Can I run this high power LED without a heat sink?

[Starlord]

Hello,

I've got a question for ya.  It may sound stupid at first glance, because I know generally high powered LEDs are run with a heat sink.  But in my application I'm space constrained and even if I wanted to, I don't know how I'd attach a heat sink, with nothing but the LED itself to attach it to and with the pads so close to the mounting holes that the head of any screw would surely contact them.

So the LED I want to drive is this one:
http://www.digikey.com/product-detail/en/XMLCTW-A0-0000-00C3ABB02/XMLCTW-A0-0000-00C3ABB02CT-ND/3711461

As you can see it's expensive.  So I don't want to blow it up.

Now, the LED is rated for up to 1A per die, which I assume means per color.  But I don't want to run it that high.  I've found a simple way to drive each color at up to 250mA.  I could limit it to 150mA but I'd prefer not to go below that.

Futhermore, I will likely be lighting only two colors at a time.  On top of that, I will be flashing the LEDs and/or fading them in and out, so they would be on around 50% of the time.  In other words, if you calculate the total power dissipation for all four colors running @ 250mA each, then in actual usage, the power that needs to be dissipated would be more like 25% of that.

That said, for simplicity's sake, I'd like like to know, if I have all four dies lit at once, what current would I need to limit each one to in order to keep the LED's junction temperature from exceeding, say 100 degrees?

I found this nice calculator:
http://support.luxeonstar.com/customer/portal/articles/179490-how-do-i-determine-what-size-of-heat-sink-i-need-includes-heat-sink-calculator-

But it wants me to input data for the heat sink which doesn't exist. 

I put in the following values:
Vf = 3.9 (worst case)
I = 250mA
LEDs = 1 (can't change this, so if I want to calculate more than one die on at a time I'd just multiple the current by 4)
And that gives Watts = 1
So if I had two of the dies on at once, it would be 2 watts.   But let's just continue with 1 for now.

Now I get to the thermal calculations...
Maximum Junction Temp = 100C
Junction to Thermal Pad = 3.5 (that's what the datasheet says anyway)
LED mounting base = 6 (I looked all over the place for information on what the termal resistance of these aluminum PCBs was but the numbers were all over the place from 2W/m-K to 150W/m-K, and that's not even C/W and I have no idea how to convert it or if it is even the same thing, but C/W was listed nowhere, so I just assume the default value of 6 is good here.)
Heatsink attachment = I just left this at the default of 4.5
Ambient Temp = 25 (again just left it the default)

And then I click calculate and it tells me...
Maximum Heat Sink Thermal Resistance = 62.9C/W

Okay...  So just for a sanity check I increased the LED current, and that value goes down.  So down means the heat sink needs to be better.

But wait..  My PCB has a thermal resistance of 6C/W.  That's better than what they say the heat sink has to be.  So what gives?

Are they making some assumption about the size of the heat sink?  Or is the PCB in fact good enough as a heat sink when powering a single die with 250mA?

If I put in 1000mA, to simulate all 4 leds being on at once and driven with 250mA, I get 5.2C/W. 

Except that forward voltage I used was the worst case for the green die.  If I average all the maximum forward voltages instead, I get 3.475V.   And if I put that in instead, then the heat sink calculation rises to 7.6C/W.

But wait again.  That thermal pad conduction...  Is that affecting the calculation?  Well, I have no thermal pad, since I have no heat sink.  So maybe I should set that to 0?  If I do that then the heat sink value rises to 12C/W.

Am I on the right track here?  I kinda doubt it.  I'd be surprised if I can drive 1A constantly into this LED (250mA per die) with no heat sink and only get what I assume would be a 50C rise (since the thermal caluclation says the heat sink only needs to be 12C/W but the PCB is 6C/W and I wa calculating for a 100C rise)

[IanB]

I looked all over the place for information on what the termal resistance of these aluminum PCBs was...

How about this?



But here's the snag. Resistance is for heat flowing in at one side and out the other. If the other side is a dead end (no heat sink), then the heat has nowhere to go and the effective resistance becomes very high. In short, the PCB will get hotter and hotter until it is nearly equal to the temperature of the die. Then the die will die 

Generally speaking, unless you are running a small LED at about 20 mA, you need a heat sink.

[Starlord]

Huh, strangely enough I just found this video:


He's running the LED without a heatsink at 1A and it takes 3 minutes to rise to 150C (300F on the meter).  According to the times he posted, it rises 129 degrees in the first minute, and then takes two minutes to rise another 100 degrees. 

I think the temperature rises more slowly as it gets hotter because the LED's forward current goes down as it heats up (Pd = Fv * current), or because heat transfers more quickly between two mediums when the temperature difference is greater.  (I don't know if that is true or not, but I think it is.)

It also takes 168s to go from 100F to 300F, but takes 250s to drop from 300F to 100F. 

Now I have no idea if this is right, but I have a hunch...

168 / 250 = 0.672

So if he was powering the LED with 1A, 1A * 0.672 = 672mA

My theory is that at 672mA, the LED would be at equilibrium... cooling off as fast as it heats up.  So if my average total current were below 672mA my LED would stay relatively cool.

And that means I could drive the four dies in my LED at 168mA each and it wouldn't heat up.  And to add a safety margin, I could run them at 150mA.

Or, if I assume that I will be driving only two at a time at a 50% duty cycle, then 600mA per die should be possible.
(My driver is limited to 250mA each anyway.)

What do you think?   Am I on the right track here?

[Starlord]

IanB:
The link I posted was for a bare LED, not one mounted on a star board.  That thermal value is for the base of the LED to the PCB, but not he PCB itself.

