Results 1 to 5 of 5
  1. #1
    mtbr member
    Reputation:
    Join Date
    Apr 2010
    Posts
    4

    Two Approaches for Effective LED Cooling

    Hi, In this post, I would like to share with you, two approaches giving better effective thermal management of power LEDs. I decided to post it in this part of the forum, because that could be as an introduction of eventual bike light project using SST-90, 2000+ lumen, and targeted more as a thrower. So, I will submit my intended initial target specs, in another thread. The specs and features may change and mature in interactive way, if hopefully, at least a portion of the popularity of the excellent quazzle's thread is achieved.
    For another project which is about to begin, probably involving SST-90W as well, I came across with some thermal issues, which I will try to observe in this small presentation.
    So using SST-90 already suggest good thermal design. In fact we've seen quite a few ways of LED mount in thermal point of view- using paste, grease, glue, tape, direct solder to copper, even copper pipe cap etc., and none of them satisfies, at least to me, the basic criteria for production, mainly concerning reliability, repeatability, QC, productivity. OK for modding, but not for production. At this point it looks like the obvious alternative is LED Star, and my answer is: NOT GOOD ENOUGH!
    Let's see what Luminus suggests:
    SST-90 W data sheet:
    PhlatLight SST-90-W devices are available on a star board for prototyping purposes
    And then:
    PhlatLight SSR-90 evaluation module consisting of a SST-90 surface mount
    device mounted on an aluminum star board.
    So, it is been stated from Luminus technical staff as well: They are for prototyping and evaluation.
    Let me put my 2c, describing two other solutions to achieve a better thermal resistance.
    Those two approaches will be compared with the thermal specifications given from Luminus in SST-90 and SST-50 data sheets.
    Both methods involve using copper as a medium base material between the component's (LED)
    thermal pad and the heatsink. And from the following descriptions, we will find out that the prime reason for using copper as a core, is not because of its superior thermal properties, but because of it's solderability. Even though heavier and more expensive, I think its particular implementation would be well worth and justified.
    The first method has been prototyped already and the results are excellent as expected.
    For example, applying 12A through the emitter, the spreading effect of the heat is immediate. Because of sufficient Al heatsing as well, looks like the LED will handle that overdriving continuously (but I am not advocating it). A few times 18A has been applied as well, but for not more than a few seconds, because of the rapid heat increase of driver MOSFET's.
    Note, that the following description is for an intended implementation, and it is slightly different from that seen in the pics from the protos - the copper baseplate is larger, using two PCB's instead of one etc.
    3.3mm thick piece of copper, is milled 0.75mm all the way, except rectangular area in the middle. In that island-like area which has the size of the thermal pad, and height of about 0.75mm above, the power LED will be solder reflowed.
    The other part of this assembly is single sided 0.7mm FR4 PCB. In the corresponding area in the middle, there will be rectangular cutout with sizes of about 0.1- 0.15mm larger than that of island-like copper area.
    After thin layer (or a few drops) of high temp glue is applied to the milled area of the cooper plate, the PCB is inserted through the slot to the main copper baseplate, and pressed against it. So we end up with very flat and even surface, with exposed copper base material in the center, surrounded by the PCB with tracks and pads, corresponding to the power LED used. That new single piece, which is in fact some kind of a MCPCB, or if we can call it a hybrid MCPCB, will be proceeded for solder reflow. The goal is to make that process compatible with automated machines as well. Manufacturer preference is to deal with a larger plate of copper of about A4 size, already CNCed - milled, drilled and perforated . FR4 PCBS are to be attached to the plate individually. Obviously, the size and shape of those individual hybrid MCPCB can vary- square, round, star etc.
    Once the emitters are solder reflowed, perforation will facilitate breaking them into pieces. In such a hybrid MCPCB, the LEDs thermal pad is directly solder reflowed to the exposed copper base and is like part of it. Because of the higher concentration of silver, the thermal properties of that joint are excellent - solder paste thermal conductivity (k) of about 60W/m-K is used. On top of that, the solder paste layer is extremely thin, because the LED is pressed by pick and place machine while the silver solder is in liquid state.

