EEVblog #744 - SMD Thermal Case Design - µSupply Part 15

[EEVblog]

Dave explains how to attach an SMD power transistor or regulator to a case to use as a heat sink in this design tutorial. And in the process talks about thermal design, the electrical/thermal analogy, and thermal vias.

This is Part 15 of the µSupply Power supply design series. Other videos are here:
https://www.youtube.com/playlist?list=PLBF35875F73B5C9B5

SilPad datasheet: http://www.bergquistcompany.com/pdfs/dataSheets/PDS_SP_A2000_12.08_E.pdf
PowerPeg thermal PCB pin/turret: http://tem-products.com/index.php/powpeg.html/
Saturn PCB Toolkit calculator: http://www.saturnpcb.com/pcb_toolkit.htm

[ColdKeyboard]

You either die a hero or you live long enough to start making crappy videos... :\

[vlad777]

Which is better : sheet mica or sil-pad?

[German_EE]

After seeing the power pegs in a mailbag I did some experiments into alternatives. A brass flat head screw driven into a countersunk hole in the PCB provides very good heat transfer, especially if the screw is used to attach a heatsink to the other side of the PCB. By clamping the semiconductor to the PCB then soldering the leads and tab pressure on the screw head will be maintained.

[T3sl4co1l]

Which is better : sheet mica or sil-pad?

Mica by a little bit IIRC, but it's a big hassle to use (has to be greased both sides).

There are also ceramic and metal products.  BeO being the classic "too good to be true" (too good to be non-toxic, more specifically), AlN and AlNO being much more popular (nearly as good performing, cheaper, non-toxic), and hard-anodize aluminum metal (very good, what with the metal core; and the advantage that it can be grounded inbetween for shielding!).  The main disadvantage being, these are all *very* rigid, very flat materials, so you need a great fit to begin with, an uneven surface may lead to fracture, the thickness wastes a lot of thermal resistance, and you still need double grease.

Kapton/polyimide is terrifically bad as a bulk material, but because it can be made in much thinner sheets than mica (which is also a bad thermal conductor, for similar reasons really), and has excellent dielectric properties, it's competitive with the others.

Rubber products are the best when you need "throw the stuff together and forget about it" simplicity.  They're generally on the low end (and the "eyed with suspicion" end) of TIMs (thermal interface materials), but they're great because of ease.

Another good way to get heat out of a board is to use a single side layout, and goop the board to the (aluminum) enclosure with one of those sticky type rubber pads (Gap-Pad seems to be the more popular trademark).  Or you can still put components on the back side, as long as the pad is tacky enough to flow between smaller components (I wouldn't suggest using anything bigger than 1206 chip capacitors if you need to do this).

If you have a thin pad (10 or 20 mil) and piles of thermal vias, you can stand to get a pretty fair power density this way (maybe up to 5 or 10W per D[2]PAK?).  Of course, you could make a sandwich pack with rubber pad and metal plate on the top side as well, which would greatly increase the heat dissipation of resistors and other low profile SMT parts.

For power applications, your best choice is to just use more parts in parallel.  Count on less than 50W per TO-220 or 100W per TO-247, and that's with a greased interface, no insulator.  Halve it if you need an insulator.  If that's not good enough, consider using a heat spreader, with a much larger insulator pad to keep the resistance low.

Tim

[rs20]

Maybe I'm missing something but where are we supposed to be sourcing these "thermal bars/spreaders" from? I realise they're pretty much just aluminium stock cut to length (although sending that contract out would get you some very crappy surface quality), but even getting this stuff simply cut to length seems like another contractor to wrangle. Not to mention getting tapped holes. Or am I missing a simpler source for these parts?

[T3sl4co1l]

You don't have a hacksaw, drill and tap?  Critical bench supplies!

No need for a huge machine shop and CNC centers!  A drill press is pretty damn handy (for straight holes to tap, and for PCB drills if you're working much FR4/G10 as well), but quite cheap as machines go, and well worth it.

Tim

[Psi]

Wouldn't more small vias be better than fewer larger ones?

The inner wall copper plated thickness would be ~the same, so it's just the surface area through the hole.
I would expect a larger hole to be worse because of the large wasted area that was drilled out (when compared with lots of smaller holes over the same area)

Imagine a large circle, now imagine one the same size but made of many smaller circles in an 360deg arc.
The total circumference of the smaller circles added up would be more than the single larger circle circumference.
And that isn't including all the extra small circles you could fit in the center.

