Paper clip heatsinking for LM317?

[ezkcdude]

Well, unfortunately, I won't be able to finish today. I'm missing 100R 0805 resistors, which are in series with the XO. And everything was going so well...

 

[Garbz]

That value is not all that critical. If you have any other resistors <120ohm they could go there.

 

[Xakepa]

Quote:

Originally Posted by tomb In fact, heat transfer may be inhibited, because the optimum surface area exposed to the air is actually a smooth surface. Increasing the area of that contact surface has benefit, but not through simple surface roughness - those are not "fins."

Hold your fire. I was recommending grooves in the millimeter range, which IMO are big enough to be considered a microscopic increase of the surface. We could argue on the airflow, but (being an experimentalist) I'd rather see the temperature reading (with the same ambient) with and w/o groves.

If I have time, I could try that...

 

[tomb]

Quote:

Originally Posted by Xakepa Hold your fire. I was recommending grooves in the millimeter range, which IMO are big enough to be considered a microscopic increase of the surface. We could argue on the airflow, but (being an experimentalist) I'd rather see the temperature reading (with the same ambient) with and w/o groves. If I have time, I could try that...

With respect, you missed the point. You can't scale the air molecules. They could care less about a millimeter ... and no, I'm not saying that air molecules are bigger than a millimeter. If you look at some surface roughness charts with respect to air movement - it takes bigger "bumps" than that before the air can see them.

An aluminum machined heat sink already has an optimum fin size, even accounting for manufacturing and structural considerations. The cost of machining already drives them to an optimum dimension on the heat transfer surface. Otherwise, they'd be selling heat sinks with sandpaper finishes.

 

[ezkcdude]

Quote:

Originally Posted by Garbz That value is not all that critical. If you have any other resistors <120ohm they could go there.

I wish I did. I didn't have any 0805 in that range.

 

[Xakepa]

Quote:

Originally Posted by tomb They could care less about a millimeter ... and no, I'm not saying that air molecules are bigger than a millimeter. If you look at some surface roughness charts with respect to air movement - it takes bigger "bumps" than that before the air can see them.

I got it. That seems counter-intuitive at first untill I realised that the flow is so slow it could be insensitive to the surface roughness.

 

[tangent]

Okay, enough speculation. Time for some actual science.

Test Setup
Six LM317s configured as 0.125 A current sources propped up by silicon diodes. The diodes are there to make the numbers come out nice: with 10 V across the circuit, 1.25 V is dropped across the current-setting resistor, ~0.7 V is dropped across the diode, leaving a nice, round 8 V across the regulator. Since the current through the regulator is nominally 0.125 A, that makes for a very easy to handle 1 W. Ya might think I planned this or something.

From left to right, front to back, the configurations are:

1a. Naked regulator; b. paper clip heat sink. (It was easy enough to change from one to the other that I didn't bother making separate board instances for these two.)

2. 4x penny heat sink.

3. Aavid-Thermalloy model 577202B00000G heat sink.

4. Penny bolted to regulator.

5. Penny epoxied to regulator.

6. Penny paper-clipped to regulator.

Results

1a: 0.127 A through regulator, 75.0° C
1b: 0.127 A, 71.3° C
2: 0.125 A, 55.5° C
3: 0.126 A, 54.5° C
4: 0.126 A, 60.0° C
5: 0.126 A, 61.0° C
6: 0.126 A, 58.0° C

Discussion

1: The naked regulator is the baseline, and as you can see, it gets pretty hot. This is about as hot as I'd let a chip get before I'd say a heat sink is no longer optional. Adding a paperclip knocked a few degrees off. I'm surprised it did anything at all, actually. It's not terribly effective, but it's literally better than nothing.

2: This is four pennies bolted together, with nuts in between each penny to give some room for air to get in there. The penny closest to the heat sink is oriented to give maximum surface contact, and the others are oriented higher to get them more into the open air. You might instead orient the pennies in a rosette sort of pattern, but I didn't have enough room here to get away with that. As-is, it's nearly as good as the purpose-made heat sink. It takes some work to put it together, but if it's late at night, you don't have a heat sink, and you can't get down to your local Radio Shack, dig up some pre-1982 pennies, fire up the drill press and get crafty.

3: Bog standard heat sink. Available lots of places. Works well. Will probably cost you less than what it's protecting. Standard lore says parts will last twice as long for every 10° C drop, so this gives a nice round 4x lifetime increase. It'll literally pay for itself unless you're using some extremely cheap parts.

4: #2's little brother, and it works nearly as well. You might be able to equal its effectiveness by lapping the mounting surface flat to maximize the surface contact area.

5: Again, I'm a bit surprised. I would have expected to give up more than 1 degree by using glue instead of heat sink compound. Now, if you're dealing with more than 1 W of dissipation, the penalty here will get larger.

6: After seeing #1b, this makes sense. It's just adding the surface area of the paper clip to the penny, so you get a few degrees additional drop. You might get another degree or two of drop if you use some heat sink compound instead of allowing there to be an air gap.

Test Details

The regulators are all Fairchild LM317Ts. They're probably even from the same tube.

Each heat generator was allowed to cook for at least 15 minutes to get it up to thermal equilibrium.

Temperatures are approximations. I would often see the regulator swing back and forth over about a 1 degree range. The number I give above is my off-hand estimate of the center of the range.

Where possible, the temperature was measured in the same location. Regardless of actual location, I spent a few minutes on each part looking for the hot spot. The location I used most is down and to the left of the mounting hole, with the temperature probe jammed in the corner between the IC encapsulation and the tab.