    In the prototype assembly, as I mentioned above, some things were done differently. Milling was done manually, on a hobby mill, as you can see the quality is not good. In fact, the image bellow shows the piece, which has been used in the factory for experimenting.

    This is the pcb's I made. Originally rectangular, but I cut the corners, providing more space for the mounting screws.

    It's been decided, instead of two pcb's, will be used one, with cutout in the center:


    And finaly, the assemblies solder reflowed







    The post is getting too big and boring, it is late here in SA, tomorrow will continue with some better staff.

    Nikola

  2. #2
    mtbr member
    Reputation:
    Join Date
    Apr 2010
    Posts
    4
    In this post an attempt will be made, in finding out what would be the benefits of using the described approach in the previous post -(Hybrid MCPCB), as an alternative of using a typical aluminum MCPCB.
    As practical examples, the data for SST-50-W and SST-90-W will be compared side by side with the calculated results for corresponding alternatives of Hybrid SST-50-W and Hybrid SST-90-W.

    In other words:
    Typical star MCPCB will be compared to Hybrid MCPCB with the same size area.

    Because the copper plate is thicker, the height of the Hybrid MCPCB is about 1.6 mm. higher. Having the same baseplate area sizes for comparison, will gain the same values for the thermal resistance base-to-heat sink -R(b-hs), which simplifies the comparison and targets the analysis towards the effects of the thermal resistance case-to-baseplate -R(c-b).


    SSR-50:

    The following data is from Luminus SST-50W data sheet, p10 on the bottom:


    Below, the equations are extended, just to show the thermal resistance case-to-baseplate R(c-b), and the thermal resistance baseplate-to-heat sink R(b-hs), as a separate values:

    R(j-c) = 2.45 C/W
    R(j-b) = 4.28 C/W = R(j-c) + R(c-b) = 2.45 + 1.83
    R(j-hs) = 4.39 C/W = R(j-c) + R(c-b) + R(b-hs) = 2.45 + 1.83 + 0.11

    Or, as add up components:

    R(j-c) = 2.45 C/W
    R(c-b) = 1.83 C/W
    R(b-hs) = 0.11 C/W
    -------------------------
    R(j-hs) = 4.39 C/W


    About R(c-b):

    I believe, the MCPCB used for SSR-50 and SSR-90 is Bergquist's Thermal Clad MP-06503, which dielectric thickness is 75 micron, and thermal conductivity (k) is 1.3W/m-K. If so, one's can calculate the thermal resistance through the dielectric layer alone, using the following formula:

    R= L/(kA)

    Substituting:
    dielectric thickness (L) of 75 micron,
    thermal conductivity (k) of 1.3W/m-K, and
    thermal emitting area (A) of 36.4mm^2,
    it yields
    R=1.58 C/W, (dielectric layer only)

    which is only 13% less, than R(c-b)=1.83 C/W, taken from Luminus measurements. That difference might due to some variations of the three variables of the above formula on the right. In fact, in this particular case concerning typical MCPCB technology, R(c-b) includes the solder reflow layer, copper layer, and the above calculated dielectric layer, but there is negligible contribution from the first two. With the calculated copper layer thermal resistance of Rcopper=0.0024 C/W (k=401W/m-K, A=36.4mm^2, L=35micron), it is not worth dealing with it at all.
    The not uniform solder layer thickness may have some variations, but for sure its distributed thickness is much thinner than the 35micron copper layer, and with its thermal conductivity of about 60 W/m-K, the contribution towards the total thermal resistance would be negligible as well.


    About R(b-hs):

    As from Luminus empirical data, R(b-hs) = 0.11 C/W, is a small contributor to the total thermal resistance.