So wouldn't you be better to have lots of 0.4mm holes, or as small/many as you can get away with before the PCB house charges you more.

[casper.bang]

Yay, finally progress on the uSupply... good on you dave!  Remember there's a non-insignificant crowd ready to throw money at you when/if this is ever finished. *Insert take-my-money smiley here*

[rs20]

You don't have a hacksaw, drill and tap?  Critical bench supplies! No need for a huge machine shop and CNC centers!  A drill press is pretty damn handy (for straight holes to tap, and for PCB drills if you're working much FR4/G10 as well), but quite cheap as machines go, and well worth it.

I hear what you're saying, and I do have a hacksaw, drill and tap (albeit only UNC 1/4"-20tpi for tripod mount threads, but hey) (and not that anyone cares, but I don't have space for a drill press but I do have a plunge router which... isn't really the same thing but I digress). I'm not sure how to get a consistent measured thickness, decent surface finish, apart from lucking out on getting the right stock to begin with, but I concede that this is just ignorance on my part. What I really don't understand is how to do this in moderate or large quantities -- I mean, he's talking about avoiding double side load for the reason of extra costs on the production line, so how do you get these bars in that sort of quantity for reasonable cost?

[EEVblog]

Another good way to get heat out of a board is to use a single side layout, and goop the board to the (aluminum) enclosure with one of those sticky type rubber pads (Gap-Pad seems to be the more popular trademark).

That stuff is ok for packages that only need to get a relatively small amount of heat out overall. But IME not so great for power packages that need real heatsinking.
I have some of this stuff lying around the lab somewhere... cost a fortune IIRC.

[senso]

You don't have a hacksaw, drill and tap?  Critical bench supplies! No need for a huge machine shop and CNC centers!  A drill press is pretty damn handy (for straight holes to tap, and for PCB drills if you're working much FR4/G10 as well), but quite cheap as machines go, and well worth it.

I hear what you're saying, and I do have a hacksaw, drill and tap (albeit only UNC 1/4"-20tpi for tripod mount threads, but hey) (and not that anyone cares, but I don't have space for a drill press but I do have a plunge router which... isn't really the same thing but I digress). I'm not sure how to get a consistent measured thickness, decent surface finish, apart from lucking out on getting the right stock to begin with, but I concede that this is just ignorance on my part. What I really don't understand is how to do this in moderate or large quantities -- I mean, he's talking about avoiding double side load for the reason of extra costs on the production line, so how do you get these bars in that sort of quantity for reasonable cost?

You can buy aluminium plates in 1mm increments in thickness,  and square/rectangular solid extrusions in increments of 2-5mm, its a matter of playing with the required thickness's required, if you choose to use a plate is cheap and fast to cut it, a bit wasteful but not that bad, and remember, thermal pads can have more than 2mm in thickness, they are squishy so you don't need to even have 0.1mm accuracy in all those parts.

And to find cheaper thermal pads, buy stuff for laptops, they use thermal pads by the metric ton, just avoid the crappy blocks of rubber from eBay that are sold as thermal pads.

You can get high thermal conductivity stuff from Fujipoly, and 3M as a huge selection, from thin to thick, more or less squishy, with glue on none, one or both sides, sold by the roll, also cheap!

[T3sl4co1l]

I hear what you're saying, and I do have a hacksaw, drill and tap (albeit only UNC 1/4"-20tpi for tripod mount threads, but hey) (and not that anyone cares, but I don't have space for a drill press but I do have a plunge router which... isn't really the same thing but I digress). I'm not sure how to get a consistent measured thickness, decent surface finish, apart from lucking out on getting the right stock to begin with, but I concede that this is just ignorance on my part. What I really don't understand is how to do this in moderate or large quantities -- I mean, he's talking about avoiding double side load for the reason of extra costs on the production line, so how do you get these bars in that sort of quantity for reasonable cost?

If it's just a few, cut slightly oversize, then use a file (or sander or grinder or..) to get the desired finish.  Using a hand file is an ancient and sorely under-appreciated task.

