Homemade MCPCB's with questions about the thermal conductivity of solder.

[Mike S]

Hopefully this isn't in the wrong place. If it is, I apologize.

There's a graph online that shows the different values for various types of solder. When the unit is given in watts per meter per kelvin (W/m-K), is it better to have a solder with a higher or lower value? The solder paste I use is the lowest on the graph and thought this would be best when reflow soldering the LED's. Now I'm wondering if this may have been a poor choice.

Lately I've been buying bare emitters and making my own version of a MCPCB by using ultra thin copper clad boards and reflow soldering the LED directly to copper. The copper board is 0.13 mm thick before it's been etched. I usually remove the all the copper from the back side to make it as thin as possible. With the copper removed from one side, it's now only 0.09 mm thick.

Since there's a small void created by the PCB layer that the solder must fill, is the transfer of heat now much worse because of it, or is it negligible?

To keep the void as small as possble, a drinking straw is used to press down on the LED while it's being reflow soldered. The hole in the center of the straw fits perfectly over the dome so that it's only pressing down on the flat substrate. Different straws work for different LED's. For example, a coffee stir fits nicely over an XP-G.

I've done some side by side comparisons between MCPCB's with LED's already mounted and the homemade version using XM-L's, XP-G's, MC-E's, ML-B's and Rebel's. The homemade version heats up a little faster, but the copper is also thinner at 1.55 mm. The aluminum MCPCB's are between 1.75 mm and 1.95 mm. Once I use up the thin stuff, I plan to order something a little thicker and test again.

I believe it is working well, but wanted to get some feedback. I've used it on everything from 13 mm "stars" on up to 100 mm x 100 mm arrays which have the linear SMT drivers mounted on board and sharing the same heat sink. Sometimes the outer edges of the PCB will slightly lift up, so I found that instead of removing all the copper from the back side of the PCB, I'll leave an isolated copper trace along the entire edge and reflow solder it to the heat sink. That will hold it down nice and tight and make the whole thing more sturdy.

If this continues to work, I plan on adding an opaque layer of "solder mask" (I think that's what it is called) over the top of the PCB to seal in the traces and make it look a little more professional.

 

[Mattaus]

Looks pretty good to me. Could open up a whole new world with custom shapes and what not...

 

[Justin Case]

You want thermal conductivity to be high to "conduct" heat away.

Note that thermal resistance (units of K/W) is given by bond line thickness/(thermal conductivity * effective contact area). So if the application allows a thin bond line, that helps. If the application has a large effective contact area, that helps. If the application and budget allows use of a higher thermal conductivity material, that helps. Notice that thermal resistance depends linearly on each of these variables. Thus, there isn't any greater thermal resistance reduction benefit by focusing on one approach over another.

Your bond line thickness is 0.09mm. The effective contact area is the area of the thermal pad under your XP-G, which is 1.3mm x 3.3mm = 4.29 mm^2.

In your solder link, the low conductivity is 17 W/m-K. The high is 78. The "classic" value is 50 (63/37 eutectic lead solder). Even for the worst conductivity of 17, calculated thermal resistance is 0.001 K/W. If you drove your XP-G at full power (1.5A*3.5V =5.25W, of which perhaps 4W goes to waste heat), your temperature rise is going to be negligible. Basically, all of your waste heat coming out of the XM-L's thermal pad gets pulled away from the LED. The inherent thermal resistance of the XP-G itself from junction to solder point (6K/W) is your limiting factor.

The reason is that you have a thin bond line vs. a relatively huge effective contact area. You probably could use toothpaste for your thermal transfer agent.

So what you need to do is make sure that your MCPCB is well sinked so that it doesn't heat up. The heat is being pulled away from the LED and it has to go somewhere. Right now, it goes into your MCPCB and then into your pill.

 

[CKOD]

.09mm is ~3mil which is a pretty standard solder bondline iirc which means your thermal performance should be on part with a normal joint. 3 mil is also about as thin as a sheet of notebook paper, where did you find copper clad so thin? Ive seen "scissor cuttable" or something like that before, but I dont think it was *that* thin.

 

[lolzertank]

Here: http://parts.digikey.com/1/parts/11513-pcb-copper-clad-pos-12x18-2-side-pc93p.html
Wow, that stuff's expensive!

 

[Mike S]

Justin Case, thank you for that explanation. I read that W/m-K predicts the power loss through metal. My understanding was that the lower the number, the easier it is for the power to pass through. So what I need to do now is switch to a different solder that is more conductive since the size of the LED's thermal pad is fixed as well as the thickness of the PCB. I'd like it to be able to power an XM-L to 3A knowing the solder joint isn't weakest link the thermal path.

CKOD, I checked the thickness of some copy paper after reading your post and it's the same, like you said. (0.08 to 0.1 mm)

It looks like the copper board in lolzertank's link is for photoresist. I prefer that method, but not for that price! Electronics Goldmine sells the plain double sided stuff. It's $5 for a 12" x 12" sheet. The description says it's 0.1" (0.254 mm) thick, but that's not what the caliper is showing.