Safe current for XML2 on Aluminum MCPCB

thisnameisvalid · 2019-08-20T15:22:02.000Z

clemence:

Get yourself a decent lux meter and 5 digit volt meter. And see for yourself. Initial drop from “virgin” output is the fastest.
I discussed this with Djozz few months ago. And he got the same conclusion too. After few hours resting, the output back to normal but requires more current. All damaged LEDs needs higher input current to get the same output/voltage/temp reading.
In your case with FET driver, as the LED voltage gets lower, the current input gets higher. This is why you don’t see much reduction in output, the driver compensates by increasing the current.

[Clemence]

Do you still have any of the test data for this? I’d be interested in looking through it. I’m not surprised that there is a loss in brightness, but the fact it’s rapid does surprise me slightly. The sinkpad marketing videos show the die being much cooler with a DTP board when viewed through an IR camera, and one might expect that to carry on when the LED is overdriven. That’s just an assumption though, I’ve never seen it tested. Next time I get a new LED for a FET light I might test the output before / after the first turbo run

clemence · 2019-08-20T16:05:33.000Z

I never really write it down because crash test any LED in my opinion, is a one way destructive test. The first time I noticed this was when I skipped a current setting (yes I do all the test manually). For example, let’s say I plan to test an LED from 100mA all the way to 5A with 100mA resolution. Then at 3,3A I forgot if it was 3,3A or 3,4A. So I wait several minutes and rerun the test from 3A, in all cases, whenever I passed too far outside max rated current, the second reading always reads lower.

I don’t think you will get accurate reading with FET based driver. As I mentioned above, damaged LED (by over current) has almost always suffer reduced forward voltage. Lower forward voltage means higher current in FET driver. It’s self balancing.
My measurement done with constant current DC power supply in a controlled total loss cooling system.

[Clemence]

WalkIntoTheLight · 2019-08-20T17:21:53.000Z

That is very surprising. With all the FET-driven lights that BLF users love, there should be more complaints about damaged LEDs. Whether it’s higher current consumption, or lower output, this board should be full of complaints about all the high-output lights people buy.

Even if the output is compensated with reduced forward voltage, that can only give a little more time before the problem is noticeable. It’s not like you can shove 10 amps through one of these LEDs.

You’re talking about damage in seconds. Everyone should notice that.

Okay, you’ve got me interested enough that I’m going to dig out my test results on some of the FET lights I reviewed when new, and re-test them now that I’ve used them for a year to two.

clemence · 2019-08-20T17:42:30.000Z

My guesstimate is at 4A your L95 = ~2000 hours, and at 5A L95 = ~500 hours. Output depreciation is not an on/off process, everything starts as soon as there’s current flowing.

[Clemence]

WalkIntoTheLight · 2019-08-20T17:45:01.000Z

Okay, I did a quick test of my most-used FET light, the Astrolux C8. I’ve had it for about 1.5 years, and put on about 15-20 hours of “full output” usage on it (just a rough estimate). It does have a timed step-down, but I always bump it back up every time it steps down. So, it does get hot, though not too hot to hold.

The output is exactly what I tested on it when it was new, a little under 1300 lumens. No current measurements, though.

Okay, I know you say you can’t tell with FET lights, but if the damage is so quick, it’s got to show up as decreased lumen levels eventually. Unless the damage stops after a certain point, and lumen and current levels stabilize.

Is the damage you’re seeing, possibly due to improper heat-sinking?

WalkIntoTheLight · 2019-08-20T17:51:21.000Z

clemence:

My guesstimate is at 4A your L95 = ~2000 hours, and at 5A L95 = ~500 hours. Output depreciation is not an on/off process, everything starts as soon as there’s current flowing.

[Clemence]

Okay, from that chart, it looks like the damaging factor is heat at the junction point, not the amount of current flowing. Current plays some role, but it appears if you properly heat-sink things, most of damage is mitigated.

And even worst-case, it doesn’t seem the L70 lifetimes are affected much at all; just the L95 lifetimes.

I’m not sure if we can extrapolate accurately past Cree’s 3 amp testing, but if you’re right that 5A gives 500 hours until the LED is only 95% as bright, then that’s why I’m not seeing any drop at all after 15 or 20 hours.