Also there is no such thing as a dead end for the heat.  When the heat reaches the back side of the PCB it either flows to the heat sink or to the air.  But the heat sink can't hold an infinite amount of heat, and it won't cool down unless the heat is flowing away from it.  So where does the heat go?  To the air. 

The PCB is a heat sink.  It's just a lot smaller and with less surface area.  The heat sink thermal value that calculator is calculating is telling us in a way how much surface area it needs to dissipate the heat to the air.  But the problem is I don't know what the thermal resistance between the PCB and the air is to tell how good it is compared to he heat sink.  I think the thermal value there is just how much resistance there is from the LED side to the heat sink side of it.

[Corporate666]

There is no way you are going to get anywhere near 1A of drive current without a heat sink.

Total guess (based on lots of experience though) I would say that you would be lucky to get more like 100mA for all 4 dies before you start having problems (assuming bare LED).

If you aren't going to be running the LED anywhere near it's max level, you would be a lot better off going with a lower power LED and running more of them.

Even 5050 LED's without thermal pads will overheat on a bare PCB without adequate ventilation. 

[Starlord]

There is no way you are going to get anywhere near 1A of drive current without a heat sink.

Total guess (based on lots of experience though) I would say that you would be lucky to get more like 100mA for all 4 dies before you start having problems (assuming bare LED).

So you think a constant current of 400mA (100mA per die) would be the limit?  Or did you mean 100mA in total?

Because if 400mA is okay that might still be workable for me.  Like I said, the dies will most likely be on a 50% duty cycle and only two would be driven at a time.  So average total current even if I ran the dies at 250mA ea would still be only 250mA.  (But I'd probably go for 150mA each just for that safety margin.)

If you aren't going to be running the LED anywhere near it's max level, you would be a lot better off going with a lower power LED and running more of them.

Better off how?  Are we talking money, or luminosity?  I can get the LEDs for less than they cost on Digikey and it would probably cost me more to have custom PCBs made, and if that's not gonna help with my thermal problem at all, (and if I used FR4 it would pboably be worse) then using what's already available and assembled seems like the best option even if I'm not using the LEDs to their full potential.

[Corporate666]

So you think a constant current of 400mA (100mA per die) would be the limit?  Or did you mean 100mA in total?

I meant 100mA in total, not per die.  It's really impossible to say what you could run the LED at without knowing a lot more about your use and your implementation.  There will always be thermal transfer.  I guess worst case would be soldering wires to the pads on the LED itself and leaving the thermal pad free in air inside a plastic enclosure.  In that case, even 25mA per die might heat things up too much.  But just from experience, I think you would be lucky to get 100mA in total without any heatsinking whatsoever.  If you truly mean no heatsink - not even an FR-4 PCB, then even 100mA in total is likely not achievable.

Better off how?  Are we talking money, or luminosity?  I can get the LEDs for less than they cost on Digikey and it would probably cost me more to have custom PCBs made, and if that's not gonna help with my thermal problem at all, (and if I used FR4 it would pboably be worse) then using what's already available and assembled seems like the best option even if I'm not using the LEDs to their full potential.

Well, I don't think it would make sense to buy an expensive LED capable of being driven at a high level and then running it at the same level as you could have run a much cheaper LED - and get the same light output.

Can you give some more detail about what your exact application is?  Like what light output level do you need?  What is the lifespan needed?  What is the housing going to be like?  What type of environment will it be used in?   How are you driving the LED - constant current driver?  What duty cycle (% on and off time and frequency) are you running the LED at?  What are your size constraints?


The more detail you can give, the better.  But in a nutshell, LED's (regardless of type, form factor, mounting option, voltage and current) operate in a somewhat narrow efficacy range.  Especially when it comes to color LED's, you're not going to find an LED that produces 1/10th the heat for a given light output compared to another LED.  It's mostly a function of current and voltage... so running your 1A (capable) LED at 50mA per channel (for example) isn't going to make appreciably more or less light than a 100mA per channel LED being operated at 50mA. 

[IanB]

IanB:
The link I posted was for a bare LED, not one mounted on a star board.  That thermal value is for the base of the LED to the PCB, but not he PCB itself.

Also there is no such thing as a dead end for the heat.  When the heat reaches the back side of the PCB it either flows to the heat sink or to the air.  But the heat sink can't hold an infinite amount of heat, and it won't cool down unless the heat is flowing away from it.  So where does the heat go?  To the air. 

The PCB is a heat sink.  It's just a lot smaller and with less surface area.  The heat sink thermal value that calculator is calculating is telling us in a way how much surface area it needs to dissipate the heat to the air.  But the problem is I don't know what the thermal resistance between the PCB and the air is to tell how good it is compared to he heat sink.  I think the thermal value there is just how much resistance there is from the LED side to the heat sink side of it.

Thanks for the education in something that happens to be my profession 

Sorry I missed the fact that you had seen the value I posted and that I misread your post. But beyond that thermal interface you are lacking hard data.

If this was an electronics question you would be asked to post a schematic so people could see what your design looks like. The thermal situation is no different. Before any analysis can take place you need a diagram of your mounting arrangements. How big is the  PCB? What are its specifications? How is it mounted? Is it horizontal or vertical? In free air, on a back plate, or in a shroud? Mounted against insulating material or some other conducting material?

That kind of information would allow some basic estimates of heat transfer to be possible. But basic estimates only. For accurate design you would need to make some measurements to get hard data or else over design the system so that you have a good margin of safety.

But without hard data you have to go with the most conservative assumption: lack of effective heat sinking will lead to a dead LED at any drive currents above a few milliamps.

