Possible LED attachment improvements

Flashlight Modding and DIY Parts
1

This thread is split-out from How to reflow solder an LED emitter on a PCB or MCPCB., reorganized and updated with new information.

1. What can be improved
There are 2 sources of lumen-loss that can be reduced by improved LED attachment:

  • thermal damage during reflow
  • thermal resistance

We have not made any quantification of the losses and improvements. All that is available at the time of writing is theoretical calculations based on incomplete data and measurements done on LEDs we don’t use. Therefore for now all improvements are purely hypothetical.

Thermal damage is hard to quantify. The only hard info I found is here:
https://www.nichia.co.jp/specification/products/led/ApplicationNote_SE-AP00032-E.pdf
It shows that 260°C reflow for 10 seconds reduces output of Nichia NS3W183 by 10. Doing it twice is a 12 reduction.
Normally we heat up the LED twice, once during reflow and again (though to lower temperature) when attaching leads.
I’ve seen somewhere that flip-chip LEDs are less susceptible to being damaged like that.
I’ve seen no manufacturer suggesting ways of improving the process to reduce peak temps. I don’t know why. I suspect that normally the effect is less pronounced than with NS3W183.
Reducing thermal damage should improve LED output for all currents all the way from moonlight to turbo.

As to thermal resistance - some LEDs have very small thermal pads for the wattage they push through. The worst affected is XHP35 HI. Osram Oslon Black is quite bad too.
For them even a very thin layer of solder can lead to real increase in die temperature.
Reducing thermal resistance would improve performance at high-turbo modes without notably affecting lower ones.

2. How to reduce thermal damage
There are 4 ways really:

  • improve temperature control (some of us do it already, some don’t)
  • use a low-temperature solder (note: our regular leaded solder is already better than unleaded SAC305 which most manufacturers recommend. By 37 °C)
  • use a conductive glue (note: glues often need high-temperature curing….but not always)
  • avoid heating twice

It should be noted that most manufacturers specify maximum junction temp of their LEDs of very roughly 150 °C. There is no point of getting lower than that.

3. How to reduce thermal resistance
Use a better thermal interface material.
It can be:

  • glue
  • solder
  • thermal paste
  • some other less popular materials

You can also reduce bond thickness. What is the regular thickness?
I don’t know.
Manufacturers say they recommend choosing stencils which lead to 70-100 µm joint.
We don’t use stencils, which reduces our control over paste application. We compensate by applying pressure during reflow to remove excess solder. It may actually be better than the regular process, but I don’t know…

4. Separation of concerns
Solder between LED and PCB has to perform 3 functions:

  • conduct electricity (on electrical pads)
  • conduct heat (on thermal pad)
  • keep surfaces together (anyhow)

The functions are really independent and can be performed by separate substances.

5. Baseline material: Sn63Pb37

  • 183 °C melting temperature
  • 50 W/mK conductivity

6. Possible process improvement 1: good temperature control
Some put more or less effort into that already. You should keep your reflow process as close to the reference as possible:

  • don’t exceed peak temps
  • don’t keep heated-up for longer than needed
  • don’t heat up too fast
  • don’t cool down too slow

7. Possible process improvement 2: better application of pressure
Now we tend to tap LEDs with our fingers to remove excess solder. This is very inconsistent and introduces the risk of shifting LEDs. Some device applying pressure vertically could allow us to increase pressure and therefore reduce bond thickness.

7. Possible process improvement 3: heat up once
One could use a suitable conductive adhesive to connect the lead wires. Do note that some glues are not resistant to high temperatures. Alternatively one could use Sn42Bi58 solder - 138 °C should be high enough not to unsolder in regular operation (right?) and low enough to not cause damage.

8. Possible material improvement 1: Indium
Indium is a low-temperature solder with extraordinary thermal conductivity as well as good melting temperature.

  • thermal conductivity 86 W/mK
  • 156.7 °C melting temp

It is not available as solder paste. However:

  • it is available as a very expensive wire
  • it is available as inexpensive scraps. Are they really pure though?

Also, it’s 26 times weaker than SnPb. It may need some other method of keeping LED and PCB together.
I thing epoxy around the LED die would work well enough. Some clamp would be even better, though harder to implement.

9. Possible material improvement 2: glue + paste
Conductonaut has thermal conductivity of 73 W/mK.
and obviously doesn’t require heating the LED at all.
However:

  • it doesn’t glue surfaces together
  • can’t withstand > 140 °C, limiting the choice of adhesives
  • is there a risk of pump-out?

