Why a good thermal path really matters

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
My how far we have come in just a few years. From top-O-the-line LEDs that put out 30 lumens to now into the thousands of lumens. From when the most powerful modern LED had a maximum drive rating of 350mA as compared to todays 13A+. Most importantly to the point of this thread is the current density has risen to almost 4Xs the original Luxeons.

Despite this massive increase in current density the industry still uses the same old mcpcb technology with little in the way of improvements. While more than sufficient for some tasks for others they fall woefully short. Running these new large die-high current LEDs-in flashlights that were never designed for them in the first place certainly falls in the latter category. Some of us crazy folk also like to overdrive our LEDs to eek out every last drop of light. :p

I am always amazed at how far people go to maximize the lumens making sure they have the latest bin and special drivers, AR coated glass, the best thermal compound they can find, resistance mods, and list goes on. All this effort and so many people miss the worst offender in their light. The mcpcb. This is the greatest bottleneck for heat getting out of most of these high power flashlights. You could have a giant chunk of dry ice as the last portion of your heatsink but if you have a saltine stuck in between your LED and the ice you wont be moving the heat too terribly fast. Remember that the closer you get to the LED die the more critical it is that there is a good thermal conductor.

What am going to be doing in this thread is using it as a repository for some of my thermal tests and showing some ways to improve the heat extraction. Hopefully you find it enlightening. The goal is to help more people put greater effort into the thermal side of things. I know it's not easy as there aren't really any proper off-the-shelf solutions so this will require some DIY work. It's worth it!

TheTests
________________________

The first one I want to show you is a test jtr1962 did of an XP-G R5 overlayed on my test of the same bin and type LED. The difference is how we managed the thermal side of things. Both of us have it mounted to a large heatsink to shed all that heat but as jtr1962 said about his setup it was a fairly lousy thermal path. Keep in mind that what I am doing here is in no way intended to take the shine off of his immense work in getting hard facts on whether these LEDs live up to there billing. His tests just provided me data to help me make a point for this thread.


So here we have my test of an XP-G R5 with thinned ceramic backer epoxied to solid copper in blue, my test of an XP-G R5 with a thinned ceramic backer epoxied to a heat-pipe in yellow, and jtr1962's in red.
6f8fd59f-1.jpg




So why the big disparity in the amount of current my XP-G could handle vs his? A superior thermal path. Some of you may remember me shaving down the backs of LEDs in the past. These new LEDs are so tiny that I had to come up with a new way to handle them. I put a mold release agent over the LED and then poured mold silicone over it(note this is not the stuff you get at the store. It has a catalyst that is added). After setting up I peeled it up off the glass with the LED being held as secure as you please allowing me to use a diamond coated disk in my dremel to shave the ceramic down. When done I gently pried the LED out of the silicone. This might also work with the HomeDepot variety of silicone as well but I can't be sure. You would want to keep the blob relatively small to avoid having a gummy center that will not dry. You might also be able to use something like a very thin layer of Vaseline as a release agent. That is not an official recommendation as I have not tried it to know for sure.





The shaved XP-G is on the right.

43178c6b.jpg




Here is the diamond coated disk I used. Should be able to find it at Lowes or Home Depot with all the other Dremel attachments. You want to tilt it ever so slightly and constantly change your direction of attack so that you wear the surface down evenly.

1ed2e66d.jpg




6fc8a9df.jpg






253161fc.jpg








The ceramic is now about .1mm thick.

34db4d82.jpg


After shaving the back I used Arctic Silver to bond it to a piece of copper 1/8th thick. It would be advisable to scrape the silicone off the top pads and tin before attaching to the copper base. Otherwise you will find it next to impossible to tin those tiny pads. You'll notice I added kapton tape to hold the wires in place. Without this the tiny solder points joining the wires to the LED would snap at the tinyest movement. You must have the wires fixed somehow before you solder.

a1d24077.jpg






Here is the simple little setup to get rid of the heat. Not pretty but it works.

99326cae.jpg






_______________________________

This next one is my test of an XL-M T6 soldered to a solid piece of copper. Some people have reported that there is not much in the way of lumens to be had past 3A. That may be true if you have a bad thermal path but as you can see there is quite a bit more past the maximum rating of 3A.