I’m okay if I get 500 hours at almost “new” output levels. That would explain why people aren’t complaining.

clemence · 2019-08-20T17:53:50.000Z

WalkIntoTheLight:

Okay, I did a quick test of my most-used FET light, the Astrolux C8. I’ve had it for about 1.5 years, and put on about 15-20 hours of “full output” usage on it (just a rough estimate). It does have a timed step-down, but I always bump it back up every time it steps down. So, it does get hot, though not too hot to hold.

The output is exactly what I tested on it when it was new, a little under 1300 lumens. No current measurements, though.

Okay, I know you say you can’t tell with FET lights, but if the damage is so quick, it’s got to show up as decreased lumen levels eventually. Unless the damage stops after a certain point, and lumen and current levels stabilize.

Is the damage you’re seeing, possibly due to improper heat-sinking?

Please check if you think my setup wasn’t cool the LED enough:

Nichia NF2W757GR-V3 [GRV3] Flashlight Modding and DIY Parts

Nichia sent me hundreds of several different LED types for my past city garden light project prototyping. I found some very interesting “BLF-worthy” LEDs. One of them is currently the only medium power Nichia LED which is OEM rated up to 2,4 watt/LED. For 3030 medium power LED, it’s like an XHP50 to XML. It’s Nichia’s most powerful medium power LED. There are two variants available: NF2W757G-V3 (3V) and NF2W757GR-V3 (6V) I’m only interested in NF2W757G*R*-V3 (GRV3) because it’s OEM rated up to …

First look at Nichia 119D D340 sm505 R70 Flashlight Modding and DIY Parts

Received the engineering sample from Nichia today. I requested the 119D instead of 219D because I’m still curious and optimistic that 119 footprint will outperforms 219 on my VR16S1 board as I suspected months ago. The larger total cooling area plus larger current capability with 119 footprint should nett higher max output than the less than optimal 219 thermal pad surface. This 119D also very interesting with it’s “glass lens” as described in the data sheet. Let’s find out if this new LED has …

[Clemence]

clemence · 2019-08-20T18:01:10.000Z

WalkIntoTheLight:

Okay, from that chart, it looks like the damaging factor is heat at the junction point, not the amount of current flowing. Current plays some role, but it appears if you properly heat-sink things, most of damage is mitigated.

And even worst-case, it doesn’t seem the L70 lifetimes are affected much at all; just the L95 lifetimes.

I’m not sure if we can extrapolate accurately past Cree’s 3 amp testing, but if you’re right that 5A gives 500 hours until the LED is only 95% as bright, then that’s why I’m not seeing any drop at all after 15 or 20 hours.

I’m okay if I get 500 hours at almost “new” output levels. That would explain why people aren’t complaining.

Please check again, it’s the current that heavily affecting. Check those tests with the same tj but different current. Roughly 1A increase in current reduce the lifetime 4 times.

[Clemence]

WalkIntoTheLight · 2019-08-20T18:11:07.000Z

clemence:

WalkIntoTheLight:

Okay, from that chart, it looks like the damaging factor is heat at the junction point, not the amount of current flowing. Current plays some role, but it appears if you properly heat-sink things, most of damage is mitigated.

And even worst-case, it doesn’t seem the L70 lifetimes are affected much at all; just the L95 lifetimes.

I’m not sure if we can extrapolate accurately past Cree’s 3 amp testing, but if you’re right that 5A gives 500 hours until the LED is only 95% as bright, then that’s why I’m not seeing any drop at all after 15 or 20 hours.

I’m okay if I get 500 hours at almost “new” output levels. That would explain why people aren’t complaining.

Please check again, it’s the current that heavily affecting. Check those tests with the same tj but different current. Roughly 1A increase in current reduce the lifetime 4 times.

[Clemence]

Temperature of 55C with 2100mA current, L95 = 66,500 hours (or 54,400 in another set).
Temperature of 105C with 2100mA current, L95 = 14,900 hours.

Same current, different temperature, and the L95 is a third the value.

But, yeah, current also does hit the L95 values pretty hard too.

The L70 seems to be relatively unaffected by current, though.
45,400 hours with 1.5 amps. 36,300 hours with 3.0 amps.

Also, based on how those values jump around on that chart, I’m betting there’s a huge margin of error.

