LM317/337 heatsinking pads

[axemaster]

I've been thinking of building fairly high power split rail power supplies with LM317/337 regulators, and I'd like to use those aluminum boxes on Digikey. For example:

http://www.digikey.com/product-detail/en/hammond-manufacturing/1590Y/HM955-ND/1090766

The aluminum box seems like it would make a superb heatsink, and I plan on bolting TO-220 packaged regulators directly to it. However, the metal tabs of the regulators are not at ground, and will short together through the case. I know that thermal pads are usually used to solve this, so I was hoping to get some insight on those. In my investigation I've found that there seem to be mainly 2 types of thermal pads:

- silicone rubber pads, which offer medium performance
- phase change pads which offer ~3x better performance

However, it's not clear to me if phase change pads will offer electrical isolation from the case. What's the deal on this? Am I overlooking some better solution?

[KE5FX]

I've been thinking of building fairly high power split rail power supplies with LM317/337 regulators, and I'd like to use those aluminum boxes on Digikey. For example:

http://www.digikey.com/product-detail/en/hammond-manufacturing/1590Y/HM955-ND/1090766

The aluminum box seems like it would make a superb heatsink, and I plan on bolting TO-220 packaged regulators directly to it. However, the metal tabs of the regulators are not at ground, and will short together through the case. I know that thermal pads are usually used to solve this, so I was hoping to get some insight on those. In my investigation I've found that there seem to be mainly 2 types of thermal pads:

- silicone rubber pads, which offer medium performance
- phase change pads which offer ~3x better performance

However, it's not clear to me if phase change pads will offer electrical isolation from the case. What's the deal on this? Am I overlooking some better solution?

Ditch the LM317s in favor of a more modern part that lets you ground the tab, e.g. MIC29512. 

[Zero999]

Yes, using the case as a heat sink is an excellent idea. I've done it before and it works well.

I soldered the tab of each regulator directly to a thick piece of copper which was mounted to the case with a piece of PVC insulation tape sandwiched in-between and fixed using steel screws with nylon bushes and washers for insulation.

[jeroen79]

Ditch the LM317s in favor of a more modern part that lets you ground the tab, e.g. MIC29512. 

That could be dangerous.
A metal case should be earthed for safety.
Otherwise some fault could put a high voltage on it.

If the output is connected to the case it would also be earthed.
That is fine if you only want to work at earth reference but you would not have the versatility of a floating powersupply.
And if the supply is connected to a non-earthed outlet the case would be at whatever potential the 'ground' output is.

[Zero999]

Ditch the LM317s in favor of a more modern part that lets you ground the tab, e.g. MIC29512. 

That could be dangerous.
A metal case should be earthed for safety.
Otherwise some fault could put a high voltage on it.

Who said anything about not earthing the case?

If the output is connected to the case it would also be earthed.
That is fine if you only want to work at earth reference but you would not have the versatility of a floating powersupply.
And if the supply is connected to a non-earthed outlet the case would be at whatever potential the 'ground' output is.

I agree, a floating supply is a good idea, as it's more flexible. As far as safety is concerned: it's not a good idea to float it at more than 60V, relative to earth.

[Mechatrommer]

Ditch the LM317s in favor of a more modern part that lets you ground the tab, e.g. MIC29512. 

whats the complementary negative regulator for MIC29512 if LM337 need to be ditched as well?

[wraper]

You can use thin mica pads + thermal paste. Much better thermal conductivity than rubber and cheap.

[jeroen79]

Who said anything about not earthing the case?

Noone did but I thought I'd mention it anyway.
But if you earth the case and you connect one of the outputs to the case the output is no longer floating but earth referenced.
That is fine if this is a conscious design choice but not good if it comes as a surpise later on.

[Gyro]

One thing to remember when using diecast enclosures is that their thermal conductivity might not be as high as you expect. Their walls are quite thin and I suspect that there is some zinc in the mix.

They do work, but you will find a hot-spot around the device so keep dissipation at a reasonable level.

[axemaster]

Oh, here's a corollary question: has anyone here ever used a FLIR camera to measure the temperature of ICs? I have access to one and it seems like the best method since a thermocouple would act as a heat sink and corrupt the measurement, especially for small parts. However, I've not had much luck finding out the emissivity of IC package plastics, or the emissivity of solder mask.

