INA181 current sense amplifier overheating

[strulo]

I tried making a dummy load, where I use an INA181A1IDBVT current sense amplifier to measure across the sense resistor.
Datasheet: https://www.ti.com/lit/ds/symlink/ina181.pdf

But for some unknown reason, the INA181 gets hot, and current runs through it rather than the mosfet.

The output of the INA181 normally goes to an opamp which in turn controls the mosfet, but I've taken the opamp
out of circuit and shorted the mosfet gate to ground, and the problem persists, I've attached the simplified schematic of this.

I've measured voltages and resistance in the circuit an deverything looks fine, the grounds are actually ground, voltage on pin 6 etc.

I have tried with two different INA181's, I'm out of ideas and I don't know where I messed up.

It looks to me that I'm doing exactly like in fig44 of the the datasheet

[Siwastaja]

Didn't see the source voltage mentioned in your post, but I'm assuming it is below +26V, right?

Photo of the setup would be helpful.

[strulo]

Yeap, well below, I can't go much above 1V as the current limit on the power supply of 200mA kicks in and the IC starts warming up

[strulo]

Picture of the board + relevant section of the layout.

[Siwastaja]

Does the current start rising quite linearly starting from almost zero (edit: way below 0.7V)? If yes -> physical short

Does the current start rising quickly after approximately 0.7V? If yes -> current is flowing through semiconductor junctions in the IC.

Double and triple checked the pinout?

Try disconnecting the sense pins, just GND and +5V, how much current does it pull from +5V?

[Siwastaja]

Unrelated comment for future reference: this layout of paralleling standard resistors gives really poor accuracy on the current measurement. You can mostly calibrate it out unit-by-unit but the temperature coeff suffers, too. Instead, use a single larger SMD current sense resistor (2512 would be typical) where you can add Kelvin connections beneath the component, basically tracks connecting in the middle of the pad from below the component.

[Siwastaja]

Schematic shows GNDREF being connected to the power supply negative, but layout+photo doesn't seem to confirm this? Are they connected or not? Not being connected would be an obvious reason, the CM voltage could be anything then and cause current to flow through subtrate diodes.

[strulo]

Thanks for the suggestions!

Current was rising quickly but seemingly linear.


> Double and triple checked the pinout?

It turns out I'm an idiot 
Somewhere along the line I switched up the leads on the power supply so I was passing voltage in the opposite direction, which this circuit was never built to handle, so any number of things could have become messed up
But I know for sure they were correct at one point - so there must have been another issue before (probably short or crappy soldering somewhere, I've been taking this chip on and off a few times now) which got fixed and then
I didn't notice due to reversing the supply

---

> Schematic shows GNDREF being connected to the power supply negative, but layout+photo doesn't seem to confirm this?

They're connected, on the layout you can see a lttle rectangle on GNDLOAD jack where it's tied to the main ground plane

> Unrelated comment for future reference: this layout of paralleling standard resistors gives really poor accuracy on the current measurement.

Thanks for the tip! I sort of assumed that the resistor errors would average out, but I guess then that the way current flows through them + temp variation just screws it all up?




Thanks for the help and sorry for wasting time!

[Siwastaja]

In parallel resistors like that, tolerance of the resistors and the layout affects the current sharing. It's quite easy to make them share current equally enough so that none of the resistor overheats; but for measurement accuracy, it's a disaster if 55% of the current flows on the right side and only 45% on the left side.

Also when you tap the measurement line from the copper which carries current, the extra resistance of this copper is added in series, it's highly variable, and has a dramatically poor temp coeff of copper (0.4% per degC, compare to some 0.01%/degC of basic current sense resistors). So measure the shunt resistor pad voltage in a place where no more current flows, i.e., "after" the pad, this is, from under the component.

This being said, for an electric load that is supposed to convert all power into heat anyway, you can be quite wasteful in current sensing (i.e., drop significant voltage over the shunt), making such measurement flaws less dramatic for performance.

2512 SMD current sense resistors are easily/cheaply available up to 3W dissipation and there is a lot of space underneath for Kelvin connections. I prefer 1210 size for tighter loops whenever needed, though.

[strulo]

Good to know! I'll keep it mind for second revision

[exe]

it's a disaster if 55% of the current flows on the right side and only 45% on the left side.

I'm not sure I follow this part. All current-sensing resistors are in parallel, so they all at the same potential (assuming their resistance much higher than copper), and roughly at the same temperature. So, to me this layout looks fine. May be I'd do kelvin connection to the middle of this "combined" resistor, like here: https://en.wikipedia.org/wiki/Four-terminal_sensing#/media/File:Kelvin_connection_layout.png

[Siwastaja]

it's a disaster if 55% of the current flows on the right side and only 45% on the left side.

