According to specs 3886 can disspate 125W max with no fancy stuff like that, so am not sure I am impressed.
Their image shows the LM3886TF (isolated package). This package has a thermal resistance from junction to heat sink of approx. 2 K/W. At 100 W dissipation, this means the semiconductor junction is 2*100 = 200 K = 200 ºC above the heat sink temperature. The thermal limiter in the LM3886 kicks in at about 165 ºC, which means the author claims the LM3886 is operating with a heat sink temperature of 165-200 = -35 ºC.
Maybe the author really did manage to keep the heat sink temperature at -35 ºC. Maybe they kept it submerged in liquid nitrogen or liquid CO2... I doubt that, though.
Others are free to believe what they want, but the laws of thermodynamics and other areas of physics relevant here have been rather well-established for the past few hundred years.
~Tom
Last edited:
You are right, physics and math don't lie, BUT just one small piece of information here is missing or unknown to you, and that is "thermal resistance" with this cooling method 🙂
Last edited:
You are correct that I don't have the thermal resistance from the die to the front of the package. I do, however, have common sense.
Go back a few pages in this thread and get the thermal conductivities for the various materials included in the IC packaging process.
The IC is connected to the DAP with thermally conductive epoxy. Between the DAP and the outside of the package is a thin layer of plastic. The total thermal resistance from the die to the heat sink through this thermal interface is 2 K/W.
To get from the die to the front of the package (to reach the clamping block), the thermal path is through several millimeters of plastic. This will have a significantly higher thermal resistance than the 2 K/W through the back of the package. Hence, the total thermal resistance, including through the clamping block, from the die to the heat sink will be approximately 2 K/W.
There is no way the clamping blocks will provide any meaningful reduction in the thermal resistance. The claims of the authors is false. Flat out false. There's no need for "time will tell". Their claims don't pass a simple sniff test and back-of-envelope calculation.
If you really want to minimize the thermal resistance, use the LM3886T (non-isolated) package. It is specified to have a thermal resistance of 1 K/W. Add a quality SilPad (1500ST would be my choice) and you're looking at about 1.05~1.1 K/W. Under these conditions, the heat sink will still need to be kept below 165-1.1*100 = 55 ºC, which for 100 W of dissipation is a challenge, but possible.
Of course, to make this possible, you will need this super duper highly proprietary patented clamping device. I happen to have one of those extremely rare devices in my lab. It's called ... are you ready for this? ... a MACHINE SCREW.
~Tom
Last edited:
What if you used the non-isolated 3886T, with no SilPad.
Connected directly to a copper 3" x 3"
[go 6" x 6" copper if you need to]
(whatever thickness you can find) [yikes electrically active: V-
]and then THIS was SILPAD to a big alum heatsink for elec - isolation
...
wait ... is making V- a big 3x3 copper plate, a bad thing ... 😱
tee-hee
Last edited:
The surface roughness of both preclude a good thermal flux transfer.
Even if one were to polish both surfaces to micron level finishes (optically flat give or take).
If you're going to go hot spreader, use that graphene material. Thermal conductivity out the wazoo..
The copper heat spreader can be 1/4 inch thick, and a square inch or two to relax the thermal requirements of the insulator between the copper and the sink.
I go into that a tad in the LA article.
jn
Depends on what you mean by "directly connected". If you mean mechanically attached, you'll need to do something about the surface roughness (as pointed to already). The most common way for dealing with this is to add a thin coat of thermal paste to fill the gaps due to the surface roughness.
Maybe if you soldered the IC package to the copper slug and then machined the copper slug flat, you could get a very slight improvement in thermal resistance. This would assume that you can solder the IC package to the copper slug without getting voids in the solder joint, which can be a bit of a challenge...
With the IC package electrically connected to the heat sink, you will need to isolate the heat sink from the chassis. Otherwise, you'll have the negative supply voltage present on the chassis, which is a recipe for disaster.
Aside from the "what-if" and general geekery, I'm not sure what you're intending to get out of all this work. The LM3886TF is capable of delivering the datasheet performance if provided with a reasonably sized heat sink. It works... Why make it any more complicated?
~Tom
All of these permutations to lower TJ seem a whole lot more complicated than a slightly larger heatsink and parallel lm3886 chip arrangement (or lm4780, although the former is better thermally).
In particular as even the isolated LM3886TF can provide the published performance when attached to a reasonably sized heat sink.
~Tom
I 'may' have to worry about these. I should have waited for the parallel-86, but was trying to get my project done over xmas with my bonus. Ran out of funds and then realized my replacement ribbons were lower impedance than the stock factory units. DC resistance is 3.1 Ohms. AC impedance I don't have an easy way to measure, but won't be much higher as going active. I'm on the limit current wise. But hey, nothing ventured etc. luckily I do not 'bang head' these days.
I'll give you that, Bill. 🙂
How low are you running these ribbons? (You might well be fine regardless)
How low are you running these ribbons? (You might well be fine regardless)
undecided. Apogee ran them at either 1KHz LR2 or 750Hz LR4. some have gone lower but not clear if anything to gain other than flappy ribbons going that way.
The amount of energy that you'll be pushing into these ribbons is pretty small, then. Isn't most of the spectral energy below 500 hz or so?
You've got the additional benefit that you're not driving all the losses in a crossover, too, so your watts go further (and so do your thermals, as result).
You've got the additional benefit that you're not driving all the losses in a crossover, too, so your watts go further (and so do your thermals, as result).
It's about 60:40 energy distribution with a 750Hz crossover. most things are stacking in my favour 🙂
The power spectral density of classical music tends to follow a 1/f slope as I recall. Rock music tends to be more like 1/(f^2). This means that at a given frequency the power is less at high frequency. However, the tweeter typically covers at least 10-15 kHz of bandwidth, hence, the total integrated power is actually higher for the tweeter than for the rest of the drivers.
I'm not sure I'd worry much about it, though. The MOD86 can deliver 60+ W into 4 ohm on +/-28 V rails, provided that you use a reasonably sized heat sink. This should be more than adequate to provide ear-splitting SPLs in a residential setting.
~Tom
I'm not sure I'd worry much about it, though. The MOD86 can deliver 60+ W into 4 ohm on +/-28 V rails, provided that you use a reasonably sized heat sink. This should be more than adequate to provide ear-splitting SPLs in a residential setting.
~Tom
Tom, are you sure about the thermally conductive epoxy die attach?
I do not recall the last time I saw power product epoxied to it's holder. Everything I've ever seen has been soldered or eutectic scrubbed to the base.
jn
The die is connected to the DAP with a method that provides very low thermal resistance. It has to be to survive the max dissipated power listed in the data sheet.
I don't know for sure if epoxy is used or if the die is attached with that scrubbing method you describe.
The point I was trying to make was that the thermal resistance from the die, through the DAP, to the heat sink will be significantly lower than that through the plastic to the clamping gizmo to the heat sink. Hence, the thermal resistance will be dominated by that from the die through the DAP to the heat sink, and the clamp has no meaningful effect.
I suppose we could dissolve the package in hot nitric acid and figure out which die attach method was used... 🙂
~Tom
- Status
- Not open for further replies.