[rob77]

Ditch the LM317s in favor of a more modern part that lets you ground the tab, e.g. MIC29512. 

whats the complementary negative regulator for MIC29512 if LM337 need to be ditched as well?

and here the silence comes.... 
all of the sudden there is no obvious replacement for such a "archaic" negative rail regulator...

Ditch the LM317s in favor of a more modern part that lets you ground the tab, e.g. MIC29512. 

so for all of you guys who immediately suggest to ditch those "dinosaur" components... either propose a complete solution or stay silent in this case please propose parts for both positive and negative rails.

furthermore - stop suggesting expensive and "modern" parts to everyone - please stop suggesting a $7 replacement (MIC29512) for a 50cent part (LM317) to hobbyists and beginners... if LM317 is capable of doing the job then that LM317 is good enough and there is no reason to replace it with a 14x more expensive part.

@ OP:  regarding thermal pads... either go for the silicone pads, or for higher power dissipation go for mica pads with a good thermal paste.

[macboy]

Another option is to use a variant of the LM317 that has an isolated TO-220, aka "Full Pack". ST Micro makes makes one. These can be bolted directly to any case with ordinary metal hardware. This will have slightly less thermal conductivity than the normal metal tab version with a properly installed mica washer + paste.

EDIT: My apologies, I see now that there is no corresponding negative device with isolated packaging.

[edavid]

If you use a dual secondary transformer, you can use a positive regulator for the negative output.

Also, mica + paste is not worth it for a medium power application like this.  Too messy.

[Mechatrommer]

all of the sudden there is no obvious replacement for such a "archaic" negative rail regulator...

not to mention the LM317 is a floating regulator, so you can make it run at anything higher voltage, thats why i guess no GND tab. the unavailability of floating and LDO negative regulator makes LM317/LM337 combo is not yet replacable.

[David Hess]

If you use a dual secondary transformer, you can use a positive regulator for the negative output.

Also, mica + paste is not worth it for a medium power application like this.  Too messy.

I was just about to post that.  The LM317 is also easier to frequency compensate than the LM337.

I disagree about mica and thermal grease being too messy though.

[axemaster]

rob77 has the right idea here - I am looking to build using reasonably cheap components. There's a very big difference between $.50 and $3.50 parts. The other thing is that the LM317/337 tolerate very considerable input voltages, whereas the GND referenced packages only seem to go to 16V or something like that.

Another option is to use a variant of the LM317 that has an isolated TO-220, aka "Full Pack". ST Micro makes makes one. These can be bolted directly to any case with ordinary metal hardware. This will have slightly less thermal conductivity than the normal metal tab version with a properly installed mica washer + paste.

EDIT: My apologies, I see now that there is no corresponding negative device with isolated packaging.

Yeah I looked at that too, very disappointing that they didn't give both versions this package.

If you use a dual secondary transformer, you can use a positive regulator for the negative output.

Also, mica + paste is not worth it for a medium power application like this.  Too messy.

Wait, you can do that? I was under the impression that using positive regulators to generate negative rails causes problems... after all, why would negative regulators even exist otherwise?

[edavid]

If you use a dual secondary transformer, you can use a positive regulator for the negative output.

Wait, you can do that?

Sure.  You just build 2 completely independent supplies, both using positive regulators.

I was under the impression that using positive regulators to generate negative rails causes problems... after all, why would negative regulators even exist otherwise?

E.g. for when you use a center tapped transformer instead of dual secondary.

[mikerj]

However, it's not clear to me if phase change pads will offer electrical isolation from the case. What's the deal on this? Am I overlooking some better solution?

Phase change materials do not guarantee any isolation by their very nature.  They are excellent in terms of thermal impedance, but not the correct choice if isolation is required.

[Phoenix]

Oh, here's a corollary question: has anyone here ever used a FLIR camera to measure the temperature of ICs? I have access to one and it seems like the best method since a thermocouple would act as a heat sink and corrupt the measurement, especially for small parts. However, I've not had much luck finding out the emissivity of IC package plastics, or the emissivity of solder mask.

Yes and it's fantastic. I've done comparisons against a thermocouple and IR camera on IC packages and PCBs - they agree within a degree or 2. IC's are packaged in a type of epoxy, which has a high emissivity. FR-4 and solder mask are also similar. 0.9 is a good estimate for most parts on a PCB. Don't try to measure solder, component leads or other metal surfaces as they will be wildly inaccurate (a reflection of the background temperature).
http://www.chipmos.com/_en/pdf/108-Poly99.pdf

[David Hess]

Wait, you can do that? I was under the impression that using positive regulators to generate negative rails causes problems... after all, why would negative regulators even exist otherwise?