If you wanted to make a hundred of a part, it would be worth making up a drawing and handing it off to a machine shop.  Let them figure it out.  As long as your drawing is correct and reasonable, they have to redo it until it meets spec.  You can also make SolidWorks models (of solid machined parts, or bent/cut sheetmetal, or..) and have them rapid-fabbed at turnkey machine shops, probably for a bit more cost, but with great lead times (almost as good as quick turn PCBs!), excellent results, and reasonable prices in higher quantities.

Obviously, one cannot make prototypes without at least a little ME ability.  If you don't know drawings, it's worthwhile to take a course on it (or find as much on/offline material as you can).

That stuff is ok for packages that only need to get a relatively small amount of heat out overall. But IME not so great for power packages that need real heatsinking.
I have some of this stuff lying around the lab somewhere... cost a fortune IIRC.

I've heard of the stuff that does, or did, cost a fortune.  Unless you're buying it by the roll, it doesn't look like that's typical of even Digikey prices nowadays, though.

Like I said, the thinner stuff is good for dissipating some power; obviously you aren't going to get much through a 1/8" (3mm) pad filling in the standoff space below a PCB -- but it's a good solution if you can handle the extra buck or so, and the board is getting just too hot on its own.  In other words, it's a pretty darn good alternative to air, if not an ideal solution.

Tim

[krivx]

I'm not sure if we're dealing with a different class of machine shop but for something as simple as an aluminium bar I don't think anyone wants a SolidWorks file. A napkin with some dimensions and tolerances on it would probably be more appreciated.

[EEVblog]

I'm not sure if we're dealing with a different class of machine shop but for something as simple as an aluminium bar I don't think anyone wants a SolidWorks file.

The solidworks model was just done for the video as a visual aid.

[max_torque]

For "real" SMC thermal performance, use an IMS pcb!  (Internal Metal Substrate)

These are aluminium or copper pcbs, with a thin vapour depositied layer of insulation, and then the normal copper layer put on top.  I've done a 500Arms mosfet inverter on a circular <100mm diameter pcb usings an IMS pcb to mount the Fets! (although it was directly water cooled on the back of the pcb)

Because the layer of insulation is so thin and consistent, the thermal co-efficient between the copper layer and the alluminium of the pcb substrate is tiny 

[senso]

That stuff is ok for packages that only need to get a relatively small amount of heat out overall. But IME not so great for power packages that need real heatsinking.
I have some of this stuff lying around the lab somewhere... cost a fortune IIRC.

I've heard of the stuff that does, or did, cost a fortune.  Unless you're buying it by the roll, it doesn't look like that's typical of even Digikey prices nowadays, though.

Like I said, the thinner stuff is good for dissipating some power; obviously you aren't going to get much through a 1/8" (3mm) pad filling in the standoff space below a PCB -- but it's a good solution if you can handle the extra buck or so, and the board is getting just too hot on its own.  In other words, it's a pretty darn good alternative to air, if not an ideal solution.

Tim

You can get thermal pads with a thermal resistance of 17W/mK, you can still dissipate a lot using those pads, even with 3mm thickness.

[djQUAN]

Something I did way back.... with an aluminum case and TO-220 devices

http://quan-diy.com/projects/charger/liion.htm

[delmadord]

The 3D visualizations done by David are awesomely spectacular! Both thumbs up!  If they only had higher FPS. The video is million resolution times gazillion FPS and the visualization is photos basically.

[ion]

Which is better : sheet mica or sil-pad?

Martin Lorton did a video recently that showed silicone to be a bit better than mica.

[German_EE]

I am sorry about the delay as this page took some time to find. The following page provides a good discussion on thermal issues and includes some practical examples:

http://ludens.cl/Electron/Thermal.html

For those who are curious a .cl domain is in Chile.

[Psi]

Just made some mockups in Altium using Saturn to get via C/W

Obviously 0.2mm is crazy as the PCB house is going to charge extra for that many holes
But 0.4mm seems doable and does give some good benefit.

[Paul Moir]

Maybe I'm missing something but where are we supposed to be sourcing these "thermal bars/spreaders" from?

A local machine shop should be able to turn them out pretty cheap, especially if you talk to them beforehand about the design. 

Speaking of which, 4 tapped holes + 4 screws or is it 2 tapped holes and 4 screws?  What about 2 clearance holes and nuts & screws?  Seems like it would be a whole lot cheaper to make, and you could have pretty wide tolerances on both the case & bar.  3 screws would get much better pressure if you found it necessary:  might be worth putting the hole in your PCB.