There are quite a few adhesives which can supplement it for electrical pads.
There is one outstanding though. Cotronics Duralco 120

  • electrical resistance low-enough to be meaningless
  • large range of working temperatures
  • cures at room temperature

I wouldn’t be surprised to find more suitable conductive glues…

10. Possible material improvement 3: UNIMEC H9890-6A glue
Years ago Saabluster used an exotic glue for XR-Es. Now there’s a really interesting glue from the same manufacturer: UNIMEC H9890-6A

  • 140 W/mK
  • 200 °C curing for 1 hour
  • very electrically conductive
  • bond thickness as thin as 10 µm

OK, curing temp may be higher than ideal. Or maybe it doesn’t matter?

11. Other thoughts

What does it take to make optical contact bond?

2

Regarding thermal resistance: here is a formula for the thermal resistance of, say, the solder joint for the thermal pad.
R=(thickness)/(area*conductivity)

For an XHP35 the thermal pad is 4.3 mm^2. If the solder has 50 W/mK and is 50 microns thick the thermal resistance is 0.23 K/W. So there is room for improvement, but don’t expect a large improvement because this is already small compared to the LED thermal resistance of 1.8 K/W.

3

Thank you EasyB, you reminded me that I failed to include calculations. I’ll do that now.
According to The Driver, XHP35 HI has thermal pad of 4.11 mm²
I’ll use this number as I suppose it comes from polarity indicator. It’s not a big deal…

I wonder why did you use 50 µm? I have seen no data on what would be joint thickness and used to use the smallest recommended stencil-produced 70 µm. At this point I’ll use both numbers, I have no idea where lies the truth.

Let’s calculate temperature increase on the solder joint when driving XHP35 to its peak (note: peak with SnPb solder; a better joint should move the peak slightly).

With 50 µm bond thickness the temp difference would be:
49W / 4.11 mm² / 50W/mK * 50 µm = 11.9 K difference.
With 70 µm bond thickness that would be 16.7 K difference.

That’s assuming there are no voids in the joint. This is unrealistic. Let’s use 10% voiding and assume that thermal resistance scales linearly with voids (it’s probably worse than that). So the real differences would be 13.1 and 18.4 K respectively.

According to PCT, dropping temperature by 10 K (150°C->140°C) improves flux by 2.5% at maximum current that they specify. Let’s assume that improvements that we make are linear to this gain.

With 50 µm bond we can gain:
table(table#posts).
|Material|ΔT between LED and PCB|ΔT improvement over SnPb|output improvement|
|Sn63Pb37|13.1|0|0%|
|Indium|7.6|5.5|1.4%|
|Conductonaut|9|4.1|1%|
|H9890-6A|4.7|8.4|2.1%|

With 70 µm bond we can gain:
table(table#posts).
|Material|ΔT between LED and PCB|ΔT improvement over SnPb|output improvement|
|Sn63Pb37|18.4|0|0%|
|Indium|10.7|7.7|1.9%|
|Conductonaut|12.6|5.8|1.5%|
|H9890-6A|6.6|11.8|3%|

Theoretically, a 10 µm bond of 9890-6A would bring improvement of 3.1% to 50 µm SnPb and 4.4% to 70 µm SnPb.

4

Very interesting, thanks for sharing! :beer:

5

I am interested in continueing the discussion here. From a practical perspective I find the Indium solder much more interesting. The US company Indium offers several types of solder wire containing Indium, this is presumeably what Agro also found. Yes, it’s quite expensive, but a small amount would suffice for most people. The problem here is that only a very high Indium content actually allows for a better thermal conductivity compared to standard solder types (see values in datasheet ).

I have found a “cheap” source of Indium wire here in Germany. I guess you would need to add flux to solder with it.

Concerning voiding:
One property of Indium is that it fills voids. That is why it is used for sealing things.

Here’s a nice article on Indium solder.

I wonder if the reduced mechanical strength of the solders bonds compared to normal solder bonds really makes a difference when solderding LEDs. They really don’t weigh much.

6

Yes, purity is needed.

You can find a bit cheaper fluxless wire in China. Example:
https://www.aliexpress.com/item/High-purity-indium-wire-plate-block-1-0mm-1-5mm-3-0mm-experimental-research-special-line/32821335021.html
It may or may not be exactly the same as what you found.

7

A very relevant CPF thread on this topic from 2012: Heat Management Issues

He basically had the same ideas as you and Enderman have here: Indium solder and liquid metal instead of thermal paste.

8

I have done some calculations with Excel:

So generally it seems that spending a lot of money on Indium is hardly worth it, but since we have found rather cheap sources, it does seem to make a little sense (because of the theoretical creeping effect in addition to the tiny temp difference). If you combine it with cherry picking LEDs, soldering under pressure and using liquid metal thermal paste under pressure, you might get noticeable gains. Cherry picking LEDs will probably have by far the biggest effect! If I were to reccomend one thing, that would be it.