Like edavid says, negative regulators are often required because of other circuit constraints.  If you have a floating source, then using a positive regulator for a negative output is often preferable for performance reasons.

Integrated processes usually lack PNP power transistors and in the early history of discrete silicon transistors, power PNP devices were expensive or low performance so old discrete linear power supplies often used floating NPN based positive regulators to generate negative outputs.  Check out the power supply for the Tektronix 453 oscilloscope for an example.

[macboy]

Wait, you can do that? I was under the impression that using positive regulators to generate negative rails causes problems... after all, why would negative regulators even exist otherwise?

Like edavid says, negative regulators are often required because of other circuit constraints.  If you have a floating source, then using a positive regulator for a negative output is often preferable for performance reasons.

Integrated processes usually lack PNP power transistors and in the early history of discrete silicon transistors, power PNP devices were expensive or low performance so old discrete linear power supplies often used floating NPN based positive regulators to generate negative outputs.  Check out the power supply for the Tektronix 453 oscilloscope for an example.

For the same reason, the "classic" (read: old) LM337 negative regulator uses an NPN power device, making it identical to a low dropoout (LDO) regulator in terms of how it operates. That means greater difficulty in getting it stable, especially with a wide range of loads as used with a bench supply. For that reason, a pair of LM317s on separate transformers or secondaries is greatly preferred. The LM317 is comparably easy to tame.

[Zero999]

Wait, you can do that? I was under the impression that using positive regulators to generate negative rails causes problems... after all, why would negative regulators even exist otherwise?

Like edavid says, negative regulators are often required because of other circuit constraints.  If you have a floating source, then using a positive regulator for a negative output is often preferable for performance reasons.

Integrated processes usually lack PNP power transistors and in the early history of discrete silicon transistors, power PNP devices were expensive or low performance so old discrete linear power supplies often used floating NPN based positive regulators to generate negative outputs.  Check out the power supply for the Tektronix 453 oscilloscope for an example.

For the same reason, the "classic" (read: old) LM337 negative regulator uses an NPN power device, making it identical to a low dropoout (LDO) regulator in terms of how it operates. That means greater difficulty in getting it stable, especially with a wide range of loads as used with a bench supply. For that reason, a pair of LM317s on separate transformers or secondaries is greatly preferred. The LM317 is comparably easy to tame.

The LM317 has a Darlington pair output stage.

The LM337 has a Sziklia pair output stage. It's not as bad as the single PNP transistor used in low drop-out regulators but it's not as stable as a Darlington pair.

https://en.wikipedia.org/wiki/Sziklai_pair
https://en.wikipedia.org/wiki/Darlington_transistor
http://www.ti.com/lit/ds/snvs778e/snvs778e.pdf
http://www.ti.com/lit/ds/symlink/lm317.pdf

[David Hess]

For the same reason, the "classic" (read: old) LM337 negative regulator uses an NPN power device, making it identical to a low dropoout (LDO) regulator in terms of how it operates. That means greater difficulty in getting it stable, especially with a wide range of loads as used with a bench supply. For that reason, a pair of LM317s on separate transformers or secondaries is greatly preferred. The LM317 is comparably easy to tame.

Unfortunately the LM337 has the stability of a positive low dropout regulator combined with the high dropout of an LM317. 

The stability issue is not a big deal; you just have to remember to include the bulk output capacitor close to the regulator which positive low dropout regulators usually require also.

The LM337 has a Sziklia pair output stage. It's not as bad as the single PNP transistor used in low drop-out regulators but it's not as stable as a Darlington pair.

Huh?  Did you check the schematics you linked?

The two transistors output pairs for both regulators use the Darlington configuration.

[Zero999]

The two transistors output pairs for both regulators use the Darlington configuration.

Oh, yes you're right. I was looking at Q11 & Q12, when I should have been looking at Q25 & Q26. 

[axemaster]

OK, thanks everybody for your help! In the next few weeks I plan on doing a pretty thorough study on pcb heatsinking since I've found the online resources pretty thin. I'll post it to this forum when I'm done. Have a nice week!