EDIT:  or one centre hole and no outer ones? 

[overthere]

The cheapest solution in cooling is always not to have to cool. Speaking replacing 7805 with switching regulators.
If you need fast regulation, an option might be to do preregulation and use a second step LDO with fast response.
But that should be obvious, and sometimes cooling is really required. So lets face the topic.

A problem what I see for reverse mounting those TO220 is the stray inductance by the leads.
Of course it does not matter for 7805 and linear stuff, but for fast switching DC/DC Converter it counts.
Also the assembly house will be angry this that method, because mounting takes a long time.
In our company reverse mounting TO220 is forbidden, for bad serviceability.
But I think Dave already mentioned that its a  huge effort to mount these,

A solution by Infineon is to introduce so called topsidecooling in CAN-Package (Can=Tin, not controller area network)
There you can directly make contact with your PCB. However this package is to infineon, and they don't seem to offer new chips.
So it seems dead. Also TI has some package allowing top side cooling, but in my eyes are not so clean solution.

LEDs and linear stuff (7805) can also be mounted on a aluminium PCB.
Heat transfer is good, but you introduce parasitic capacistance, especially deadly if you're working on higher voltage. (>100V)
Soldering those with a good iron is also a challange, a reflow oven is highly recommended.
Heaving the Aluminium PCB, anyway a good heat contact to the thermal sink is requiered.

My current solution I have in mind for DC/DC Converter cooling.
Cutout by termal interface material
1. There are some really small high packages out for power transistors, like SuperSO8 by Infineon.
These have a high of 1mm. Top side cooling is no real option, as heat transfer is bad. (20k/W)
But you can extend the copper area of the pad a bit and make a recantagle with the IC as cutout in the interface material.
The interface material is available in 1mm high. So a good contact to the case could be made.

Let me make a rough calculation:
10mm² Area, 1mm Thickness, 1.5 W/mK:: 0.15 W/K. Perfect.
Capacitance introduced (Epsilon_R=5.5): 490pF. Not good, but okay. Maybe make the area smaller or use a ticker foil.
Or Put two over another.
The advantages of the solution is:
1. Small thermal "road". (Little bit longer then TO220 solution)
2. Cheap Mounting / of the shelf components
3. No via stiching required. / Components can be place on the top.
Disadvantages:
1. Small high package required (SON, SuperSO8, etc.). But should be no problem for new designs.
2. Introduces stray-capacitance (bare in mind, other solutions too, always calculate it.)
3. Dual side mounting (cost)

I attach a picture showing the proposed solution in detail.

Another solution, but too expensive in my eyes is to machine a custom cooling element, like Dave proposed.
The mounting can be simplified, by using pressfit, and a termal intace material. The capacitance problem of above will still exist and bigger,
and the termal road will be long.

[boffin]

I happy to see the MicroSupply back on the table.

Only thing I worry about is producing low voltages with higher currents, the DC-DC will be off, the linear reg will have 8.4v (ish) volts in, and if you ask for 2.4V out at 1A, you're dissipating 6W across the linear reg.

My wish list for the microsupply:

• single rotary encoder control with two additional buttons (load on/off, lock) - not unlike the Gopher supplies
• simultaneous display of max current and current
• 20V, 1A:  CV and CC modes
• Wireless Ethernet (what's the point of a battery power supply if you still need a piece of Cat5e)  perhaps an E.Imp on the i2c to do that?
• solid aluminium case

[NiHaoMike]

Actually, there are 100W audio amplifiers out there built on SMD with little additional heatsinking. The modern designs just don't get particularly hot to begin with.

[krivx]

I'm not sure if we're dealing with a different class of machine shop but for something as simple as an aluminium bar I don't think anyone wants a SolidWorks file.

The solidworks model was just done for the video as a visual aid.

I was referring to T3sl4co1l's post. The model in the video was a very nice visual aid 

[free_electron]

just insert a copper coin in the board and be done with it.

first picture : press fitted copper coin
second picture: embedded coin this gets laminated inside the board.

the coins can be shaped in any form you want.

[c4757p]

second picture: embedded coin this gets laminated inside the board.

That sounds expensive.