    Hybrid MCPCB SST-50:

    Using the same pattern, Hybrid's MCPCB Thermal Resistance can be derived:

    R(j-c) = 2.45 C/W
    R(j-b) = 2.55 C/W = R(j-c) + R(c-b) = 2.45 + 0.10
    R(j-hs) = 2.66 C/W = R(j-c) + R(c-b) + R(b-hs) = 2.45 + 0.10 + 0.11

    Or, as add up components:

    R(j-c) = 2.45 C/W
    R(c-b) = 0.10 C/W
    R(b-hs) = 0.11 C/W
    -------------------------------
    R(j-hs) = 2.66 C/W


    About R(c-b):

    Contrary to the usual star MCPCB with three layers for R(c-b), the Hybrid MCPCB has only one silver solder layer. As it been mentioned above its thermal resistance is negligible. Even at some extremes, the calculated value is about 0.016 W/m-K. And that result is valid, if the layer thickness is 35 micron (as much as typical FR4 copper layer). Never mind that, I have placed a value more than 6 times higher than calculated one:

    R(c-b) = 0.10 C/W

    About R(b-hs):
    As it seen, the same value of R(b-hs) = 0.11 C/W, is used for the hybrid version as well. Assuming the above described Hybrid MCPCB, with the same area as the MCPCB of SSR-50, and using the same or similar thermal interface material, it can be concluded an equality between the two, as far as R(b-hs) is concern.
    So the only portion which makes difference for the thermal resistance, when comparing the usual MCPCB with Hybrid MCPCB, until the thermal path reaches the (aluminum) heat sink is R(c-b) the thermal resistance case-to-baseplate .


    Even in those circumstances, the thermal resistance, up to the heat sink is:

    R(j-hs) = 2.66 C/W
    Or, a massive difference of 1.73 C/W compared to usual star MCPCB (SSR-50).




    Now, shortly for SST-90:

    SSR-90:

    Typical Thermal Resistance (SSR-90 MCPCB, from SST-90-W data sheet) :

    R(j-c) = 0.64 C/W
    R(j-b) = 2.02 C/W = R(j-c) + R(c-b) = 0.64 + 1.38
    R(j-hs) = 2.15 C/W = R(j-c) + R(c-b) +R(b-hs) = 0.64 + 1.38 + 0.13

    Or, as add up components:

    R(j-c) = 0.64 C/W
    R(c-b) = 1.38 C/W
    R(b-hs)= 0.13C/W
    -------------------------------
    R(j-hs) = 2.15 C/W


    R(c-b) = 1.38 C/W - gained empirically
    R(c-b) = 1.26 C/W - derived from the formula

    In this case, 9% difference, because of the reasons already mentioned above.


    Then, one can notice this:
    R(b-hs) = 0.11 C/W for SSR-50
    R(b-hs) = 0.13 C/W for SSR-90
    Above values are empirical. If derived from the formula, the values should be the same. From real measurements however, that difference might due to some variations of the three components involved, as been mentioned above as well.



    Hybrid MCPCB SST-90

    R(j-c) = 0.64 C/W
    R(j-b) = 0.74 C/W = R(j-c) + R(c-b) = 0.64 + 0.1
    R(j-hs) = 0.87 C/W = R(j-c) + R(c-b) + R(b-hs) = 0.64 + 0.1 + 0.13

    Or, as add up components:

    R(j-c) = 0.64 C/W
    R(c-b) = 0.10 C/W
    R(b-hs)= 0.13C/W
    -------------------------------
    R(j-hs) = 0.87 C/W

    Or, a huge difference of 1.28 C/W compared to usual MCPCB (SSR-90).
    LEDs, dissipating heat of 25W for each case, will end up with 32C Junction Temperature difference.





    ***********************************************
    Some practical comparison examples using SST-50:

    Again, from above:

    R(j-hs) = 4.39 C/W for usual star MCPCB (SSR-50) data sheet
    R(j-hs) = 2.66 C/W for the hybrid version of SST-50 (Hybrid MCPCB)


    Case 1.