[IanB]

Huh, strangely enough I just found this video:

That's an interesting video, and it can be analyzed, for that LED under those conditions: sitting in free air with significant convective cooling. The data from the video can be plotted and extrapolated to estimate the maximum temperature that would be have been reached if the experiment had continued:



Extrapolation of the data predicts a maximum temperature of about 330°F, a temperature rise of 260 degrees above ambient. Now to a first order approximation the temperature rise may be assumed proportional to the power input, which is proportional to the drive current for a nearly constant forward voltage. So if 1 A gives a temperature rise of 260 degrees, then it may be estimated that 0.5 A would give a temperature rise of 130 degrees, and 0.25 A a temperature rise of 65 degrees.

But these estimates are for that particular situation. If the LED were mounted differently then the results would be different. Also, an LED may not fail immediately with a puff of smoke. It may fail slowly with diminishing light output if not cooled adequately.

[Corporate666]

That's an interesting video, and it can be analyzed, for that LED under those conditions: sitting in free air with significant convective cooling. The data from the video can be plotted and extrapolated to estimate the maximum temperature that would be have been reached if the experiment had continued:

The other thing to keep in mind (and I know you know this, of course, IanB), but that LED is mounted to a metal core PCB - a heat sink.  Not a great heat sink, but a massive amount more surface area than simply the LED's thermal pad.  Furthermore, he is measuring the temperature on the base of the PCB, which is going to be significantly lower than the junction temperature.  Third, that's a Luxeon III LED, which used a pretty giant die and those LED's were bulletproof by today's standards. 

Not to mention, heat in an RGB or RGBW LED is even more critical to manage than a single color LED, because the dies will use different chemistry.  The red die will likely be AlInGaP and the others InGaN... and looking at page 4 on the datasheet, at 100C, the red die is only emitting half the flux it does at 25C. White and green have dropped to 85% and blue is still close to 100%.  So if the RGB LED is being used for color mixing, maintaining color accuracy becomes a significant problem when things heat up.  And it's not just "meh, so it changes slightly - big deal" because losing half your red compared to blue would mean a huge change in the color of purple light, for example.

LED's don't like to be run hot... a lot more info on the exact application is needed, Starlord.

[Psi]

I think the temperature rises more slowly as it gets hotter because the LED's forward current goes down as it heats up (Pd = Fv * current), or because heat transfers more quickly between two mediums when the temperature difference is greater.  (I don't know if that is true or not, but I think it is.)

Yep, the temp rises slower because the hotter it gets the faster heat gets pulled away by the surrounding material (be that a heatsink or free air)
Like water traveling down a slope, where the temp difference is the angle of the slope.

Eventually it reaches equilibrium where the heat produced equals the heat being drawn away and the temp stops rising all together.

[Starlord]

So you think a constant current of 400mA (100mA per die) would be the limit?  Or did you mean 100mA in total?

I meant 100mA in total, not per die.  It's really impossible to say what you could run the LED at without knowing a lot more about your use and your implementation.  There will always be thermal transfer.  I guess worst case would be soldering wires to the pads on the LED itself and leaving the thermal pad free in air inside a plastic enclosure.  In that case, even 25mA per die might heat things up too much.  But just from experience, I think you would be lucky to get 100mA in total without any heatsinking whatsoever.  If you truly mean no heatsink - not even an FR-4 PCB, then even 100mA in total is likely not achievable.

When I say no heat sink, I mean no heat sink with fins.  The LED will be mounted on a 20mm wide aluminum star board like the video above. 

Also, while The LED in the video was in free air, the enclosures my LED is installed in will vary in enclosed-ness.  Two possible cases are several inches inside a plastic tube, or at the bottom of a small wooden box.

Well, I don't think it would make sense to buy an expensive LED capable of being driven at a high level and then running it at the same level as you could have run a much cheaper LED - and get the same light output.

I can get these LEDs for around $6, already mounted on an aluminum star board.  If I were to design my own PCB it would take several days, and it would be made of an inferior heat dissipation material - FR4.  I also have limited space, and I don't think just one of those cheap square neopixel-like LEDs would have the kind of light output I need. 

I'd been using these bulbs made for car tail lights that were one color that use LEDs like that and they needed a minimum of five of a single color to achieve sufficient brightness.   They also claim to be 1W for that small one, and up to 2.5W for a larger model with 13 LEDs on it.  But they're all tiny - the smallest is like a 1/2" long 1/2" diameter cube of LEDs, and the largest is the same diameter but an inch long.  And these things are meant to be installed inside, say your enclosed interior lamp and left on for long periods of time and they seem to have no issue.  Also they're just made of FR4.  I guess the actual surface area is greater...  in the case of the 2.5W one you'd have 25mmx10mm per side, and four sides, so that's 25x40mm, twice as much as the aluminum star board.  But the aluminum star board has both sides exposed to air, while the 2.5W LED has them arranged in a capped off tube.  So the star board has 20mmx20mm x 2 exposed to air.  So all other things being equal, and the aluminum providing no improvement in dissipation in such a small board size, my LED and the 2.5W automotive bulb have the same surface area exposed to air.

Can you give some more detail about what your exact application is?  Like what light output level do you need?

Well I need it to be as bright as possible within my other limitations.  I chose this LED not only because it is already mounted on a PCB for me, but also because it has a ridiculously high lumens per watt.  But I couldn't tell you exactly how many lumens I need.  I have no way of measuring how many lumens those bulbs I was originally using actually put out and even if the Chinese site I got them from had specs, who knows if they'd be correct.

What is the lifespan needed?

I guess three years, but they're not going to be used constantly.  This is something you take out occasionally and use for a  few hours and then put it away for weeks, like a flashlight.

What is the housing going to be like?  What type of environment will it be used in?