[zapta]

Yay, finally progress on the uSupply... good on you dave!  Remember there's a non-insignificant crowd ready to throw money at you when/if this is ever finished. *Insert take-my-money smiley here*

+1.

And, hopefully it will work with a standard micro usb USB charger (up to ~10W) instead of a yet another special wall wart.

[overthere]

Actually, there are 100W audio amplifiers out there built on SMD with little additional heatsinking. The modern designs just don't get particularly hot to begin with.

As mentioned above, a good idea is always to minimize losses first, e.g. use a switching regulator. Comparing a Class D Amp is basically a switching amplifier, but can absolutely not be used for a power supply.
(You would have a hard time getting good performance.). I propose a switching pre-regulator and a following LDO for fast response and noise reduction.


A coined solution sound interesting for me. Actually pressfitting copper could work, i now some assembly houses doing that. But thats definitly not a cheap solution.
(And: You need to remove the heat from the copper coin).

[Vgkid]

That is a great video, thanks.

[T3sl4co1l]

just insert a copper coin in the board and be done with it.

first picture : press fitted copper coin
second picture: embedded coin this gets laminated inside the board.

the coins can be shaped in any form you want.

Or soldered in,

http://seventransistorlabs.com/Images/EmbedHS1.jpg
http://seventransistorlabs.com/Images/EmbedHS2.jpg

I've also heard of PCBs with not just "heavy copper" but "ludicrous copper"; surely they use machined parts instead (much as above).  I'm sure a typical quote for a board runs "if you have to ask, you don't want to know"...

Tim

[BobC]

The thermal issues bring me back to the overall packaging plan.

First, this isn't a rack-mount instrument.  The top and sides are usable, in addition to the front and back.  (Well, maybe not the sides, since that's where the case halves meet.)  Assuming simple case modifications (drilling holes) are still free/cheap in the quantities anticipated.

Even with rubber feet, the bottom will have the least air circulation, so it is the worst surface to use as a heat sink.  Which leaves the other 5 surfaces.  Since it would be convenient for the top to be removable and still have the supply operable, heat removal shouldn't be seriously impacted when the top is absent (but it would be nice if it were enhanced!).

To me, this means an internal 1" heatsink that is screwed to the top would be very convenient, with minimal layout restrictions.  Remove the screws and the top lifts off, and the exposed heatsink may work better than with the top on.  If more heat removal is needed, an exterior heatsink may be mounted on the top.

This eliminates any need for custom aluminum bars, or difficult assembly and attachment.  The cost is that the parts would not be surface-mount, but the heat path is simplified by not having the PCB be part of it.

If we're drilling holes in the top, why not use flat-mounted (upward-facing) encoders with long shafts that stick out the top?  Remove the knobs and the top can lift right off.  This not only permits encoders to be used (yay!), but keeps them far from the binding posts, and also keeps fingers from blocking the display.  And also keeps fingers from hitting the encoders when leads are being plugged/unplugged.

With no penetrations in the bottom, feet are not needed for clearance, but may be desired for non-slip purposes.

That leaves the front and rear panels available for connectors, buttons and the display.  And how about putting the binding posts on the rear?  It would improve finger access to front panel buttons, and could even leave room to have the encoders come out the front (again).

Other thoughts:

1. Don't have the PCB use both case side slots: Have it supported primarily by the front and rear panels (and optionally use one side slot).  This would permit both the front and rear panels to be PCBs soldered to the main PCB (if desired), and the united assembly dropped into the lower half of the case (rather than sliding in).  If additional support is desired, put rubber feet on the PCB, to brace it against the bottom.

2. Use 18650 battery clips soldered onto the main PCB.  These provide enough clearance for SMD parts to fit underneath.

Did I miss anything?

[free_electron]

second picture: embedded coin this gets laminated inside the board.

That sounds expensive.

not really. the press fit stuff is cheap. they simply stamp these. same principle as those press-fit connectors.

the embedded one is more pricy due to the special shape of the coin. but if you need it you need it.

[SA007]

There is one thing I am missing from the whole setup.

Why have the cooling pad of the part electrically connected to the tab?
I think in a lot of cases the added silpad is not really needed if the cooling is isolated.

Of course the board will conduct heat worse is there is no direct copper connection, but how much worse.
It think the added cost and complexity of getting a silpad nicely fitted underneath the pcb is not really needed.