    R(h-a) < 1.73 C/W (or equal)

    If a Hybrid MCPCB is attached to a heat sink, with thermal resistance for natural convection environment of let's say 1.70 C/W, the total Thermal Resistance junction to ambient R(j-a) will be
    2.66+1.70=4.36.C/W. In other words, the total thermal resistance junction-to-ambient
    R(j-a)=4.36 C/W for the hybrid version, is already less than the thermal resistance of junction-to-heat sink R(j-hs)= 4.39 C/W for the star SSR-50. So practically and theoretically, achieving the Hybrid MCPCB total thermal resistance is not possible by any means, no matter how huge the heat sink for SSR-50 (star) will be chosen.

    In fact, a heat sink with 1.7 C/W thermal resistance is quite big, but it is still a practical example. From this Online calculator for natural convection heat sinks, that example corresponds to size of about 110x110mm, when 12W of heat is applied. The heat sink I am using now, for the test of SST-90W, has even slightly better thermal resistance. For bike lights heads, that value (1.70 C/W) is quite easily achievable as well, not to mention the air forced cooling factor there.

    Case 2.

    R(h-a) > 1.73 C/W

    Choosing a heat sink, with R(h-a) > 1.73 C/W , where the Hybrid MCPCB will be attached, makes it possible to choose another larger heat sink for the usual MCPCB (SSR-50), in order to match the same thermal resistance as the Hybrid MCPCB counterpart.

    Let's now choose a heat sink with R(h-a)=3.0 C/W for the Hybrid MCPCB. A matching heat sink for the SSR-50 must be with R(h-a)= 1.27 C/W .

    Total Junction-to-ambient resistances R(j-a) in [ C/W]
    2.66 + 3.0 = 5.66 for the Hybrid MCPCB,
    4.39 + 1.27 = 5.66 for the typical MCPCB (SSR-50)

    While the required thermal resistance on the typical MCPCB (SSR-50) version must be a little bit more than twice less then that of the Hybrid MCPCB, the difference in size is much more than that ratio.
    For example, heat sink with R(h-a)=3.0 C/W resembles the default one on the above mentioned online heat sink calculator (70mm x 70mm, 10 fins, heat source 12W),
    heat sink with R(h-a)= 1.27 C/W must be about 400mm long, in order to reach the total thermal resistance, if the same extrusion has to be used.


    Conclusions
    The positive effects of using alternatives for the typical MCPCB technology components, like Hybrid MCPCB, may have different interpretations. The comparison examples above cover only the case, when providing the same conditions for the power LEDs, affect the heat sink size.
    Another interpretation could be, how the same heat sink size, but used differently, will provide different thermal resistance, thus temperature rise, and LED junction temperature.
    The thermal management is prime consideration in LED designs and its importance is more than ever.
    Overheated LEDs are less efficient with immediate effect, and end up short lived in a long term.
    Keep them cool!



    Appendix

    From Bergquist's site :
    Many of the world's top LED manufacturers recommend Thermal Clad for use with their Power LEDs. Link to Avago, Cree, Lite-on, Lumex, Lumileds, Nichia, Osram and Seoul Semiconductor for additional information.
    That's why it will be adequate to make some thermal resistances calculations based on using Thermal Clad MP-06503. The same pattern, as used for SSR-50 and SSR-90, will be used calculating junction-to-heat sink thermal resistance R(j-hs):

    R(j-hs) = R(j-c) + R(c-b) + R(b-hs)

    R(j-c) - from corresponding data sheet document
    R(c-b) calculated, based on using Thermal Clad MP-06503
    R(b-hs) chosen arbitrary =0.1 C/W


    Cree:

    XP-G =19.5 C/W (6.0 + 13.4 + 0.1)
    XP-E = 24.1 C/W (9.0 + 15.0 + 0.1)
    XR-E = 9.6 C/W (8.0 + 1.5 + 0.1)
    MC-E = 7.2 C/W (3.0 + 4.1 + 0.1)


    Seoul Semiconductors:

    P7 = 4.2 C/W (3.0 + 1.1 + 0.1)


    Luminus:

    SSR-50-W = 4.39 C/W ( 2.45 + 1.83 + 0.11) all from data sheet document
    SSR-90-W = 2.15 C/W ( 0.64 + 1.38 + 0.13) all from data sheet document

    The bold value in the middle of the brackets is for R(c-b).