I'm not sure I follow this part. All current-sensing resistors are in parallel, so they all at the same potential (assuming their resistance much higher than copper), and roughly at the same temperature. So, to me this layout looks fine. May be I'd do kelvin connection to the middle of this "combined" resistor, like here: https://en.wikipedia.org/wiki/Four-terminal_sensing#/media/File:Kelvin_connection_layout.png

But the assumption of "much" higher R than copper is most of the time incorrect when it comes to current measurement. For example, if 1% accuracy is required, the R must be like 100x higher than the R of copper to make your assumption hold true. The OP's case likely hits right on this edge but changing Rsense to maybe just 100 mOhms would already make things more critical.

Resistor tolerance alone causes inequal current sharing, but given that 1% parts are cheaply available, this isn't the biggest problem. When small resistance values, like often in current sensing, are involved, the resistance of the copper tracks are significant portion of the total resistance. This means that current sharing is severely affected by some paths having longer path of copper. In this case, the middle resistors have lower total resistance. You can fiddle with the layout until cows come home, but the best thing to do is to remove the copper resistance from the equation by allowing one path for the current (single resistor) and not measure in the middle of the copper leading to that resistor, but use simple Kelvin sensing techniques.

In this case, 200mOhm is quite a lot for current sensing so even if you have an unstable (between units, between temperatures) 1 - 2 mOhm of copper resistance in series, this variation isn't disastrously bad, but think about it when the sense resistor is just 1mOhm which is often the case in high-efficiency designs measuring inductor or motor current!

Copper sheet resistance for 1oz copper is 0.5mOhm/sq. Counting squares roughly, middle resistors take 2 + 1 = 3 squares, the ones on the edges take some 3 + 4 = 7 squares, difference in copper resistance being some 2mOhm, causing additional 1% sharing error in top of the resistor tolerance (which is a matter of luck).

Kelvin connection into the middle of the "combined" resistor is of course somewhat better than what we have now, but unless there is strong reason to do so, a single part is both better and easier.

[exe]

In this case, 200mOhm is quite a lot for current sensing so even if you have an unstable (between units, between temperatures) 1 - 2 mOhm of copper resistance in series, this variation isn't disastrously bad

Yeah, I'm talking about this particular case. There are five resistors, 1Ohm each. I'd say with such high resistance current sharing should be very good, and I'd expect a thick copper pad not to be an issue (I can't prove this with numbers, so hand-waving here).

Speaking of single shunts. There are very few devices that are precise. While copper is very bad, through-hole power shunts are not perfect either. I have a power supply project where I'd like to get as accurate shunt reading as possible. It's not that trivial to find a sweet spot between shunt resistance, temperature rise, TCO, accuracy and price. Also, by changing one variable (say, shunt resistance) we change others. It seems lower-value shunts are made from different material than higher-value shunts.

I found this shunt for my project: https://nl.mouser.com/ProductDetail/IRC-TT-Electronics/OAR1R100JLF/?qs=e7YCq4YpkUe4JXAXnEMHIQ%3D%3D . Seems to be a good value for money to me (less than $1 for claimed 20ppm) , but I'm open for alternatives. I also considered a shunt in to-220 package so I could mount it on a heatsink and reduce temperature rise, but those are expensive.

[Siwastaja]

For simple SMD, I use some 1mOhm when high efficiency is important and accuracy within less than say 5% isn't, like torque limiting motors and protection of semiconductors. I try to use some 5-10mOhm when higher accuracy is needed, like lab supplies/measurement devices. Of course, unit-to-unit calibration may still be required. These resistance numbers are for approx. 20A of current.

This is with two-pad SMD parts, proper Kelvin sensing and single resistor. It isn't perfect, just maybe almost two orders of magnitude (definitely more than one order of magnitude) better than what shown by the OP.

When even higher accuracy is needed, special (expensive) 4-terminal parts need to be used.

The part you linked has 5% tolerance with suggests there may be unexpected drifts even if you calibrate it, and on such through hole part, Kelvin sensing is a tad difficult, I think a 2512 2W SMD part specified for 1% tolerance would do better job with similar price and no TH assembly step (assuming your design already uses SMT parts). Note that if you derate power, it's running with smaller dT, and slightly inferior TC can be accepted to produce same result. The part you linked needs a very low TC because it's designed to run very hot. I prefer a derated SMT part cooling into the ground plane below.