[T3sl4co1l]

There are devices with isolated grounds, but they almost always perform worse.

Generally the tab is the substrate (lowest voltage in the IC), and the package generally connects that to the center pin (as with TO-220, etc. -- oddly enough, if you crack open a TO-92, you find the 'tab' is probably off-center, due to pinout variants -- but these aren't thick or extended tabs, either).

It's not convenient to attempt to make analog ICs with an isolated substrate (say by building *all* the transistors in 'wells' which isolate them over the substrate by a diode junction), nor to add an extra insulator (such as DBC (direct bonded copper), which is popular -- and necessary -- in large switching modules).  Some devices (such as MOSFETs and diodes) are available with isolated metal tabs (most likely using a DBC ceramic insulator), and many things are available in "full pack" (encapsulation all around everything, e.g. TO-220F -- also includes FWB modules, even the ones with a metal backing plate).

Some devices *do* have natural isolation (SOI and SOS, processes for CPUs and RF), but they're more expensive or special purpose, so it doesn't matter much here.

Tim

[IanB]

You can get thermal pads with a thermal resistance of 17W/mK, you can still dissipate a lot using those pads, even with 3mm thickness.

At 17 watts per millikelvin thermal resistance, no kidding! 

[Dave]

At 17 watts per millikelvin thermal resistance, no kidding! 

No. It's watts per meter kelvin. That is actually specific thermal resistivity, not resistance.

dQ/dt = k * A * deltaT / d

dQ/dt - heat transferred per unit time (W)
k - specific thermal conductivity (W/(m*K))
A - cross sectional area (m^2)
deltaT - temperature difference between sides (K)
d - thickness of material (m)

I think they keep writing it as meter kelvins, even though they might confuse some, to avoid even greater confusion of people thinking that Km denotes kilometers. Go figure.

On a related note, why the heck doesn't the forum support greek letters other than mu, when engineers use them all the time?

[c4757p]

On a related note, why the heck doesn't the forum support greek letters other than mu, when engineers use them all the time?

Because there exist programmers too stupid for Unicode, and sadly a great many of them have written web code.

[SA007]

There are devices with isolated grounds, but they almost always perform worse.
...

Hmm, didn't even know they existed, I was aiming for an isolation on board level.
As in, place a device isolated from the cooling tab on the board.

I recently saw an image with an d-pak mounted on a board with obvious heat sinking surrounding it but not electrically connected to the heat sinking.
Of course I don't remember where I saw that board, but still.

I think there are plenty of cases where that would be enough cooling for a component.

[langwadt]

Just made some mockups in Altium using Saturn to get via C/W

Obviously 0.2mm is crazy as the PCB house is going to charge extra for that many holes
But 0.4mm seems doable and does give some good benefit.

whether those numbers mean anything depends on what you are doing. If for example you are trying to get heat from a dpak into a ground plane all but the outer ring of vias are
mostly redundant

[T3sl4co1l]

I recently saw an image with an d-pak mounted on a board with obvious heat sinking surrounding it but not electrically connected to the heat sinking.
Of course I don't remember where I saw that board, but still.

If you have a multilayer board, you can interpose layers, so for example, you might have:
1. Top Layer: pad, heatsink tab, copper pour, thermal vias connected
10 mil prepreg
2 . Top Inner: GND Plane
40 mil core
3. Bottom Inner: copper pour, thermal vias connecting to pad
10 mil prepreg
4. Bottom Layer: GND Plane

And then further heatsinking via the GND Plane(s), which could include SMT heatsinks for example.

Because the prepreg layers are thin, you get not-terrible conductivity between the pairs on each side (layers 1-2 and 3-4).  Not enough that it can be net pad size, so you still want copper pours, and doubling up helps that much more (even though the second pour is only connected by vias).

Again, this can be done with vias around the periphery.  The GND-GND stitching (needed to get heat from layer 2) has to go outside the pad, obviously (or use buried/blind), but the vias from the connection itself can be ViP or peripheral (in accordance with the usual advice concerning that).

This doesn't work on 2 layer boards because there's no layer pair with close spacing.  A four layer board typically has two pairs with close spacing, so you can use one, or both in parallel, to your advantage.

