% failures are a combination of lumen degridation and non-lumen degridation failures. I would suspect that the overwhelming majority of failures are due to lumen loss, and only a very few failures would be due to some other catastrphic failure.

For InGaN, the B10,L70 and B50,L70 plots are different, but not by much. To reduce degridation from 50% to 10% at 60,000 hours only requires a reduction in junction temperature by 10C. This is because InGaN LEDs are much more robust at high temperatures than other LED materials.

Second, the AlInGaP are not older Rebels, they are for Red, Red/Orange, and Amber Rebels. RD07 was released months before the 100 lumen rebels were. Thus, the 100 lumen and 50 lumen rebels will have similar lifetime characteristics, since they are all based on InGaN.

AlInGaP (red/orange/amber) LEDs have different lifetime and performance charachteristics, and thus are called out seperately in reliability documentation.

The reason they did both plots was for the different LED materials, not for old vs. new.

 

[Luminescent]

You're a bit mistaken in your explanation.

First, your definition of B50, L70 and B10, L70 are off. B50 L70 means that 50% have failed catastrophically OR lost 30% output due to lumen degridation. The 50% failures are a combination of lumen degridation and non-lumen degridation failures. I would suspect that the overwhelming majority of failures are due to lumen loss, and only a very few failures would be due to some other catastrphic failure.

Huh??? I thought that's what I said, I guess you are right it was confusing.

Perhaps a better way to say it would be that the two plots give us plots at different confidence factors. For the [B50, L70] plot, 50% of the devices can either fail outright or degrade to 70% output, but for the [B10, L70] plot we are looking at the limits where we can be much more confident that the devices will continue to meet their specifications, because only 1 in 10 (10%) will either fail outright or drop below 70%.

For InGaN, the B10,L70 and B50,L70 plots are different, but not by much. To reduce degridation from 50% to 10% at 60,000 hours only requires a reduction in junction temperature by 10C. This is because InGaN LEDs are much more robust at high temperatures than other LED materials.

I agree completely with your comment about InGaN being more robust at high temperatures, but if we are being hyper-technical I would say about the [B10/L70] vs. [B50/L70] plots, that they are not plotting reduced 'degridation'[SIC] they are plotting reduced mortality (with mortality defined as a failure to continue to meet the L70 criteria for any reason, including outright failure).

Could we agree that the differences in the two curves represents both an increase in outright device failures, and greater variability in lumen loss at higher temperatures?

Second, the AlInGaP are not older Rebels, they are for Red, Red/Orange, and Amber Rebels. RD07 was released months before the 100 lumen rebels were. Thus, the 100 lumen and 50 lumen rebels will have similar lifetime characteristics, since they are all based on InGaN.

AlInGaP (red/orange/amber) LEDs have different lifetime and performance characteristics, and thus are called out separately in reliability documentation.

The reason they did both plots was for the different LED materials, not for old vs. new.

Thanks for pointing this out.

I stand corrected. I was in a hurry to make a point and was looking at the wrong table.

I'll remove the reference to new and old Rebel 100/50 from my comments to avoid misleading anyone.

I just wanted to make the point that looking at these curves will help us understand how well we can expect these devices to work at high drive levels. I will stand by my comment that we can see from the larger shift in the 700mA curves, that the AlInGaP devices do NOT like high drive levels. We have to shift the temperature curves way over to get to the 90% confidence level (B10) at 700mA, which would correspond to either living with an unrealistically large heatsink, reduced reliability, or both.

 

[evan9162]

Totally agreed on the AlInGaP devices. Several years ago, we were noticing the severely reduced output at high temperatures that these devices suffered from, in stark contrast to the InGaN devices that hold up quite well to high temperatures.

In fact, these devices suffer from heat induced output loss so severly, that even at modest drive currents, output levels off and the increased heat from higher input power more than offsets the extra output from increased drive current.

We found that in some cases (poorer heatsinking), a red Luxeon I (roughly equivalent in performance to a red rebel) would actually level off at about 500-600mA, a drive level that a white luxeon I would shrug off and keep chugging along at.

It seems that the loss of output due to temperature goes hand-in-hand with decreased lifespan at increased temperatures.

Several years ago, I used a thermoelectric cooler (TEC) to supercool a red-orange luxeon I. By dropping the junction temperature by some 60 degrees celcius, I was able to increase output by at least 50%. It was fun to explore the output vs. junction temperature curve in those parts.

 

[VidPro]

so there is a simple solution, Socketing. Every other light in the world is socketed. mabey its about time , before they SOLDER them into home lighting fixtures , to have a copper heat sink socketing system.

patent anyone

have the base of the led, and mabey one of the connecting poles, come down into a copper connector on the side. and the socket has a copper connecting reciever, both thick enough to transfer the heat, or say liquid filled heat tubes.
then the sockets have the copper spreaders, and aluminum sincs.
and just incase i forget, they would be Polar sockets, so the avearge consumer, can only stuff them in one direction.

@ <-- le bulb
|]
_____[||]_____ <--- el Socketo

grrr anything so i dont have to solder on them hair wide connections again.

at LEAST they should be socketing them weasily 5mm bulbs, a array made of them is still not cheap, and desoldering 100, aint easy.