9

Interesting 2010 paper from Luminus:
http://www.nbetech.com/pdfs/Application_supplemental_3.pdf
They attach their dies with sintered silver.
An interesting note is that their process requires temperatures of ~260 °C. Probably for no less than an hour. Therefore such temps are not problems for LED dies. Is it for silicone? Is it for phosphor?

Also, saabluster used Diemat DM6030 with XR-E.
This glue requires curing at 225 °C for a quarter + 200 °C for 30 minutes + 175 °C for 45 minutes.
saabluster knows his stuff and if heating up like that reduced LED performance - he would probably find out.
So it seems that at least XR-E is resilient to prolonged heatup to 225 °C.
I’m somewhat puzzled as Nichia’s 260 °C is quite close.

Though if 200 °C is not a problem, pressureless sintering looks like the best way.
In this paper you can see a nice list of makers and users of this tech. It’s quite extensive.

10

That Luminus Paper is indeed interesting, thanks!

Saabluster certainly tried a lot of things to improve heatsinking and luminance if his LEDs.

11

OK, some details found on low temp pressureless sintering so far:
Namics XH9889-2 offers 300 W/mK. With curing at 200 degrees. That looks like the top choice available.
But another interesting options is Nihon Hanada MAX 112. Cures at just 150 °C, that’s certainly safe. 182 W/mK ain’t bad.

But something tells me that this indium wire might be cheaper. :wink:

12

I was just thinking - why not just use SnAg solder paste (or wire)? Sn96.5Ag3.5 supposedly has a thermal conductivity of 78W/mK and a low melting temp of 220°C. It's important to differenciate between SnAg and SnAgCu solder. The latter has a lower thermal conductivity of around 60.

Some solder wire products I found:

Felder ISO-Core VA Sn96,5Ag3,5 (Sn96Ag4 according to DIN EN ISO 9453) (221°C)

Furutech S-070 SN96,0Ag4,0 High Performance Solder (220°C)

WBT WBT-08x5 SN96,0 Au4,0Silver Solder (216/219°C)

Solder paste:

Edsyn CR 88 Sn96.5Ag3.5

I do wonder though if the "negative" effect of adding a little bit of copper can be negated by having a much larger percentage of silver as in

Mundorf MSolder SUPREME SilberGold Sn88,6Cu1,8Ag9,5Au0,1 (290°C)

Shouldn't this be way better than the others?

One thing to note when using solder wire with LEDs: one needs to put flux onto the contacts of the LED because the flux in the solder wire is probably gone when you put the LED onto the tinned pcb contacts.

I just noticed that I have discussed this topic before a few years ago (here).

13

Solder Alloys: Physical and Mechanical Properties shows Sn96.5Ag3.5 to have 55 W/mK.
https://www.ametek-ecp.com/-/media/ametek-ecp/v2/files/cw_datasheets_sds_cfsi/datasheets/sn96ag4%20data%20sheet.pdf?la=en shows Sn96Ag4 to have 53 W/mK.
https://qualitek.com/nc600_SnAg_tds.pdf has Sn96Ag4 at 55.3.

14

Ok, you convinced me. Maybe the number I found is a typo and they meant 58.

15

Did you find any further methods of improving the thermal resistance of led flashlights?

I plan to implemt some of your/our findings in my big thrower when the LED is replaced the next time.

16

Nope, I found nothing better for this kind of stuff.
There may be some improvements for lights that have bottlenecks down the path (f.e. DQG Tiny 18650), but I can’t add anything to improve on what’s said here.

17

It’s probably a bit related. Just got an IC Graphite thermal pad kindly shipped to me by zak.wilson. Here’s a quick test comparing it to the MX-4 thermal paste. No TIM for reference. It seems to suck heat off the MCPCB just as well as normal paste. It’s reusable too, and no cleaning required so perfect for LED testing.

18

Thanks for the test! That is indeed very interesting for lights which are modded vers often.
Could you repeat the test, but measure the brightness of the LED?

19

OK, so I got some Indium off AE (advertised as 99.99% pure….) and tried soldering it. While it’s very very soft, about as soft as pure lead, joint strength was surprisingly good. Two strips of FR4 PCB soldered together in an area of ~1cm² wouldn’t come apart until the FR4 itself was severely deformed. Wetting, as expected, was very good (I used some standard rosin flux).

I found one possible downside to pure indium though:

This could be an issue with ceramic packages, though I doubt it applies to any but the most expensive sources.

(This paper also somewhat relieved my concerns about creep failure though I’d take extra precaution not to stress the emitter via the centering ring, just to be sure.)

20

Nice! The strength shouldn’t be a problem because you can use normal solder for the plus and minus connections.

I have also been researching this quite a bit and will post something here soon.

Optimizing the solder connection is the most fundamental thing we can improve with a given LED. I don’t know why nobody started doing this earlier.