Obviously the added capacitance is not insignificant, so you want to direct it somewhere appropriate.  Ground is probably the best, for typical applications (switching node?).  A fully isolated heatsink node would carry essentially full noise, which wouldn't be real great on something like an offline SMPS.

Tim

[loneoceans]

I'm not sure if I'm convinced that this is a good method for heat-sinking.

In the classic approach, one needs to incur added cost of hand-soldering, but it seems like the rest of the assembly is easier - single hole to be made in case and cheaper to get this manufactured, then a screw and a sil-pad.

In Dave's approach, you not only incur greater cost of having the large device on the PCB (e.g. D2Pak is significantly larger in footprint than TO220 standing up), but you also need to have more space on -both- sides of the PCB, adding cost in drills (can add up depending on your volume), and having to design and manufacture the aluminium bar. Finally you still need to get the bar assembled (4 screws required in this case) and use sil-pads. Overall I'm not sure if I'm convinced this is any sort of cost or assembly efficient method at all.

Previously when tearing down some electrical systems, another method I came across was to have the legged device installed normally but with a cut-out on the PCB. The case is shaped to fit the device, and a sil-pad 'jacket' is slipped over the device. This is then clipped to the case with a clip that looks like a binder clip, but very strong. Aside from the one-off cost of design the custom case (probably not useful for you in this case but it's a good alternative for large volume runs), installation is quick and simple, and the result is very good.

In the picture I've attached below, you can see the TO220-shaped device (I've removed the clip), which doesn't need a jacket because it's the plastic kind which is entirely insulated. The board also had another TO247 device which was heat-sinked to the case in the same way with a silicone jacket.

[mjt]

The other day I was looking at power mosfets and I was wondering - how do companies expect you to heatsink a D2PAK / TO-263 ?

I mean, take the Infineon IPB010N06N or the  IR IRFS7530-7PPbF - surface mount chips, 300W+ power dissipation.

Obviously you can get little surface mount heatsinks but you're not going to dissipate 50 W through a 3°C/W heatsink, let alone 300W.

All the power electronics of this scale that I've seen has used big through-hole TO-220 and TO-247 packages, bolted directly onto heatsinks - but someone must be buying these surface mount power mosfets, as loads of companies manufacture them.

Does everyone use expensive insulated-metal-substrate boards? I haven't seen many of them in products, and Eurocircuits charge €148 for 10 boards at 50mm x 50mm - quite a price jump for my hobby projects, when I'm used to paying $14 for the same in FR4 and they have to be single-sided to boot.

Custom copper inlays in the PCB sound even more expensive. Or can I order them in small quantities and fit them manually? I can't find anything for thermal coin, thermal inlay or copper coin on farnell or digikey.

Would a 20mm x 20mm area of 0.4mm thermal vias drop the thermal resistance to 0.15 °C/W or would the thermal resistance of the top layer of the PCB lead to diminishing returns as you get further from the component?

More generally, how do manufacturers expect us to dissipate power from these 300W+ TO-263 MOSFETs?

[rs20]

More generally, how do manufacturers expect us to dissipate power from these 300W+ TO-263 MOSFETs?

I think they don't expect you to actually dissipate 300W+. These headline specs are theoretical capabilities, available there in reserve for those few customers that are actually using IMS board. I mean, what do you expect them to do, sell themselves short and claim that their transistor craps out at 5W just because the typical customer can't dissipate 5W cheaply enough? The transistor can dissipate 300W, so they're allowed to claim it. (Others would go further and say that the headline specs are pure marketing bull.) For the rest of us, we look at the resistance spec and get a mosfet that only dissipates 1W or less in our particular usecase.

[T3sl4co1l]

Simple: manufacturers lie about ratings.

Specifically, what they do is submerge the device in nucleated boiling liquid (frothing freon at 25.00C) and perform the test.  When they say "case temperature = 25C", they literally mean all points on the case are this temperature.  The dissipation is no different from a TO-220, as should be expected: under such fantastic cooling force, that little extra tab of copper ain't gonna do jack.

Their excuse being, they did the test.  They typically don't tell you how they did it (if so, it's buried in an obscure appnote).  Whether you can actually achieve even the teensiest fraction of that in practice, f**k you...

So you must willfully ignore the ratings they give (unless they give reasonable numbers for specified board layouts), and either calculate it yourself, or at least remember offhand what kind of ratings to expect.  You're in the single digit watts, for the most part -- true whether it's anything from SOT-89 to D2PAK.  You're primarily board-area limited.