I mentioned that above already.  Environment?  Indoors, outdoors, same conditions as you would use a flashlight.

How are you driving the LED - constant current driver?  What duty cycle (% on and off time and frequency) are you running the LED at?  What are your size constraints?

Yes, I will be using a constant current driver, and I already mentioned the duty cycle is around 50% per die, with one or two dies lit at a time.  But I'd like to maintain flexibility and be able to do my own calculations as to if I can boost the current in certain instances, so I'm really looking for numbers that tell me what will be okay with a 100% duty cycle and all four dies lit.

The more detail you can give, the better.  But in a nutshell, LED's (regardless of type, form factor, mounting option, voltage and current) operate in a somewhat narrow efficacy range.  Especially when it comes to color LED's, you're not going to find an LED that produces 1/10th the heat for a given light output compared to another LED.  It's mostly a function of current and voltage... so running your 1A (capable) LED at 50mA per channel (for example) isn't going to make appreciably more or less light than a 100mA per channel LED being operated at 50mA.

You sure about that?  I mean Digikey doesn't list specs for color LEDs but they do list lumens per watt for white LEDs, and as you can see there's a wide range of values:
http://www.digikey.com/product-search/en/optoelectronics/led-lighting-white/

Also this RGB led, which is one of the few Digikey has:
http://www.osram-os.com/Graphics/XPic5/00115404_0.pdf

States that its efficiency is 45lm/W for red and green and 15lm/W for blue, whereas the Cree I chose... going by the lm @ the test current of 350mA, since they don't list the numbers in their datasheet, appears to get more like 68lm/W for red, 87lm/W for green, 25lm/W for blue, and 101lm/W for white.

In other words, if I use the Cree it will be 1.5x as bright for the same power, or run 33%? cooler at the same brightness.

So yeah, I guess it's not 1/10th the heat, but 1.5x as bright for the same power is nothing to sneeze at when I'm trying to push the thing as far as possible.

[Starlord]

If this was an electronics question you would be asked to post a schematic so people could see what your design looks like. The thermal situation is no different. Before any analysis can take place you need a diagram of your mounting arrangements. How big is the  PCB? What are its specifications? How is it mounted? Is it horizontal or vertical? In free air, on a back plate, or in a shroud? Mounted against insulating material or some other conducting material?

The problem is, it's variable.  I can tell you one configuration would be that the LED is mounted on the aluminum star board (always on the star board) 6" inside a piece of acrylic tube with one end open to the air.  That's probably the worst case. 

As for the PCB, it's a star board like the one in the video.  Like this:
http://www.digikey.com/product-detail/en/MTG7-001I-XML00-RGBW-BC02/1125-1176-ND/4031704

But without hard data you have to go with the most conservative assumption: lack of effective heat sinking will lead to a dead LED at any drive currents above a few milliamps.

But what is an effective heat sink?

I know the star board is a heat sink.  It may not be a very effective one, but it is one.  Problem is, I don't know how effective it is.  Nor how to compare its effectiveness with a real heat sink.  For heat sinks I see one unit for power dissipation W/m-K, while when I try to look up alumiumum pcbs I get another unit.  C/W.  I don't know how to reconcile these.   I tried doing some calculations with a PCB thermal calculator, like how you'd calculate heat dissipation for a chip, but I don't know if the material being aluminum makes a difference or not.  The calculator was assuming FR4 of course.   

http://circuitcalculator.com/wordpress/2007/02/16/pcb-thermal-copper-area/

If I enter 3.5 for Theta JC there and leave it at 1W dissipation (if I supplied two LED dies at 150mA each constantly, that would be around 1W) and leave the rest of the values as they are, it tells me I need 10cm^2 of PCB area, but both sides can be counted, and the star boards have 3.14cm^2 of PCB area on each side, so that's 6.28cm^2 total.

Which is close to my desired value.  If I'm flashing these LEDs so they're on only 50% of the time, then according to this calculator I ought to just scrape by.

I wonder though if that value for Theta SA is accurate for aluminum.

[mikerj]

But what is an effective heat sink?

I know the star board is a heat sink.  It may not be a very effective one, but it is one.  Problem is, I don't know how effective it is.  Nor how to compare its effectiveness with a real heat sink.  For heat sinks I see one unit for power dissipation W/m-K, while when I try to look up alumiumum pcbs I get another unit.  C/W.  I don't know how to reconcile these.   I tried doing some calculations with a PCB thermal calculator, like how you'd calculate heat dissipation for a chip, but I don't know if the material being aluminum makes a difference or not.  The calculator was assuming FR4 of course.   

http://circuitcalculator.com/wordpress/2007/02/16/pcb-thermal-copper-area/

If I enter 3.5 for Theta JC there and leave it at 1W dissipation (if I supplied two LED dies at 150mA each constantly, that would be around 1W) and leave the rest of the values as they are, it tells me I need 10cm^2 of PCB area, but both sides can be counted, and the star boards have 3.14cm^2 of PCB area on each side, so that's 6.28cm^2 total.

The bit you are missing is that a heatsink has to get rid of the heat it has sunk, usually into the surrounding air.  If the heatsink is in still air, and mounted within an insulating container then it's effective thermal impedance will rise dramatically, far above the figures that you will see within a heatsinks specifications.

There are so many variables that trying to predict a safe value of current is nearly impossible.  There are very expensive modelling tools such as Flowtherm that can help model these conditions, but even then nothing beats doing some actual testing.

FWIW I helped someone who was converting a traditional miners helmet lamp to use a CREE LED.  The LED was mounted on a 'star' and the star was thermally bonded to the largest aluminium plate he could fit inside the lamp (a circular plate, maybe 40mm diamater).  However, the lamp body is made from some kind of plastic (bakelite possibly) and at 750mA the heatsink would reach 100C within a minute or so.  At typical room temperatures the LED could not be run continuously at more than about 200mA, and rather less than that to keep the LED comfortable.  Fortunately the design included thermal monitoring via a small micro and a thermistor (my job), so the micro could turn the LED current down to maintain a maximum permissible temperature

[Starlord]

Here is an example of the kind of bulbs I was using previously:
https://www.superbrightleds.com/moreinfo/festoon/3710-led-bulb-6-smd-led-festoon/661/

95 lumens.  Intended to run continuously.  Even has heat sensitive electrolytic caps on the back.  And the festoon base it plugs into is a rather poor heat sink, being a couple metal clips in a plastic base.

The Cree XM-L RGBW LED that I was looking at on the other hand is mounted on a star board of similar size and at 350mA the white die should be puttng out 93lm.

So you can see that these two are pretty similar.  Furthermore, the automotive bulb is designed to be put inside an enclosed space - a dome light, where it will have no direct exposure to air.  My LED will be exposed to fresh air, through a 6" acrylic tube.

Furthermore, if you look at the specs of the LED, it draws 136mA @ 12V.  That's 1.6W!  And for the same lumen output as my LED at 1W no less.  That would be like driving my LED at 560mA.  And that's equivalent to 140mA per die.

It can also be assumed that whoever designed those automotive LEDs didn't plan for them to fail a month after they were installed.

So how can this be?  The numbers you're giving me - 100mA max - is like 0.3W.  That's 5x less than what this LED is typically run at.  And I have had no problems at all with these LEDs.  I even built an Iron Man Arc Reactor for a friend out of three of them and he wore the thing all day without issue, aside from a slight "sun burn".


And here's another high power LED for cars:
https://www.superbrightleds.com/moreinfo/festoon/universal-high-power-led-kit--cob-led-27-pcb-w-festoon-base/1237/2887/

150mA @ 12V = 1.8W.  Even higher! 
(And only 90lm, so even lower efficiency!)

And this one doesn't even have metal to metal contact like the festoon one, it's just got some double sided tape to stick to whatever.  Probably not a very good heat conductor.

[DmitryL]

Here is an example of the kind of bulbs I was using previously:
https://www.superbrightleds.com/moreinfo/festoon/3710-led-bulb-6-smd-led-festoon/661/

95 lumens.  Intended to run continuously.  Even has heat sensitive electrolytic caps on the back.  And the festoon base it plugs into is a rather poor heat sink, being a couple metal clips in a plastic base.
It can also be assumed that whoever designed those automotive LEDs didn't plan for them to fail a month after they were installed.

[...]

You can buy any cheap garbage that overheats and therefore LEDs degrade pretty soon.
Do you have a document that guarantees you that this lamp will "run continuously 24/7 in a confined space" with all nice graphs, bells and whistles (i.e. datasheet)?
What do you do when this cheap stuff dies because of the overheat ? Probably go and buy another piece...

[Starlord]

There are so many variables that trying to predict a safe value of current is nearly impossible.  There are very expensive modelling tools such as Flowtherm that can help model these conditions, but even then nothing beats doing some actual testing.

Unfortunately, there are just as many variables in my usage case.  The one I've described is just one of many.  Imagine if you were an engineer at Adafruit making Neopixel boards and someone asked you to describe the enclosure and whether they'd be attached to it in some way to provide heat sinking.  They couldn't tell you because they don't know what the end user is going to do with them.  And in my case I have many different projects I want to use these for, and others will be using them as well and I have very little control over how they get installed.

So I'm just looking for something like probably safe current, when mounted on an aluminum star board, with no additional heat sink, in a small plastic enclosure with limited air flow, with only one or two LED dies in use. 

But for more info on typical usage -
In one case I would have a single white LED lit constantly.
In another I would have white and another color flashing at a 50% duty cycle constantly.
In yet another I might have a white led die lit constantly, with another color led die flashing at a 50% duty cycle, but for around 30 seconds at a time, with long pauses between uses.

[IanB]

In the video I looked at we saw an experiment performed and the results from such an experiment are far more useful than paper calculations. For example, from that video the thermal resistance from PCB to ambient can be estimated as follows:

Input power: 1 A x 3.4 V = 3.4 W
If we assume a radiant efficiency of 20% the heat dissipation would be 3.4 x 0.8 = 2.7 W
At steady state the PCB temperature was estimated to be 260 degrees above ambient, so the thermal resistance PCB to ambient was 260/2.7 = 96 degF/W = 53 K/W.

In the experiment the PCB was in free air providing convective cooling. If the PCB were to be mounted on a plastic or wooden back plate inside a long tube there would be little free air circulation and the thermal resistance would be expected to be much higher.

But you can do similar experiments yourself with your exact set-up and get hard data. That will tell you much more about the constraints than we can tell you.

[Corporate666]

When I say no heat sink, I mean no heat sink with fins.  The LED will be mounted on a 20mm wide aluminum star board like the video above. 

Also, while The LED in the video was in free air, the enclosures my LED is installed in will vary in enclosed-ness.  Two possible cases are several inches inside a plastic tube, or at the bottom of a small wooden box.

So your limiting factor is going to be how much heat the star PCB can reject to ambient given the insulating properties of your enclosure.  It's absolutely impossible to give a number on that unless you could provide a lot more detail on the housing - and even then, it would be a big guess.  The best way forward is to experiment, but given a wood or plastic enclosure (and I am guessing the star PCB will be screwed to one surface?), I don't think there is any way you're going to come close to dissipating 1W or more continuous.  Now, if you're pulsing at a reasonable frequency, it comes down to your average power dissipation.

I'd been using these bulbs made for car tail lights that were one color that use LEDs like that and they needed a minimum of five of a single color to achieve sufficient brightness.   They also claim to be 1W for that small one, and up to 2.5W for a larger model with 13 LEDs on it.  But they're all tiny - the smallest is like a 1/2" long 1/2" diameter cube of LEDs, and the largest is the same diameter but an inch long.  And these things are meant to be installed inside, say your enclosed interior lamp and left on for long periods of time and they seem to have no issue.  Also they're just made of FR4.  I guess the actual surface area is greater...  in the case of the 2.5W one you'd have 25mmx10mm per side, and four sides, so that's 25x40mm, twice as much as the aluminum star board.  But the aluminum star board has both sides exposed to air, while the 2.5W LED has them arranged in a capped off tube.  So the star board has 20mmx20mm x 2 exposed to air.  So all other things being equal, and the aluminum providing no improvement in dissipation in such a small board size, my LED and the 2.5W automotive bulb have the same surface area exposed to air.

I'm not sure the point of the comparison here - there are a huge number of variables in your original idea of using the star PCB, so adding more by comparing to a different kind of LED in a different setup just makes it a lot more complicated, to the point where I don't think you're getting any useful data whatsoever, especially if you didn't measure their power draw and temperature increase inside your enclosure.

Yes, I will be using a constant current driver, and I already mentioned the duty cycle is around 50% per die, with one or two dies lit at a time.  But I'd like to maintain flexibility and be able to do my own calculations as to if I can boost the current in certain instances, so I'm really looking for numbers that tell me what will be okay with a 100% duty cycle and all four dies lit.

You said 50% duty cycle but then said it will be used for a while then put away - so that's not a constant 50% duty cycle.  The housing will have air inside which will heat up and act as a thermal battery.  Maybe you can run it at 50% duty cycle for 20 minutes before it gets too hot - is that enough?  Do you need an hour of runtime?  10 hours?  50% duty cycle implies constant operation at 50%, but it seems like that's not the case.

You sure about that?  I mean Digikey doesn't list specs for color LEDs but they do list lumens per watt for white LEDs, and as you can see there's a wide range of values:
http://www.digikey.com/product-search/en/optoelectronics/led-lighting-white/

Also this RGB led, which is one of the few Digikey has:
http://www.osram-os.com/Graphics/XPic5/00115404_0.pdf

States that its efficiency is 45lm/W for red and green and 15lm/W for blue, whereas the Cree I chose... going by the lm @ the test current of 350mA, since they don't list the numbers in their datasheet, appears to get more like 68lm/W for red, 87lm/W for green, 25lm/W for blue, and 101lm/W for white.

In other words, if I use the Cree it will be 1.5x as bright for the same power, or run 33%? cooler at the same brightness.

So yeah, I guess it's not 1/10th the heat, but 1.5x as bright for the same power is nothing to sneeze at when I'm trying to push the thing as far as possible.

Yes, I'm sure.  As I said, there is not a 10 times difference in light output between two LED's run at the same power level.  You're also sort of comparing apples to oranges above.  An LED doesn't really have a fixed efficacy.  It has a range that varies based on various factors (heat, drive level, variation among bins, and more).  Looking at the LED's you mentioned, the Cree is rated at a proportionally lower drive level, so it's going to be more efficient.  Take the green for the Osram... page 4 of the datasheet shows the light output has already fallen off substantially by 140mA (compared to a linear relationship between VI and lumens).  At 50mA, it's about 47.5% the light output of 140mA.  The datasheet doesn't list lumens but it says 45lum/watt @ 140mA, and Vf at 140mA is 3.4, or 0.476 watt which = 21.42 lumens.   At 50mA, the green die is just below 3.1V, say 0.15 watt.  Light output is 10.17 lumens, but that's 68 lumens per watt.   The Cree says 87lum/watt for it's green die.  A difference, yes... but like I said, nothing like a 10x difference. 

As for the white LED's you mentioned, I referred to colored dies specifically since white LED's introduce the difference of phosphor conversion, and that makes a big difference.  But even so, Digikey's search tool has it wrong.  If I select everything from 5 to 20 lumens/watt, all the results are the Cree MT-G2 LED.  Looking at page 3 of the datasheet for that LED, it's listed at 1100mA @ 6V (6.6 watts), and just taking the first bin (the lower flux bin), @ 5000k it's at 863 lumens which is 130 lumens/watt.   If you go down to the highest lumens/watt LED listed on Digkey, it's 265 lumens/watt for the LZ9-00CW00
from LedEngin.  But on page 5 of the datasheet, it lists the typical luminous flux at 106 lumens/watt at 350mA.  I am sure there is some unobtainable bin of the LED that when run at 100mA in a 6500k color temp and 50 CRI will achieve 250 lumens/watt with a 10ms pulse, but that's apples to oranges.  Like I said, there just isn't going to be a 10x variation between LED's.

On top of that, the eye's response to light levels isn't linear, it's logarithmic.  It takes about double the light for a person to look at two LED's and say "Yep, A is definitely brighter than B", and it takes at least 4 times the light output to appear twice as bright.  So the 22% difference in real world efficacy between the Cree and Osram LED's listed above isn't likely to be noticeable.



I think that in summary, the question you are asking isn't answerable, and with all due respect, it sounds like you have an answer in mind that you're looking to hear.  Nobody is going to be able to tell you what level you can drive your LED's at for "the most brightness possible" in an unknown housing with an unknown usage case with an unknown run time, environment and so on.  There are just too many variables - the only way you are going to get to an answer is to buy some of the LED's you're thinking of and test it experimentally. 

[Corporate666]

Here is an example of the kind of bulbs I was using previously:
https://www.superbrightleds.com/moreinfo/festoon/3710-led-bulb-6-smd-led-festoon/661/

95 lumens.  Intended to run continuously.  Even has heat sensitive electrolytic caps on the back.  And the festoon base it plugs into is a rather poor heat sink, being a couple metal clips in a plastic base.

The Cree XM-L RGBW LED that I was looking at on the other hand is mounted on a star board of similar size and at 350mA the white die should be puttng out 93lm.

So you can see that these two are pretty similar.  Furthermore, the automotive bulb is designed to be put inside an enclosed space - a dome light, where it will have no direct exposure to air.  My LED will be exposed to fresh air, through a 6" acrylic tube.

Furthermore, if you look at the specs of the LED, it draws 136mA @ 12V.  That's 1.6W!  And for the same lumen output as my LED at 1W no less.  That would be like driving my LED at 560mA.  And that's equivalent to 140mA per die.

It can also be assumed that whoever designed those automotive LEDs didn't plan for them to fail a month after they were installed.

So how can this be?  The numbers you're giving me - 100mA max - is like 0.3W.  That's 5x less than what this LED is typically run at.  And I have had no problems at all with these LEDs.  I even built an Iron Man Arc Reactor for a friend out of three of them and he wore the thing all day without issue, aside from a slight "sun burn".


And here's another high power LED for cars:
https://www.superbrightleds.com/moreinfo/festoon/universal-high-power-led-kit--cob-led-27-pcb-w-festoon-base/1237/2887/

150mA @ 12V = 1.8W.  Even higher! 
(And only 90lm, so even lower efficiency!)

And this one doesn't even have metal to metal contact like the festoon one, it's just got some double sided tape to stick to whatever.  Probably not a very good heat conductor.

I didn't say 100mA max regardless of mounting/substrate.  I said the bare LED connected to nothing.  And that might even be too generous by a lot.

And again, drawing comparisons to other LED products out there and trying to infer something from it isn't going to get you anywhere.  You could come up with a number you think works and end up being off by an order of magnitude.

Those little festoon LED's are Chinese crap, and they are going to be used in map lights, dome lights and stuff like that (i.e. not on constantly).  Not to mention - it's "designed" to run continuously - designed by who?  And what level of engineering is behind it?   There are tons of "LED replacement bulbs" that have nowhere near enough heat management to run continuously, but if it's in your brake light - maybe it can run for 2 minutes before it gets too hot - and maybe for 10 minutes before it burns out.  Who sits at a red light with the brakes on for 10 minutes?  So the comparison won't give you any meaningful data - you have to build your enclosure, get some of the LED's in question, see what you can run them at, dial it back a bit for breathing room, and then you'll know the answer.

[Starlord]

I didn't say 100mA max regardless of mounting/substrate.  I said the bare LED connected to nothing.

I mentioned in my first post that the LED would be attached to a PCB and that it would be aluminum, so you can hardly blame me for assuming you were talking about the usage scenario I described, and not something crazy like trying to attach wires directly to the bottom of a high powered SMT LED with no leads.

and with all due respect, it sounds like you have an answer in mind that you're looking to hear.

Well of course I do, I have a target I'm aiming for.  But I'm listening to all the feedback, and I've learned a lot from the answers I've gotten to my questions.  I understand that nobody can give me a definitive answer without knowing exactly how the LED will be used.  But the information I've been provided with has helped me narrow down what might be realistic, and what is wildly unrealistic and which of my calculations were on the right track and which weren't.

The bottom line is I was only looking for a best guess, so I have a starting point.  I have a rough idea of how the LEDs will be used, and what their duty cycle will be, and how long they'll have to cool down between uses, and with the information from that video and Ian's power dissipation calculations, I think I can now pick a drive current which will meet my goals.  But if I'm wrong, well at least I didn't make a wild shot in the dark and I should be CLOSE to where I need to be.

Who sits at a red light with the brakes on for 10 minutes?

 

Nobody, but everyone sits in a traffic jam with the brakes on for 10 minutes at some point.

So the comparison won't give you any meaningful data

Except that most of my usage scenarios are a lot like a brake light.  On for a little while, off for a long time.  I'm asking about continuous current though because I may have instances where I need to do that, and it's the worst case scenario.  I can work backwards from there more easily than I can extrapolate forwards from a specific usage case.

[Starlord]

In the video I looked at we saw an experiment performed and the results from such an experiment are far more useful than paper calculations. For example, from that video the thermal resistance from PCB to ambient can be estimated as follows:

Input power: 1 A x 3.4 V = 3.4 W
If we assume a radiant efficiency of 20% the heat dissipation would be 3.4 x 0.8 = 2.7 W
At steady state the PCB temperature was estimated to be 260 degrees above ambient, so the thermal resistance PCB to ambient was 260/2.7 = 96 degF/W = 53 K/W.

In the experiment the PCB was in free air providing convective cooling. If the PCB were to be mounted on a plastic or wooden back plate inside a long tube there would be little free air circulation and the thermal resistance would be expected to be much higher.

But you can do similar experiments yourself with your exact set-up and get hard data. That will tell you much more about the constraints than we can tell you.

Cool , that information helps a lot.  As did that nice graph you put together.  Thanks a bunch!  What did you use to make that graph anyway?

[IanB]

I made the graph in Excel by plotting data points read from the video and doing a curve fit.

A heated object like that LED approximates a first order dynamic system, for which the equation is:

    (T - T_amb) / (T_max - T_amb) = 1 - exp(-t/RC)

I used the solver add-in to regress parameter values that best fit the data.

RC is the time constant of the system, which worked out at about 83 seconds.

Since R was found in an earlier post to be about 53 K/W we can calculate the value of C:

    C = 83 / 53 = 1.57 J/K

The heat capacity C could be used to help make predictions about small duty cycles and the temperature rise from transient heat loads.

[Corporate666]

Well of course I do, I have a target I'm aiming for.  But I'm listening to all the feedback, and I've learned a lot from the answers I've gotten to my questions.  I understand that nobody can give me a definitive answer without knowing exactly how the LED will be used.  But the information I've been provided with has helped me narrow down what might be realistic, and what is wildly unrealistic and which of my calculations were on the right track and which weren't.

So why not just say what the target is?  All of the numbers given are in such large ranges as to be meaningless.

The LED has to be "as bright as possible", but there is no clarity on what that means.  The LED will be pulsed, but it might be 100% on, and it won't be used continuously, except when it is.  It will be inside a housing made of wood, or maybe plastic, but sometimes exposed to the air.  It will be used indoors and outdoors.  Even wild guesses can't come from that requirements list.

The bottom line is I was only looking for a best guess, so I have a starting point.  I have a rough idea of how the LEDs will be used, and what their duty cycle will be, and how long they'll have to cool down between uses, and with the information from that video and Ian's power dissipation calculations, I think I can now pick a drive current which will meet my goals.  But if I'm wrong, well at least I didn't make a wild shot in the dark and I should be CLOSE to where I need to be.

The video tells you nothing.  It's a different LED, different thermal resistance, different substrate, driven to failure in free air.  It has no bearing whatsoever on a Cree XML on a star board in an enclosed housing - you can't even get within an order of magnitude with that sort of dataset.

Nobody, but everyone sits in a traffic jam with the brakes on for 10 minutes at some point.

I doubt there is a single person on this whole forum who has ever sat with the brakes on for 10 minutes.  You're either moving and there is a duty cycle involved, or you shift into park.  But that's neither here nor there in terms of LED current.

Except that most of my usage scenarios are a lot like a brake light.  On for a little while, off for a long time.  I'm asking about continuous current though because I may have instances where I need to do that, and it's the worst case scenario.  I can work backwards from there more easily than I can extrapolate forwards from a specific usage case.

There is no way to calculate what you're asking based on the information provided and be within even an order of magnitude of the right answer.  You need to buy the LED, construct your worst-case enclosure and measure the temperatures involved.  Anything else is like asking how much fuel it takes to drive between work and home, without knowing where each is, or what kind of vehicle, or time of year, or traffic, etc.  You just have to try it out and see where you end up.

[Starlord]

So why not just say what the target is?  All of the numbers given are in such large ranges as to be meaningless.

The LED has to be "as bright as possible", but there is no clarity on what that means.

It means I want to drive it as hard as I can without blowing it up.  I was using 1-2W LEDs previously, but they weren't RGB LEDs.  I'm trying to achieve an equivalent brightness to those.   If you want to know the precise lumens, well the superbright LED site has a lot of 1-2W LEDs that list what their lumens are, so you could make a guess from that.  But I have no idea what the lumens of the bulbs I used were, I bought them from Chinese vendors on Ebay.

The LED will be pulsed, but it might be 100% on, and it won't be used continuously, except when it is.  It will be inside a housing made of wood, or maybe plastic, but sometimes exposed to the air.  It will be used indoors and outdoors.  Even wild guesses can't come from that requirements list.

By that logic, you could never design an Arduino because you don't knows how it might be used.  How powerful a voltage regulator should it have?  How small does it need to be?  What clock rate should it run at?  What kind of USB and power connectors should you use?

Sometimes you don't have a clear set of parameters and you have to make a best guess at what typical usage might be and design for that.

And this is one of those times.  I want to use these LEDs in a variety of things.  I don't have a single usage case I can give you.  But as long as I understand what the limits are in free air, on a star board, with no heat sink, I can make a guess on my own based on a wide variety of potential uses what's safe to run it at and I can alter the design as needed in the future. 

The video tells you nothing.  It's a different LED, different thermal resistance, different substrate, driven to failure in free air.  It has no bearing whatsoever on a Cree XML on a star board in an enclosed housing - you can't even get within an order of magnitude with that sort of dataset.

Ian said one LED is not more efficient than another by any significant amount. I checked the datasheets and he's correct.  Therefore that video does tell me something.  It tells me that whatever LED I use, if it's mounted in free air, on a star board, that 1A is going to be way too much current for it.

Also, since all LEDs are about the same efficiency, the data Ian collected should be valid for other LEDs similarly mounted on a star board, in free air. 

And his data indicates how good the star board is at dissipating heat - in free air.

Now as for in an enclosure, I don't have data for that, but if I can calculate what the LED can do in free air then I know my current has to be below that.  By how much, I don't know.  50%? 75%?   But at least I have a starting point. 

There is no way to calculate what you're asking based on the information provided and be within even an order of magnitude of the right answer.  You need to buy the LED, construct your worst-case enclosure and measure the temperatures involved.  Anything else is like asking how much fuel it takes to drive between work and home, without knowing where each is, or what kind of vehicle, or time of year, or traffic, etc.  You just have to try it out and see where you end up.

I know 1A is too much.  And based on Ian's calculations, at 750mA the LED in the video would reach thermal equilibrium at 150C.
And if I want to limit the junction temperature to 100C as suggested by Cree, then the current limit would be 450mA.

And an order of magnitude below that is 45mA.  I'm fairly certain that the LED will survive in an enclosure at 45mA just fine.  Which means the maximum current is somewhere above that. 

And that means that I'm within an order of magnitude of the correct result.