Heatsink on OPAMPs

[guitarkill]

Just because you can see something on a scope doesn't mean you can genuinely hear a difference. The only way to prove if there really is an audible difference is by a double blind listening test where the listeners don't know what is what so that they won't have any preconceived notions or bias about what they are hearing. I have my doubts that anyone could hear a difference in the sound in an overdrive pedal like a TS808 which is already producing lots of distortion intentionally. What is the point of a heat sink in that scenario? Or really any other? I can see if we were talking about a voltage regulator or something but an opamp in an audio application? To each his own I guess.

I just read this article this morning. I thought it was really good with respect to his concept of "sighted listening" (or in other words, pyschoacoustics)  http://nwavguy.blogspot.com/2011/08/op-amps-myths-facts.html

[M23Bomber]

Hello all,

Just as a joke, Sometimes I use my equipment for "experiments" like this one. opening a window in the room will cause differences. Obviously almost anyone can listen,lets say 1  But its fun to know that a pedal in sun will produce different distortion than the one in the shadow, etc... These were mainly observations .

Regards,
M.

[Ice-9]

I've always been fond of painting on racing stripes and supergluing on tailfins.

I like to grind off the top plastic of the op-amp package and watch those lil'ol elctrons rush around.  

That's not too far from the truth actually when I think about it. When I have had repairs to do to circuits and it has been unavailable IC's and those same IC's have broken off legs, I have ground down the package to reveal the the leg internally on the chip. Yes it's a last resort job to getting something working again but it does work.

[Mark Hammer]

Hello all,

Just as a joke, Sometimes I use my equipment for "experiments" like this one. opening a window in the room will cause differences. Obviously almost anyone can listen,lets say 1  But its fun to know that a pedal in sun will produce different distortion than the one in the shadow, etc... These were mainly observations .

Regards,
M.

Every year, there are at least 4 or 5 threads started here whose central thrust is "It worked fine before I boxed it up".  Which raises the I'm-so-close-to-it-I-can't-see-it element of the fact that the circuit goes into a box which is:
a) unvented, hence containing any heat generated internally (and what would a heatsink inside a small unvented box do anyway?);
b) made of the same (or similar) material that heatsinks are.

And that, in turn, raises the matter of whether what is reliably measurable on the bench from an unenclosed circuit can be generalized to something stuffed inside of an unvented aluminum box.

But while we're in the neighbourhood...
1) do chips like the LM833 and NE5532, (capable of sinking lots of current into low-impedance loads, hence long-time favourites for headphone amplifiers), get warm when used as headphone amps?
2) does the behaviour of diodes change substantively with temperature?
3) Do people stick those frilly chef-hat heatsinks on GE transistors in any of their builds, and if so, does it make the build less temperamental?

[R.G.]

But while we're in the neighbourhood...
1) do chips like the LM833 and NE5532, (capable of sinking lots of current into low-impedance loads, hence long-time favourites for headphone amplifiers), get warm when used as headphone amps?

Yes. Internally, the output stage is much like the output stage of any audio power amplifier, and it's biased into Class A or AB. Mother Nature insists through Her math that such a stage will dissipate at most about 40% of the AC output to the load for a sine wave signal. It can be much more if you feed it a half-power-supply square wave. This makes it dissipate heat equal to the power delivered to the load. So, paraphrasing Johnny Cochrane, "If feeds out watts, it will get hot."  Correspondingly, if the load is trivial, generally over 2K to 10K, the power is negligible and is mostly the power spent on internal biasing conditions, which can be calculated from the datasheet numbers.

2) does the behaviour of diodes change substantively with temperature?

Yes, a great deal. Their leakage increases with temperature, and their forward voltage at any given current goes down. How big these changes are depends on the material used  - germanium, silicon, gallium arsenide, silicon carbide, etc.  This comes right out of Mother Nature's math again, and is highly, highly calculable. About half the circuitry inside an opamp is at least partially used for making the changes with temperature in one device be cancelled by another, so the overall tempco is as near zero as the highly experienced design and layout folks can make it. Diffamps, for instance, are used not least because their differential output tends to be compensated by the drifts of the two transistors cancelling.

3) Do people stick those frilly chef-hat heatsinks on GE transistors in any of their builds, and if so, does it make the build less temperamental?

Interesting point. To be clear, I don't know the answer to that as I've never tried it. But I can speculate.   

If something - anything! - doesn't produce heat, it eventually becomes the temperature of its surroundings. If it produces little heat compared to its ability to transfer heat to the surroundings, its temperature rises by an amount modelled by the heat flow through a thermal resistance, in the thermal analog of Ohm's law. So a device that dissipates 1W inside, and has a thermal resistance from the insides to the surrounding air of 50 degrees C per watt (50 C/W), then its insides will be 1W* 50C/W = 50C hotter than the outside air.

Thermal resistance adds like electrical resistance, so electronistas tend to separate the thermal resistance into a resistance from the chip to the case, and then from the case to the ambient air. A heat sink stuck to the case affects only the case-to-ambient-air resistance.
Here's a good outline of this view of things:   http://www.altera.com/support/devices/power/thermal/pow-thermal.html and this is a good, concise table of the typical thermal resistances of various packages: http://cds.linear.com/docs/en/packaging/Linear_Technology_Thermal_Resistance_Table.pdf.

The TO-5 package has a thermal resistance from chip to case of about 40C/W, and without a heat sink, 150C/W to ambient air; of this last, 40 of it is the internal 40C/W, 110 is the thermal resistance of the case to air. If you were able to reduce the case-to-ambient by attaching a heat sink, it would do nothing at all to reduce the internal resistance.

Note that TO-92 plastic packages have about the same thermal resistance to ambient as metal - 150C/W. A plastic 8 pin dip has junction to case of about 45-50, and junction to air of 100 to 150. In fact, the thermal resistance of the plastic is high enough that a lot of the heat is conducted out through the pins.

So connecting a heat sink to an opamp might conceivably cut the temperature rise to something like 40/100 to 40/150, to roughly 1/3 or 1/4 for a GREAT heatsink. Piddly little glue-on heat sinks aren't great. Here's a glue-on heat sink for a DIP 16: http://www.aavid.com/products/standard/501200b00000g
It's heat rise is 68C/W. Gluing one of these onto a 16 pin dip reduces the case-to-air resistance to 68C/W, but doesn't help the internal resistance at all. A similar heat sink for a DIP-8 opamp would be higher - it's tempting to say 138C/W. But-but-but the resistance from chip to air in the IC is already down at about 110C/W - how can a heat sink make it worse?

It probably doesn't, and that's because not all that much of the heat comes out through the plastic. Probably most of the heat is conducted out through the metal pins. They're not a great thermal path, but they're much better than filled epoxy. Probably the heat sink glued to the epoxy is in parallel with only the epoxy part, and lowers the total thermal resistance a little.

A little of what? In a fuzz face, the second transistor is commonly biased to about half the supply, more or less. So it's conducting 4.5V/8.6K = 523uA - ish. That makes power about 4V*523uA, or 2.09mW. (note that these are broad generalizations of the operating conditions!)

The first transistor has its collector sitting at about a volt, and flowing through it 8V/33K = 242uA, less the base current of the first device, about 5uA for a gain of 100, or 237uA. It's power is about 237uW.

The temp rise you'd expect from internal dissipation on the second transistor is 2.09mW times 150C/W = 0.3C. For the first transistor, it's - um, smaller.    My best guess is that with only 0.3C from internal heating, that ambient temperature changes of 30C or more make more difference - even if my guesses about operating temps were off by 3:1 or so.

Lets look at opamps. The TL072 has a standing current of typically 1.4, max 2.5ma per amplifier. So with 9V across the power supply, each amplifier dissipates 9*0.0025 = 22.5mW, or 45mW for two amplifiers. It's internal rise from its bias is then about 0.045*150 = 6.7C. The change in internal conditions is designed to be compensated, and is best expressed as a drift in input offset voltage with temperature. The datasheet obligingly states this, as temperature stability is a virtue in opamps. For the 072, it's 18uV/C. So we can expect drift on the order of 18uV*6.7C = 122uV.

If it drives a load, it will increase dissipation by the amount of internal heating needed to drive the load. The maximum load they specify for is 2K, and +/-10V into 2K. This is 14V rms, and an output power of 98mW. Of this, we can guess that no more than half as much will be dissipated inside, so call it 50mW. That makes internal heating come up to 14C or so, and input offset drift come up to 257uV. There are many charts of the changes with temperature of what the thing does, with the theme that "it doesn't change much".

But I'm running on. Opamps are designed to not change much with changing temperatures. We had some early tube and discrete opamps in analog computers back at school. Their output would change wildly if you touched a part or didn't let them warm up for an hour or two before you ran your tests. Eliminating this dependency has been one of those common themes in opamp design for a good half-century.

[Mark Hammer]

Delightful reply.  Many thanks.

In fact, the thermal resistance of the plastic is high enough that a lot of the heat is conducted out through the pins. .........

.......It probably doesn't, and that's because not all that much of the heat comes out through the plastic. Probably most of the heat is conducted out through the metal pins. They're not a great thermal path, but they're much better than filled epoxy.

Some older power-amp chips, like the PA237, have built in heat "fins" that protrude out of the epoxy.  Still others, like the LM380, have pins that don't harbour any electronic function, but which users are instructed to solder to a large ground plane, expressly for the purpose of dissipating heat.

I gather that many of the effects circuits that use LM386s (and here I am explicitly NOT including circuits for driving low-impedance reverb pans) are feeding loads that are in the non-problematic range.  At the same time, chips like the 386 do come in several power ratings (from 1 to 4 in the 386's case), and I suspect that one can potentially, and safely, up the output of a lower-rating one via a suitable heatsink, keeping your distinction between thermal resistance of the epoxy sitting between the silicon and the heatsink atop the chip in mind.  I suppose it might not be a bad idea for the pad that the output pin is soldered to to be oversized.

So, is the epoxy thermal resistance of flatpak SMD chips lower than that of traditional DIPs?  If so, one would think that SMD parts are more amenable to use of heatsinking.  On the other hand, if a chip lays flush against a PCB (as opposed to elevated by legs, in the case of DIPs and SIPs), how much heat dissipation would any additional heatsink accomplish, compared to what is already being done by the board itself, by being in physical contact with the underside of the chip?

[R.G.]

Some older power-amp chips, like the PA237, have built in heat "fins" that protrude out of the epoxy.  Still others, like the LM380, have pins that don't harbour any electronic function, but which users are instructed to solder to a large ground plane, expressly for the purpose of dissipating heat.

In fact, some more modern high-power surface-mount chips have copper thermal pads on the bottom side of the chip. This is to be soldered to a matching pad on the PCB by reflow soldering when the chip is mounted, so the heat can be conducted out from the bottom side of the chip to the metal pad, through the solder, and into the ground plane to get it out of the chip. The trick with dissipating heat is to get it from where it's generated and spread it out without incurring a lot of temperature rise to force the heat out.

Something most people don't appreciate is that the heating problem is especially acute for spaceships. They're surrounded by a high-quality vacuum, so there is *zero* conductive or convective heat transfer. Heat transfer is entirely by radiation. Spaceships have to have radiator panels that let the heat radiate away in the form of IR or they'll quickly become high quality heating ovens. While space is indeed very cold, it's also a very high thermal resistance, so unless you worry a lot about how to get rid of heat, first you'll fry, then you'll die, then you'll freeze.

I gather that many of the effects circuits that use LM386s (and here I am explicitly NOT including circuits for driving low-impedance reverb pans) are feeding loads that are in the non-problematic range.  At the same time, chips like the 386 do come in several power ratings (from 1 to 4 in the 386's case), and I suspect that one can potentially, and safely, up the output of a lower-rating one via a suitable heatsink, keeping your distinction between thermal resistance of the epoxy sitting between the silicon and the heatsink atop the chip in mind.  I suppose it might not be a bad idea for the pad that the output pin is soldered to to be oversized.

Correct. And companies do pull special tricks to make packages get rid of heat better. They own everything inside the epoxy covering, so they can do things like doing metal heat spreader plates, wider plates before the pins, and so on.

The real bottom line on thermal resistance is that, like the inside of an electrical resistor, the total resistance is (rho)*l/A, where rho is the resistivity of the material. l/A is the form factor of the block of material. It makes sense that the wider the area over which heat is spread, the less temperature is needed to force it through. Likewise the longer (l) the distance through the resistive material, the higher the resistance to heat passage. Rho is the intrinsic resistance of the material. Copper, silver, and aluminum have remarkably low rho, although much worse than the reining champion for solid materials, diamond. Diamond, silver, copper, and aluminum, in that order are good as heat spreaders. Yes, NASA has used diamond heat spreaders. You want heat transfer to be through very wide, big-area passages, and very, very thin layers till it gets to the material layer that will carry the heat away. This is why heat sinks have fins - they want to get as much surface area as possible with the smallest amount of material that has to conduct the heat and incur a temperature rise to do it.

So, is the epoxy thermal resistance of flatpak SMD chips lower than that of traditional DIPs?

Good question. In the LM386 case, from the datasheet, it's "no". They list the SO and MSOP as being worse than the DIP package. You'd think that the thinner the epoxy, the lower the Tr would be. But heat transfer is incredibly full of special cases, and very, very complicated indeed when you get past the vastly oversimplified Ohm's law analogy.

If so, one would think that SMD parts are more amenable to use of heatsinking.  On the other hand, if a chip lays flush against a PCB (as opposed to elevated by legs, in the case of DIPs and SIPs), how much heat dissipation would any additional heatsink accomplish, compared to what is already being done by the board itself, by being in physical contact with the underside of the chip?

It's very hard to answer that without looking at the actual chip itself and how its package is mounted. There are some MOSFETs and bipolars that are like short-tab TO-220s that are intended to be soldered to flat PCB areas. Solder isn't a great heat conductor, but it's many thousands of times better than the air gaps that would be there if they were not soldered. Heat sink goo is, again, not a great heat conductor, but it is thousands of times better than air, and is an electrical insulator, where that is needed.

Simple heat flow stuff is easy to estimate by the Ohm's law analogy. Real, true, no-fooling heat flow design can need rafts of super computers or concocting and measuring actual models.

[PRR]

> TL072 has a standing current of typically 1.4, max 2.5ma per amplifier. So with 9V across the power supply, each amplifier dissipates 9*0.0025 = 22.5mW

Worst-case dynamic heat of a TL072 at 9V, driving a dead-short, is OTOO 45mW. This on top of the 23mW standing power, so 68mW.

Yes, 5532 can get hot working hard, at higher supply voltages. Possibly OTOO 800mW with 40V supply and 600 Ohm load. With classic op-amps it is hard to make anywhere near such heat with 9V.

The low-volt power chips can get warm. A large class of them are, or are derived from, ADSL drivers. In ADSL work they run a large part of maximum signal for long periods of time. They should also hold their parameters from a cold-start (when line calibration is done) until full-hot.

> what would a heatsink inside a small unvented box do anyway?

It is a series-string of thermal resistors. The biggest resistor is the bottleneck. In thermal transfer, small surface is big resistance.

Die to case is very small area but it is solid-to-solid.

Case to air is small area and now one side is air, very poor thermals. Say 0.1 square inches. Or 0.5 sq.in. if we use a heatsink and the fins are fully effective.

This air has to move to the box surface, an uncertain resistance, though convection can be pretty effective.

From air to box inside.

Box inside to box outside (solid, often metal).

Box outside to The Whole Atmosphere.

While there are two air-box transfers, the surface area is likely to be 5-10 square inches. Far larger than the 0.1-0.5 square inches of chip or chip-sink.

If heat is a problem, then heatsinking the chip inside the box is helpful.

(Gooing the chip directly to the inside of the box is far more effective; and often done on Power amplifiers. Avoids a metal-air-metal resistance, and hides the heatsink spreading cost in the case-designer's budget.)

I say again-- there's no reason to think "gear sounds better" at a temperature suitable for humans. Chips ain't people; aren't even the same chemistry. Yes, we pick electronic stuffs because it is happy at "our" temps.... when we can. Recall a major branch of electronics which does not start working until red-heat-- vacuum tubes.

[Mark Hammer]

Thanks, gents.  Always a pleasure to have such thorough and thoughtful explanations of things many of us all too often think of in unrealistically simplistic terms.

You want heat transfer to be through very wide, big-area passages, and very, very thin layers till it gets to the material layer that will carry the heat away. This is why heat sinks have fins - they want to get as much surface area as possible with the smallest amount of material that has to conduct the heat and incur a temperature rise to do it.

The school I initially attended for my M.Sc. had our department, and several others, situated in some temporary buildings that, viewed from above, were designed EXACTLY like a heatsink.  Lots of poorly-insulated "fins" extending from a common narrow corridor.  You had to set the thermostat to 80 all day to even dream of getting it up to 65.

[PRR]

Your classrooms were the chip in a box problem. The heat radiator was a couple square feet. The "box" area was hundreds of square feet. Adding surface area to the heater would have increased heat flow.

[Mark Hammer]

If memory serves, the heater was a baseboard under the window....strategically situated to foster the most heat loss.

[PRR]

Radiation under windows and on outside walls is Standard Practice.

If you don't, you have a hot-side cool-side room which is unconfortable and can be drafty.

Of course with all that new-born heat going out the window, high fuel costs, the landlord turns-down the thermostat so it's cold all over.

We have "wrong practice" in our add-on living room. To avoid extending an over-long project I just cut holes in the existing hot-air ducts in the main house to throw into the new room. It isn't as bad as it could be, because the outside walls are obsessively insulated and the windows are decently insulated. Also the oversized furnace runs only 20% of the time so the "hot side" is momentary, mostly the room runs on stored heat in the drywall. (It is astonishing how much heat is stored in drywall and my 2" floor, especially compared to the relatively small heat this well-insulated house takes on average.)