All of the pennies are 1981 or earlier vintage, that being the last year US pennies were made mostly of copper. After that, they're mostly zinc, with some copper plating. (Source) The pennies in cases 4, 5, and 6 are all 1981 pennies made in the same mint. I didn't feel a great need to find identical pennies for case 2.

In cases 2, 3, and 4, I used a silver-bearing heat sink compound. Not Arctic Silver ®©℠, but something of that sort. In cases 1 and 6, there's nothing added between the heat sink and the regulator. And in case 5, it's a fast-dryng epoxy.

In all penny heat sink cases, the surface towards the regulator is the reverse side, as it looks like the obverse has more relief. Also, Miss Manners says it's not nice to apply sticky substances to ex-Presidents' faces. (Miss Manners has nothing to say on the topic of power tools, so I'm also okay in cases 2 and 4.)

All pennies were cleaned in an acid bath to remove dirt and corrosion. I used CLR initially, but switched to a vinegar and salt solution that was recommended on a coin collector's site. I flushed the acid off thoroughly with tap water.

Holes drilled through the pennies are the same size as the holes through the Aavid heat sink and the regulators' tabs. I used the thinnest screws I had on hand and clipped their length as well as I could using my wire stripper's bolt cutting holes to minimize their contrubution to the thermal bulk. (#2-56 size.) Normally you'd use something more like #6-40, or maybe even #8-32.

 

[tangent]

Oops, forgot one of the configurations I wanted to test:

Using 600 grit wet/dry sandpaper, I lapped the reverse side of the previous #6 configuration's penny flat. Then, with a 400° C iron, plenty of liquid flux and about 10 minutes of futzing, managed to solder the penny to the regulator. (I could do it in just a few minutes now that I've figured out the technique.)

So, what did that do? It saved another degree relative to using heat sink compound: 59° C. Not as much as I'd guessed. Probably more worthwhile at higher wattages.

 

[ezkcdude]

Tangent, that was great

.

So, can I sum it up by saying, it is best to have a true heatsink, but in a pinch, paperclipping a penny to the regulator will do just fine?

 

[Xakepa]

Mister Tangent,

Thank you very much for the time and effort.

From what I see, it's hard not to note that all penny-based cooling solutions are <10% from one another, and an engineering-oriented mind tends to sum this up as "they produce similar results". I guess Voodoochile was right...

 

[tangent]

Quote:

Originally Posted by ezkcdude Tangent, that was great .

Thanks, but I'd be careful about drawing conclusions just yet. After writing that up, I began to be more and more bothered by a few things:

1. There were several places where I opined that a heavier load would show the differences more clearly. Okay, Mr. Smarty, so why don't you test with a heavier load?

So I am. I'm up to 2W now, the highest I can do with my power supply without changing out all those resistors, while staying with integral wattages.

2. I'm not sure I removed all the environmental and human factors, and test method variability. Is 15 minutes enough warmup time? I'm trying 30. The temp probe never settled on a value, and its swings weren't regular. Can I fix this? I didn't use the same probe point each time. I've decided I can't justify that.

To try and fix all these sorts of problems, I'm using my helping hands to hold the temp probe against the regulator, always in the same position and always at the same angle. In addition, I'm using my meter's data logging function so I can stay some distance away from the measurement. For all I know, my own breathing was affecting it, because as I move around, that changes how much air is moving over the test area!

3. I swapped #2 and #4, so I can fan the pennies out in the rosette pattern. We'll see if that makes a difference.

4. I thought I was being cute by rigging the voltages so the test was right at 1 W even. Except that it isn't. The current setting resistors are well within spec (they're more like 1% apart, not 5% as they're allowed to be), but there is still some difference. So, I'm taking the actual current, calculating the correct voltage drop, and adjusting the test voltage to give the same wattage in all cases.

EDIT: 5. My excuse of not cutting the mounting bolts flush with the nut was thin. I used a different method to cut the bolts so I could get rid of all the excess.

So, let me redo these tests, and get back to you. It'll take a day or three.

 

[ezkcdude]

Of course, you're right to check all these things. But, I doubt we're talking order of magnitude differences here. You never know, though.

I'm sure by the time you get these tests done, I'll have received my heatsinks, but I'll still interested to see your results.

 

[Garbz]

Tangent I'm not sure what probe you are using but when I check heatsinked parts with my K-type probe, I find it helps to dunk it into some heatsink paste. Every time I've tried this I found the probe settled faster and in some cases on a slightly higher temperature.

Noadays I just measure temps where a bit of heatsink paste is spread out from underneat the regulator during clamping. As always milage may vary.

 

[tangent]

Quote:

Originally Posted by ezkcdude I doubt we're talking order of magnitude differences here.

Ah, no, definitely staying within the same order of magnitude. But, I've redone three of them, and relative rankings have definitely changed. In particular, the lone penny isn't coming out so good. I'm not saying more...I need to complete the tests before I'm even sure of what I've got.

Quote:

Originally Posted by Garbz I find it helps to dunk it into some heatsink paste.

An interesting idea. This might be enough to get me to start again. Actually, more of a last straw sort of thing. I've been doing better data analysis this time around, and I'm finding weaknesses in the test method again. Just one example: my gathered data isn't strictly following a Gaussian curve. There's no reason this shouldn't be Gaussian, so the method's got to be flawed. Most likely I either am not taking enough data samples, or I still haven't eliminated all the external variables.

 

[ezkcdude]

Actually, since your doing these tests, anyway, could you try a setup with a paperclip and a piece of aluminum foil? I would think this would beat the penny.