  3. #3
    mtbr member
    Reputation:
    Join Date
    Jul 2005
    Posts
    1,681
    So what's the benefit?

    For some reason, I think you'd get more light with less power using a 7-up setup or two of Quazzle's units.

    The SST90 is pretty inefficient isn't it?

  4. #4
    mtbr member
    Reputation:
    Join Date
    Apr 2010
    Posts
    4
    Quote Originally Posted by chelboed
    So what's the benefit?

    For some reason, I think you'd get more light with less power using a 7-up setup or two of Quazzle's units.

    The SST90 is pretty inefficient isn't it?

    Thanks for the reply chelboed, and yes, I may agree with you, but, this is not my point, and I am really sorry, that my presentation ended up not clear enough.

    My point rather is, how with a given LED emitter, using the described technique in the first post, can improve a LED application. For simplicity, I refer to that technique as a Hybrid MCPCB, because it is a hybrid of copper baseplate and FR4 PCB attached in certain way together and it can be done just as small, as a regular aluminum star MCPCB. In other words it could be a drop-in replacement.

    Really, this approach better suits applications, involving powerful LEDs, thus the examples in post2 are with SST-50 and SST-90, but can be applied using any of the popular LEDs at this time.

    The benefit?
    It might be multi directional, depending on the design target. For example, simple drop-in replacement of Star to Hybrid will achieve less temperature rise, lower junction temperature, which means higher luminous flux, longer usable live, combined with higher reliability. Applications, where high temperature rise is an issue (ex. high luminance flashlights), such an implementation would be a definite relieve.
    From other perspective, complying to certain specifications, can make the targeted unit smaller, because it would need less heat sink area. Or, the benefits could be any sort of combination of those mentioned.
    In the first post, I have tried to describe how to make it, so you can use it. That is it!

    Nikola

  5. #5
    mtbr member
    Reputation: Cat-man-do's Avatar
    Join Date
    May 2004
    Posts
    3,664
    Quote Originally Posted by kolyo
    Thanks for the reply chelboed, and yes, I may agree with you, but, this is not my point, and I am really sorry, that my presentation ended up not clear enough.

    My point rather is, how with a given LED emitter, using the described technique in the first post, can improve a LED application. For simplicity, I refer to that technique as a Hybrid MCPCB, because it is a hybrid of copper baseplate and FR4 PCB attached in certain way together and it can be done just as small, as a regular aluminum star MCPCB. In other words it could be a drop-in replacement.

    Really, this approach better suits applications, involving powerful LEDs, thus the examples in post2 are with SST-50 and SST-90, but can be applied using any of the popular LEDs at this time.

    The benefit?
    It might be multi directional, depending on the design target. For example, simple drop-in replacement of Star to Hybrid will achieve less temperature rise, lower junction temperature, which means higher luminous flux, longer usable live, combined with higher reliability. Applications, where high temperature rise is an issue (ex. high luminance flashlights), such an implementation would be a definite relieve.
    From other perspective, complying to certain specifications, can make the targeted unit smaller, because it would need less heat sink area. Or, the benefits could be any sort of combination of those mentioned.
    In the first post, I have tried to describe how to make it, so you can use it. That is it!

    Nikola
    Koylo, I think the problem here is that most of the content of your earlier posts were WAY over-the-head of most people who frequent this forum. Matter of fact it was a tad over MY head to tell the truth BUT I get the gist of what you are saying. Because I really like the SST-50 I made the attempt to follow your post even though I basically skipped over the math section.

    I think your idea of a better "star" platform for power LED's is a great idea. Anything that makes the heat transfer better to the heat sink gets thumbs-up and Kudo's from me. Your idea of a raised copper section ( on the star ) to help manage heat better sounds like a winner to me. Of course it makes no matter what I think unless the manufactures of the LED's recognize the same thing and make an effort to accomodate the issue. No doubt an up-dated version of the star platform using all the things you mentioned ( copper platform and free-flow soldering ) would raise the cost of the star/ LED.

    I would be super nice though if the manufactures did offer it as an option.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •