cooling a very hot chip. And how many volts can a 386 take?

[brett]

Hi
Despite it being a simple question, I couldn't find an easy answer to this on the web...

I'm considering building an over-voltage, over-loaded LM386 amp for giggles and as a challenge to see how far I could get.  To stop the power junctions melting, I'll need plenty of cooling. Maybe with ice or a peltier (heat-pump) device?
Anyway, my question is - Is it worthwhile connecting heatsinks to the legs of the device?  Do the legs provide a path of low thermal resistance to the power junctions inside? Or should I just chill the body?

The junction-case resistance is 37C/W, so cooling the case to 0C would allow for 4W of output before melting, which would be great.

Also, does anyone know what determines the maximum supply voltage for a 386? Will an LM386-4 (rated to 22V) survive at around 40V -if the internals are cool?
thanks for any help

[Nasse]

Years ago I designed LM380 pcb. Only few components, less than handful. I did read all datasheeds I could get (before internet). Never got it working, totally solid rock silent, did two boards for stereo. Can´t figure what went wrong. Still have the boards somewhere...

But there was something said in the datasheet, that soldering direct to pcb was recommended, and they suggested that areas on pcb do cool the chip, and you should solder all unused pins too, but can´t remember if 386 datasheet suggested such

But I fried a 386 like amp twice with just 9 volts supply (discontinued Philips TAA300 chip) it was round metal case and I had a small recommended heatsink on the case

Anybody remember TDA2020 power amp? It was a d.i.l. chip, okay the top of the chip was metal and you bolted a heatsink over the chip... Perhaps you could mill the 386 top very smooth (not too much) and put some heat conducting paste and small heatsink on it... or sink the 386 pcb in oil can... just some silly ideas not much real knowledge. But I believe if the chip is not designed for external extra cooling all you do wont help much if nothing (double power gets just 3dB). Been wondering what my 2x12" speakers inductance with huge magnets kicked back to the chip when I punished those small amps to death with half dead 9 volt battery. Could you get more if you have a steep lowpass filter in front (lows take most power)

Bought http://www.musikding.de/product_info.php/language/en/info/p3147_Little-Rebel---Verstaerker-Bausatz.html but have not soldered yet

[Mike Burgundy]

Cooling is one thing (liquid nitrogen!), overvoltage another. It's not like a tube that can often take a LOT of overvoltage before electricity will arc through vaccuum to where you don't want it. No idea there. If you can find a safe way to stick this thing in a bucket of N2 it might work, although stuff seeping in from the air might cause conductivity problems. Perhaps it's a good idea to take the best real-world solution and scrounge up a sealed nitrogen processor cooler and max that.

[gritz]

While I fully endorse the torturing of all electronic components in the name of science I have a feeling that your local 386 stockist is going to see a lot of you! 

In no particular order (and in no way wishing to dampen your enthusiasm)

Your °C/Watt thing doesn't refer to the output power, but relates to the current through*volts across the device (same as it does with a resistor, or LED etc). Have a google for power transfer (and be prepared to be bamboozled by an overly complicated Wikipedia entry copy/pasted from a degree-level paper by a spotty student that didn't understand it either!)

I've never looked inside a 386, but it's likely that the actual die is just joined to the legs by super thin gold wires (power stuff like voltage regulators have the die in thermal contact with a tab etc. that can be attached to a heatsink) so in your 386 it's likely that the legs only suck heat from the die via the plastic package. That said, the bigger the copper traces on the pcb the more heat the IC can lose through it's legs, up to a point. That's why it's sometimes. recommended to fix these things direct to the board, rather than in a socket.

What decides maximum voltage? Maybe something as simple as the reverse voltage ratings of the semiconductor junctions. Remember too that any constant current sources on the chip will generate more heat with higher voltages across 'em. Also junction forward voltages drop with temperature and may cause thermal runaway somewhere.

Short of sticking a 386 on a board with flying leads and dunking it in a low viscosity, non-flammable, non conductive liquid I can't think of a way to cool one that doesn't involve a bunch of mechanical faff. Plus, deep thermal cycling of electronics can also lead to premature failure.You could attach one to a heatsink (dead bug style) and solder fat single core wires to each leg (to maximise dissipation), but it would be a pita to replace when it got fried.