Lightbringer · 2019-08-20T18:21:59.000Z

WalkIntoTheLight:

Okay, from that chart, it looks like the damaging factor is heat at the junction point, not the amount of current flowing. Current plays some role, but it appears if you properly heat-sink things, most of damage is mitigated.

Think of it like this… On a macro scale, a light with lousy thermal path doesn’t even get very warm to the touch, yet the LED is screaming because it’s pretty much on fire. Cold outside air, etc., keeps the body of the light cool, yet the LED still screams in searing pain.

Same thing on a micro scale. The LED can be cooled from the bottom (Al or Cu, doesn’t matter) as much as you want, but the current density inside the chip still cooks it from the inside as well.

That’s why an LED that isn’t reflowed well to its star has a lousy thermal path and can go blue and die with much less current than one that’s reflowed perfectly.

And yeah, those figures for θ_JA are made up of a whole bunch of “partial” θs. First one’s from junction to the chip’s substrate or something.

clemence · 2019-08-20T18:36:50.000Z

Lightbringer:

WalkIntoTheLight:

Okay, from that chart, it looks like the damaging factor is heat at the junction point, not the amount of current flowing. Current plays some role, but it appears if you properly heat-sink things, most of damage is mitigated.

Think of it like this… On a macro scale, a light with lousy thermal path doesn’t even get very warm to the touch, yet the LED is screaming because it’s pretty much on fire. Cold outside air, etc., keeps the body of the light cool, yet the LED still screams in searing pain.

Same thing on a micro scale. The LED can be cooled from the bottom (Al or Cu, doesn’t matter) as much as you want, but the current density inside the chip still cooks it from the inside as well.

That’s why an LED that isn’t reflowed well to its star has a lousy thermal path and can go blue and die with much less current than one that’s reflowed perfectly.

And yeah, those figures for θ_JA are made up of a whole bunch of “partial” θs. First one’s from junction to the chip’s substrate or something.

+1
This is exactly what I wanted to say earlier. It’s the thermal bottleneck “above” the Aluminum Nitride substrate. Every material with thickness has thermal impedance/resistance. With every resistance there is heat build up. Don’t forget that the current flow inside the chip also face resistance.

[Clemence]

WalkIntoTheLight · 2019-08-20T20:20:01.000Z

I get that, and I don’t disagree with the later stats you posted. But they seem to contradict your earlier statements, where you claim that the LED is damaged in just seconds at moderate over-driving, and you can measure that effect.

I’m not at all surprised that a modestly over-driven LED might show a 5% decrease in output after 500 hours of use.

What I was surprised at was your assertion that it happens in just seconds, even with your good heat-sink.

So, I’m not sure what your belief is now. Is it seconds, or hundreds of hours?

clemence · 2019-08-20T20:27:28.000Z

WalkIntoTheLight:

I get that, and I don’t disagree with the later stats you posted. But they seem to contradict your earlier statements, where you claim that the LED is damaged in just seconds at moderate over-driving, and you can measure that effect.

I’m not at all surprised that a modestly over-driven LED might show a 5% decrease in output after 500 hours of use.

What I was surprised at was your assertion that it happens in just seconds, even with your good heat-sink.

So, I’m not sure what your belief is now. Is it seconds, or hundreds of hours?

Seconds. It’s like 10 seconds, less with fragile E21A
EDIT: My rule is to never get past 1,5x rated max current.

[Clemence]

WalkIntoTheLight · 2019-08-20T21:24:41.000Z

clemence:

WalkIntoTheLight:

I get that, and I don’t disagree with the later stats you posted. But they seem to contradict your earlier statements, where you claim that the LED is damaged in just seconds at moderate over-driving, and you can measure that effect.

I’m not at all surprised that a modestly over-driven LED might show a 5% decrease in output after 500 hours of use.

What I was surprised at was your assertion that it happens in just seconds, even with your good heat-sink.

So, I’m not sure what your belief is now. Is it seconds, or hundreds of hours?

Seconds. It’s like 10 seconds, less with fragile E21A
EDIT: My rule is to never get past 1,5x rated max current.

Okay, but you later said,

clemence:

My guesstimate is at 4A your L95 = ~2000 hours, and at 5A L95 = ~500 hours. Output depreciation is not an on/off process, everything starts as soon as there’s current flowing.

That seems to say 500 hours when driven at 5A (1.66x rated max current). Or are you saying that the full 5% damage gets done in the first 10 seconds, then almost nothing after that?