Tim

[rs20]

Simple: manufacturers lie about ratings.

Specifically, what they do is submerge the device in nucleated boiling liquid (frothing freon at 25.00C) and perform the test.  When they say "case temperature = 25C", they literally mean all points on the case are this temperature.  The dissipation is no different from a TO-220, as should be expected: under such fantastic cooling force, that little extra tab of copper ain't gonna do jack.

I think this is a bit harsh. When you design with a MOSFET, you always have to independently check 3 things (plus a few others ): the current limit, the power dissipation limit, and the junction temperature limit. When I see 300W power dissipation limit, I say "OK, that's one less limit I need to worry about" rather than "I will angrily ignore these lies grrr". Real designs with modern MOSFETs are junction-temperature limited, so that's what you design by; the 300W figure is there to tell you "there's practically no fixed power dissipation limit, move along to the current and junction temperature limits". IRF even provides junction-to-ambient thermal resistances for certain FR4 board layouts, the actual power limits in those situations are a single division away.

Also, one day someone will might use the transistor in a bath of boiling freon on an insulated metal board. No reason to sell the part short for that user just because us mere FR4 mortals are junction-temperature-constrained.

[jb79]

Hi Dave!

Nice to see that you are working on the µSupply again. Will you update the project page in the near future, or is there a plan when the µSupply should be finished?

best regards Jürgen

[dino]

Just curious, how would one cool down a bga package (5-10w) using the enclosure? Would enclosure stress compromise the bga solder balls?

[cypherpunks]

Remember, the outer layers start with 1/2 oz (17 µm) copper.  Then you drill PTHs, add a ghost-thin layer of electroless copper, and plate on an additional 17 µm to get the final desired trace thickness.

So the inside of vias are generally 1/2 oz thick.  You may lose a bit because the plating solution doesn't mix as rapidly in the confined area.

This is important when computing electricalor thermal resistance.

[T3sl4co1l]

BGAs can be heatsinked from on top, same as any; many high-performance FPGAs, etc. have a metal top.  Power devices are also designed to sink through the pads (which could be pretty good, seeing as solder balls are solid metal... but vias are not).

Usual heatsinking method is a spongy thermal pad, so the force is small.  A thinner pad (or straight grease) could be used, but then the surfaces need to mate closely.  An example might include the spring loaded brackets common in laptops (which also use BGAs or flip-chips for all the important parts: CPU, GPU, system chip, memory..).

Tim

[EEVblog]

Nice to see that you are working on the µSupply again. Will you update the project page in the near future, or is there a plan when the µSupply should be finished?

No plan. But yes it's being worked on.

[jb79]

Hello!

I'm trying to redesign Rev. C of the µSupply for myself to get more current out of it (3A), therefore I'll use the LT3083, maybe 2 of them in parallel.
It should be powered by 24V DC, so I'll place a DC/DC preregulator instead of the battery powered solution.

To get a more exact current measurement, I plan to use the INA219 in the circuit but I'll remove one of the 10Ohm resistors (R1) and replace it by a 1Ohm resistor in series with 2x10Ohm resistors in parallel. So the total shunt resistance for the opamps there will be still 1 Ohm but I get the possibility to get 1/10 of the voltage over the current shunt to measure by the INA219. Together with /8 measurement this will give me up to 3,2A total measurement range on the INA219. Of cause I take care about the power dissipation in the shunt resistors.

To get 3A out it is also necessary to change the voltage divider R25/R36. At it's output there should be 3,072mV to get 1mA/bit out of the 3,3V PWM signal.
To keep the different parts low I came out with the following values: R25=11k (10k+1k), R36=150k, resulting in only 0,08% error for the divider, the tolerance of the resistors is much higher. If only one 10k resistor for R25 is used, the error would be 0,71%.

Any other issues to deal with, when I try to get 3A out of the supply?

[BobC]

Nice to see that you are working on the µSupply again. Will you update the project page in the near future, or is there a plan when the µSupply should be finished?

No plan. But yes it's being worked on.

I just hit a situation where I need precisely this supply!  While I'd love to get the thing already built and boxed, I'd be more than happy settle for the Gerbers (or PCB) and a BOM (or kit).