09d349f7.jpg






MCPCB tests_________________________


This section will contain information on the best and worst MCPCB performers. I will reserve my initial opinions until I can do proper testing on all my boards. I am waiting on a few to come in from across the ocean so the full test will not be posted until then. If you have some you want me to test PM me.



First off let's see why I don't care for MCPCBs. MCPCB stands for Metal Core Printed Circuit Board. Nothing wrong with the metal core part. It is the material used to create electrical insulation between the metal core and the top side traces that is the culprit. Here is a picture of a board for the XR-E bought from Cutter. The chip I pealed off is laid upside down. If you look close you will notice the glass fibers that make up the insulatory layer. The glass is part of a composite along with epoxy.

e1403905.jpg


Here is how thick I measured the laminated layer without the metal traces taken into account.

cd22e004.jpg




Here's a couple more that will be tested.

b59ab6da.jpg


__________________



More to come. Among other things I will also be testing LEDs on heat pipes.
 
Last edited:

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
Well this was one of the threads affected by the great database purge of 2011. Fortunately the info I had compiled was saved. I have updated the XP-G graph to reflect the additional test done on a heatpipe. A picture is worth a thousand words. It clearly demonstrates the higher lux that can be attained by improving the thermal path.

All future tests of MCPCBs will be standardized on the XP-G R5. Still waiting on some boards to arrive before I can finish that testing.
 

jtr1962

Flashaholic
Joined
Nov 22, 2003
Messages
7,506
Location
Flushing, NY
Interesting work. I had written a long comment before the database went down but unfortunately neglected to save it. The main point I made was that I realized the importance of a good thermal path when working with thermoelectrics. Even a slightly thick layer of thermal grease can measurably affect performance. In my prior post I had also calculated that the ceramic base accounted for approximately one third of the total 6° C/W die to thermal pad impedance. My guess is your superior results aren't so much because the ceramic is thinner, but due to the fact that you have a superior heat-spreading material in closer proximity to the die. Maybe Cree should attach the die to a piece of copper to spread the heat evenly over the entire thermal pad. On another note, these tests also show that optimal products aren't always commecially viable, and vice versa. Notice that at 1.5 amps or less your results aren't that much better than mine. Maybe Cree didn't bother optimizing the thermal path any more because lifetime at higher currents would be unacceptably short regardless due to higher current density. And thinner ceramic would be much more fragile. Remember that these LEDs are geared towards general lighting where we're aiming for 50,000 hours. A few hundred hours may be acceptable for flashlight use, but this is a small part of Cree's market.
 

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
Interesting work. I had written a long comment before the database went down but unfortunately neglected to save it. The main point I made was that I realized the importance of a good thermal path when working with thermoelectrics. Even a slightly thick layer of thermal grease can measurably affect performance. In my prior post I had also calculated that the ceramic base accounted for approximately one third of the total 6° C/W die to thermal pad impedance. My guess is your superior results aren't so much because the ceramic is thinner, but due to the fact that you have a superior heat-spreading material in closer proximity to the die. Maybe Cree should attach the die to a piece of copper to spread the heat evenly over the entire thermal pad. On another note, these tests also show that optimal products aren't always commecially viable, and vice versa. Notice that at 1.5 amps or less your results aren't that much better than mine. Maybe Cree didn't bother optimizing the thermal path any more because lifetime at higher currents would be unacceptably short regardless due to higher current density. And thinner ceramic would be much more fragile. Remember that these LEDs are geared towards general lighting where we're aiming for 50,000 hours. A few hundred hours may be acceptable for flashlight use, but this is a small part of Cree's market.

What a bummer to have lost your other post.:sigh: You always have something of substance to say. I agree that all the techniques used here may not be commercially viable. But what I share is intended for the overdriver crowd here at CPF. It should be noted however that Luminus(among others) has LEDs where the die is directly attached to solid copper so you can't say all of what I discuss here is not commercially viable.

I am also going to be doing a test without the thinning to see what effect it is really having.

Also, as to the point of there being no difference until past 1.5A it should be noted that these are far more ideal conditions than would be seen in most flashlights and that when heatsoak sets in the graph will lower and the split will move farther down the line. One of the tests I want to do is a hot test to see a more "real world" result as far as what people can expect with these in lights.
 

jtr1962

Flashaholic
Joined
Nov 22, 2003
Messages
7,506
Location
Flushing, NY
Yes, as current increases you need a better thermal path-hence attaching the die to solid copper as is done by Luminus. And I do hope Cree takes note here. A simple thing like having a heat spreader under the die might significantly decrease the thermal impedance of the XP-G.

I'm looking forward to your real-world tests. I'm sure the better thermal path will make a measurable difference once temperatures stabilize. The question is will it be enough to make it worthwhile. This all harkens back to my memories testing all sorts of different setups trying to get the most performance out of my thermoelectrics, obsessing over tenths of degree improvements.

Thanks a bunch for doing these tests! Besides the practical applications for anyone overdriving, it's great fun seeing how far these LEDs can be pushed. I still remember the old days when we were afraid to push a Luxeon past 500 mA. Now we have single LEDs rivaling the output of a 100 watt incandescent.
 

doctaq

Newly Enlightened
Joined
Feb 15, 2010
Messages
174
when you thin the ceramic back does the back become eletrically neutral? can it still be directly soldered to a metal surface?
 

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
when you thin the ceramic back does the back become eletrically neutral? can it still be directly soldered to a metal surface?
Thinning the back side naturally removes the electrical contact points but you still have the gold vias that could potentially short out if you used an electrically conductive epoxy. So don't;) Since the metal is removed off the back there is also no way to solder it as solder does not bind with ceramic. I suppose, and I have plans for this, that you could go all the way through the ceramic and leave the back metalized portion of the LED and solder that. That would most certainly not be electrically neutral though and there is a very large chance of error in doing the process.
 

ma_sha1

Flashlight Enthusiast
Joined
Jan 10, 2009
Messages
3,042
Location
CT, USA
Nice testing! Love to see data like this!

DX just came out with XML-T6 on what looks like a 20mm Copper star board: SKU 54704,
may be worth a test?

My order just got shipped, I am curious to see if I could drive it to 6 Amp,
if I mounted it to a Copper heat sink.
 

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
Nice testing! Love to see data like this!

DX just came out with XML-T6 on what looks like a 20mm Copper star board: SKU 54704,
may be worth a test?

My order just got shipped, I am curious to see if I could drive it to 6 Amp,
if I mounted it to a Copper heat sink.
I ordered that copper core board for testing. Just waiting for it to arrive.
 

Walterk

Enlightened
Joined
Jan 21, 2010
Messages
755
Location
Netherlands
Got mine today. They are about 0.8mm thick. Will benefit from some polishing on the underside but star really is red copper.
(Waiting for adjustable driver before I will put it to test compared to alu star.)
 

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
Got mine today. They are about 0.8mm thick. Will benefit from some polishing on the underside but star really is red copper.
(Waiting for adjustable driver before I will put it to test compared to alu star.)
Good to hear they are arriving in people's hands. I ordered mine some time back. Getting tired of waiting but I suppose that's the nature of the beast. Can't wait to hear how they work for you.
 

shao.fu.tzer

Flashlight Enthusiast
Joined
Jul 13, 2006
Messages
1,076
Location
P-Town, TX
Coming from an overclocking background before I discovered the "new high powered LED lights", I had always know the importance of thermal conductivity in your setup. I would go so far as to lap the bottom of my (ususally solid copper) heatsinks to a mirror finish, and then use a credit card to apply Artic Silver in a ultrafine layer between the sink and the CPU die. The best heatsink I ever owned was a giant hunk of copper with a ton of aluminum rods protruding out in a fan like pattern, but very densely packed. I have tons of old CPU and chipset heatsinks I often think about incorporating into larger builds. ...oh yeah and to all the people putting artic silver in their P60 hosts, be careful. It does contain silver and you can experience a short if you get it in the wrong place. I like to use Arctic Ceramique between the threads of the reflector and pill on drop ins, but be careful, just a dab will do you.
 

Th232

Flashlight Enthusiast
Joined
Dec 25, 2008
Messages
1,064
Location
Sydney, Australia
I think JTR is talking about the Cxx-xx series of LEDs, like the CST-90, which is a die sitting on solid copper.
 

Curt R

Enlightened
Joined
Mar 22, 2009
Messages
486
Thermal conduction from a manufacturers point of view.

We never thought much of the MCPCB that everyone used with the K2 and P4/P7
LEDs. You do get some spreading of the heat to the thermal conductor of the
flashlight but the ruggedness of the LEDs themselves just didn't seem to require
the cost and design limitations imposed on the flashlight as opposed to a well
designed thin fiberglass board. The XM-L and the SST-50/90 introduce a whole
new set of requirements as related to their concentrated almost point like source
of generated heat from higher drive power levels.

From a manufacturers' point of view we require good thermal to the outside air and
at the same time keep cost down. At first we wanted to use the heat pipe concept.
Using samples made from C12200 copper tube was not cost effective vs solid C14500.
We silver solder the C14500 directly to the back of the LEDs heat pad. Then this is
inserted into the aluminum thermal heat sink part of the head with the XM-L or with
SST-50/90 into a larger C14500 heat sink and then into the aluminum thermal spreader.
All of these are interference fits done by heating the receptor of each operation. This
gives us a direct mechanical thermal path from the LED to the air. We call it DTT,
or Direct Thermal Transfer.

The problem with the MCPCB is that each layer not only acts as a thermal barrier but
that each surface also reflects a small amount of heat back. And as you look at the
thermal radiation pattern of each layer there is a shallow spherical segment of decreasing
heat like the layers of an onion as the distance increased from the LED. Each layer has
it's own onion like effect. A layer is LED to MCPCB. MCPCB to heat sink. Heat sink to head.
Flashlight head to air. As the entire unit become heat saturated the point source area of heat
directly under the LED increases in temperature.

With the DTT method the LED generated heat is routed down into the head and then radiated
outward to the cooling fins of the head in a circular cone pattern and not that of the shallow
spherical segment pattern. This also helps to decrease the build up of the point source area
directly under the LED as the head becomes thermally saturated.

I am not aware of any other manufacturer using this type of method as it is more costly than
just bolting down a MCPCB. As those of you making your own flashlights, this is one method
that you might consider. The circuit board that we use is designed only for electrical connections.

Curt
 

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
I am not aware of any other manufacturer using this type of method as it is more costly than
just bolting down a MCPCB. As those of you making your own flashlights, this is one method
that you might consider. The circuit board that we use is designed only for electrical connections.

Curt
I am also unaware of any others that use an interference fit. It is a very good design as long as you never ever need to change out that LED.:eek:
 

vestureofblood

Flashlight Enthusiast
Joined
Jun 19, 2008
Messages
3,210
Location
Missouri
Saabluster,

I really appreciate you putting this thread together. I'm a big fan of these type of experiments. Having all the equipment to document the results accurately is icing on the cake.

Just to be clear are you saying that in a side by side test with lets say an XML or SST that grinding down the base material and epoxying it to copper would defeat soldering the stock LED directly to the same copper bar?
 

saabluster

Flashlight Enthusiast
Joined
Oct 31, 2006
Messages
3,736
Location
Garland Tx
Saabluster,

I really appreciate you putting this thread together. I'm a big fan of these type of experiments. Having all the equipment to document the results accurately is icing on the cake.

Just to be clear are you saying that in a side by side test with lets say an XML or SST that grinding down the base material and epoxying it to copper would defeat soldering the stock LED directly to the same copper bar?
Glad you like it. We all know that better materials helps heat transfer better and keep the lumens up but sometimes seeing it in clear graphs can really drive the point home.
I really have not done an ablated backer vs soldered test in this thread just yet. That is coming though. I am just waiting to get a high precision setup first. I have so far just been doing simple relative measurements. The new sphere I'm building will allow me to make absolute measurements and that will be nice. I just bought an Extech 407026 to use for testing as I believe it will be more accurate than some of the more common ones around here since you can choose your lighting type. The sphere will also be coated on the inside with barium sulfate and the sphere input will be homogenized. I can't wait to get this all set up and running.

I just got my DX copper XM-L today as well. Was thinner than I expected it to be. Can't wait to test it.
 

VegasF6

Flashlight Enthusiast
Joined
Dec 5, 2007
Messages
1,449
Location
Las Vegas
That DX copper XM-L was heavier than I expected though. It was a nice surprise. Could have been smoother.

FYI I measure mine at .035 inches or about .89 mm thick.
 
Last edited:
Top