Are we talking about the same LEDs? I thought this was about Cree LEDs, like the XML or XPL. What current are you using to do significant damage in 10 seconds? It’s got to be way more than 5A. If you’re trying to shove 8 or 9 amps into the LED, then I can totally believe 10 seconds. But if it’s 5 amps, I don’t think Cree’s LM80 stats or personal usage of high-output lights backs that up.

BlueSwordM · 2019-08-20T21:53:04.000Z

Yeah, let’s be careful about what the datasheets says.

The LEDs are probably tested on non DTP MCPCBs, or high performance aluminium MCPCBs.

The type of MCPCBs we use transfer heat much more effectively to the body, meaning the LED is probably running much cooler under load vs on an aluminium MCPCB.

And most of our lights can’t overdrive LEDs for very long because of thermal control.

clemence · 2019-08-20T21:56:36.000Z

That 500 hours was my guesstimation derived from Cree’s LM80 data. But Cree never published anything above 3A - their max rated current.

I tested below LEDs:

- Nichia: 219C, 319A, E21A, 144AM/AR, 757GRV3, E17A

- Cree: XPL, XPL-HI, XPG, XTE, XPG2, XPE2, XHP50, (sorry, no XML or XML2)

- Some Citizen COB

Some unknown LED bulb LEDs

Cree is tougher than other LED I tested, but still suffer invisible damage (output reduction) at above 1,5 - 2x rated current. Whenever I get past max rated current, whenever I redo the output test, the voltage always reduced. The farther away from max rated current, the lower the voltage became. Lower forward voltage is a clear sign that output has permanently reduced. My test requires me to keep the LED at certain current for almost 2 minutes in each increment. I only write down the output after the voltage has stabilized. LED Vf is a very accurate good thermal sensor.
But as I said earlier, I sometimes forgot or skipped something and made me step backwards. Whenever I did this above max rated current, the voltage always permanently reduced. Even if I just stayed at the higher current step for few seconds

[Clemence]

djozz · 2019-08-20T22:56:00.000Z

I have never done specific tests for this, so what I know is some peculiarities that many leds showed that I tested.

1) after a led test (that usually goes well beyond specifications), in cases that they did not seem to have damaged the led, I often go back to a reasonable current within specs and let the led stabilise at that current. The output is always a bit (often just 1 or 2%, nothing serious usually) lower than the initial output at that current, and the voltage has dropped a bit. If I increase the current a bit to get back to the voltage that I originally measured, the output is exactly what I originally measured with non-increased current (so after overdriving the output seems to “follow” the voltage rather than the current). Of course now the output per Watt has suffered a bit.

2) there seems to be a recovery of the led over a longer period than a few minutes. Not just with bare leds but also with output/throw measurements of flashlights, I noticed several times that a re-measurement after a couple of minutes always gives a lower result, even when the light was fully cooled down, but when waited a day the measurement was back up again at the first value.

But as I said, this is all circumstantial observations that I gathered over time without specific testing for it, Clemence has already done more proper testing on this than I.

You can do some real quantitative testing specificially looking for the effects and get much more information, about how partial this recovery is for instance.

thisnameisvalid · 2019-08-20T23:16:58.000Z

Has anyone ever done a long term lumen maintenance test on an LED? Something like running an XP-L HI at 5A (typical DD current) for a few hours or days to see how it holds up.

WalkIntoTheLight · 2019-08-20T23:22:08.000Z

Interesting stuff. 1% or 2% doesn’t sound too bad, if you were driving things really hard to test failure currents. Strange about the long-term recovery of the LED, though. I would have thought if it didn’t recover within a few minutes, it wouldn’t. I presume it wasn’t just the battery recovery you were using.

WalkIntoTheLight · 2019-08-20T23:26:02.000Z

thisnameisvalid:

Has anyone ever done a long term lumen maintenance test on an LED? Something like running an XP-L HI at 5A (typical DD current) for a few hours or days to see how it holds up.

Well, as I mentioned earlier, I re-tested a Astrolux C8 (FET driver with Samsung 30Q) that I’ve used for about 15-20 hours on max output over the past year-and-a-half. It gave the same output as when it was new.

However, my measurements are much less accurate than the 1% or 2% that djozz was talking about. And I’m not sure how much output variance clemence was seeing.