[Gurner]

My  tuppence  worth (i.e. 2 pence, or 3 cents worth at today's exhcange rates)   ....if you want to push an LM386, don't go higher than the rated max voltage spec (well not much higher, as there's normally a little leeway), but take the load down....so instead of an 8 Ohm speaker, try a couple of 4 ohm speakers in parallel (2 ohms)...that'll make the littel bugger break into a sweat!

[R.G.]

Despite it being a simple question, I couldn't find an easy answer to this on the web...

It's worth considering that if there isn't an answer at least opined on the web that there may not be an easy answer. Just a thought.  

I'm considering building an over-voltage, over-loaded LM386 amp for giggles and as a challenge to see how far I could get.  To stop the power junctions melting, I'll need plenty of cooling. Maybe with ice or a peltier (heat-pump) device?

In the dual inline package, the problem is not what you cool it with, it's the thermal resistance getting the heat out, as you suspect.

Anyway, my question is - Is it worthwhile connecting heatsinks to the legs of the device?  Do the legs provide a path of low thermal resistance to the power junctions inside? Or should I just chill the body?

The legs are the primary path for heat to leave. The epoxy case is several orders of magnitude more insulating. Soldering a copper fin to each leg would improve heat removal substantially.

The junction-case resistance is 37C/W, so cooling the case to 0C would allow for 4W of output before melting, which would be great.

It is quite difficult to do this as a practical matter. At a minimum, there is another thermal interface at the surface of the case to whatever cools it. This will add, at a guess, another 1 to 2 C/W, best case.

Also, does anyone know what determines the maximum supply voltage for a 386? Will an LM386-4 (rated to 22V) survive at around 40V -if the internals are cool?

All semiconductor junction processes have a specific voltage limitation. They are designed for and manufactured for a certain breakdown voltage. Since this is a primary specification number, the manufacturers spend a great deal of effort to make sure that what they ship meets the spec. However, since getting performance in terms of current and speed generally needs lower voltage diffusion junctions, there is a competition between voltage and power or voltage and speed, or both.

A logical outgrowth of this situation is that there isn't much leeway on the maximum voltage spec. A 22V specified part might live at 24V, or might live a while at 24V. But at 40V, nearly twice the maximum spec, it's not likely to survive (period!) because whether or not the chip is generating any heat in operation, the voltage is making all of the internal junctions leak so much that they form local hot spots which get even hotter than operation would. And then it dies. Junction leakage and hot spot formation in doped silicon junctions is a thermal runaway process. Once it starts, it self reinforces.

You might make it to 25, 27, or 27V. The next supposedly identical chip might fail at 28. The next at 23. Depends on what the doping ovens were running that day and whether there was enough arsene in the process atmosphere. Probably phase of the moon, too.

[brett]

Thanks everyone. Great answers.
That the legs (thermally conductive metal) are important makes sense, but I've never seen that written down before.
Bummer about the maximum voltage.
I looked at the datasheet again, and the maximum voltage swing falls at low loads (roughly, the load in ohms should be almost equal to the supply voltage). So maybe 22V into a 16 ohm load is the way to go.
The datasheet doesn't go to 22V, but it looks like there'd be roughly a +/-8V swing, which is ~ 5V RMS, and 5/16 A into 16 ohms, which gives........... about 1.5 W.

Glass half full says that's almost 4 times the power available at 9V (which many people find acceptable).
Glass half empty says a 1.5W amp is too wimpy, and sure won't be seen on AC/DC's next tour.

Thanks again.

[gritz]

If you're dead-set on the 386 you could try running two in bridging mode (one to each speaker terminal running in antiphase to each other). This effectively makes each 386 "see" half the speaker's impedance, so I wouldn't try it with anything less than a 16 ohm speaker.

It *may* be possible to stack 386s in parallel with each output connected to a common point via current sharing resistors (maybe 1R? not sure...).

All this is pure conjecture, so don't blame me for any appearance of the magic smoke!

[R.G.]

Two ways to look at this:
1- how far can I push (specifically) an LM386 before Sudden Death Overtime?

2 - how much power can I get out of a chip in X space/price/powersupply/yada/yada?

For the first, just turn up the juice and load til it becomes a Light Emitting IC or Smoke Emitting IC.

For the second, as the machinists, mechanics, and carpenters say, don't break it, get a bigger hammer.  There is a slew of other power chips and boosted chips available.

[Paul Marossy]

Can I ask a dumb question? What is the point of wanting to operate an LM386 at twice its maximum voltage anyway?!

[R.G.]

Can I ask a dumb question? What is the point of wanting to operate an LM386 at twice its maximum voltage anyway?!

Not a dumb question. He did say:
"... for giggles and as a challenge to see how far I could get" which I consider a valid if not noble goal - as long as you're willing to say "Hmmph. So that's all it'll do. Ah, well. Now where did I put my beer?"

[Paul Marossy]

Can I ask a dumb question? What is the point of wanting to operate an LM386 at twice its maximum voltage anyway?!

Not a dumb question. He did say:
"... for giggles and as a challenge to see how far I could get" which I consider a valid if not noble goal - as long as you're willing to say "Hmmph. So that's all it'll do. Ah, well. Now where did I put my beer?"

I could see if it were maybe trying to run a power tube at its absolute limits or something, but a solid state device will just blow up. I guess there's the "I saw it with my own eyes" factor, though...  

[thedefog]

I think I've learned more from R.G. on this message board alone than most of what I've read elsewhere on the net.

[PRR]

LM377 rated 28V run at 34V dies every 10 years. My guess is electromigration pulls metalization between on-die traces until it quits.

'386 being basically a 15V part, I have to suspect the "22V" part is already very closely specced and has "no" leeway. 40V is probably instant death. 25V may be death in days (or less).

But the bigger issue: the output transistors in '386 are quite tiny. Current is proportional to area (the 200A starter in your car uses large-area wire, 0.050A pedal-power uses small-area wire). Copper just gets hot and can be abused for short periods. Semiconductors actually "choke up" like a freeway at rush-hour at some current density. In '386 this is about 0.45 Amps peak output. This foils the low-Z loading trick we use on less teeny amps.

Taking 20+V supply, 10V peak, and 0.45A peak, optimum load is 22 ohms and sine power output is 2 Watts.

The LM386 die is on the negative supply pin. Solder large copper to that. If that is not sick enough for you, go ahead and file the top and bottom very flat and sink those surfaces also. However I don't think heat is the biggest problem. It won't take voltage and it won't pass curent.

Look what the '386 was designed for. National had the LM380, a fine 5 Watt chip. But at the time there were millions of _cheap_ battery pocket-radios with 0.1W-0.4W power from a few transistors and a transformer. National thought they could price a very cut-down chip cheaper than the transformer and wipe-up a vast market. It had to be really cheap though. Which means minimum voltages and inky-dinky output devices. It's an OK 0.4W chip. It's wasteful as a 0.1W chip. It will do over 1W but things have to be just-right, no allowance for line or battery or load variation.

> there's the "I saw it with my own eyes" factor, though...  

If not posted on YouTube, with steady closeups of all connections and meters, it didn't happen.

[gritz]

I think it's easy to forget that we're talking about human nature here. A person spend thousands on e.g. strapping a turbocharger and nitrous oxide onto an old Pinto. His friends will say "stop wasting money on that old heap and buy a faster car" and "it'll grenade", but he ignores them and ignores the wisdom of those who have taken that route before. Of course, it all ends with a holed piston, or a broken conrod. So he tows the car back to his garage and starts again with lower compression forged pistons, forged rods etc and the cycle of wrench / wreck / repeat starts until nirvana is attained.

It's how some people roll.

386s are cheaper than pistons, but when they die they do so quietly and without fuss...

The defence rests. 

[Paul Marossy]

there's the "I saw it with my own eyes" factor, though...  

If not posted on YouTube, with steady closeups of all connections and meters, it didn't happen.

Ha ha, YouTube IS reality. 

[Paul Marossy]

386s are cheaper than pistons, but when they die they do so quietly and without fuss...

I'd be willing to bet that they will make a popping noise after they blow up when you put 40V on them. 

[Mark Hammer]

FWIW, I have a nice little heat sink and some thermal goop on my LM380-based battery-powered amp (run into an 8R load at around 11V when the rechargeable C-cells are fully charged).  Pics of it can be found here: http://hammer.ampage.org/files/Miniamp.zip

One of the difficulties of chips like the 386 is that they are small and low profile.  So affixing some sort of heat sink will obstruct the placement of very many components within a reasonable radius.  Not only would you need the heatsink itself, but you'd have to allot space for whatever means you're going to use to attach it to the board.  Of course, few people buy/use a 386 in order to have a bigger footprint.

The other thing not mentioned yet is how the gain loop (pins 1-8) are going to be used.  Most folks here generally plunk the suggested 10uf cap in there to goose gain up to 200x from the stock 20x.  More gain = trying to pass more current through those poor little transistors.