0.001 ohm 3W resistor at 30A gets too hot; impact on accuracy and alternatives?

[Alex Wolf]

If you place rectangular resistors parallel to each other, with individual current-sensing traces to each resistor, there will be uncontrolled currents through those current-sensing traces due to imbalance in the total resistance (resistor plus current traces) between the individual units.

Such a stack of resistors should be considered as one resistor; accordingly, only one pair of current-sensing traces is needed to their common contact pads. In this case, no problems will arise.

I'd like to see an equivalent circuit sketch + accuracy formulas of such a setup including the copper resistance (on which temperature has a large influence).

I'm sorry, but what you're asking for is extremely time-expensive in-depth research. Furthermore, high level of accuracy is clearly not required for a project with such low-precision components. If a high level of accuracy is required, this method is not suitable. So I can only offer a rough sketch of the PCB layout of what it should look like. The equivalent circuit looks very straight forward, just like the layout looks. The error is negligible at the given level of accuracy.



Edit: The resistors must be hand-selected and matched, that's obvious.

[magic]

Awesome, the outermost resistor that you are sensing has a significant fraction of mΩ more in series with it than the central resistors and the exact value of "significant fraction of mΩ" has a significant thermal coefficient too.

I doubt it would work well.

[Alex Wolf]

Awesome, the outermost resistor that you are sensing has a significant fraction of mΩ more in series with it than the central resistors and the exact value of "significant fraction of mΩ" has a significant thermal coefficient too.

The question is how significant this fraction will be, taking into account the fact that the operating temperature will be significantly lower, and along with it TCR becomes less significant.

Of course, you can route the current-sense traces from the central resistor if you need higher accuracy, as ahbushnell noted before.

[coppice]

"real metrology grade"
says 1% on top
    

Please, can you just go away now, we are dying out of laughter. I think you have totally lost all credibility by now.

Oh come on man, what's wrong with you? Can you just stop acting so childish, trying to prove to everyone that you are absolutely right and the superiority of your design, which is objectively wrong (against industry standards)? I’m not even talking to you, but you continue to try to start a flame with arguments to the individual. Just stop it, it's not your personal blog about how cool you are in engineering.

Oh come on man, what's wrong with you? Can you just stop acting so childish, trying to prove to everyone that you are absolutely right and the superiority of your design, which is objectively wrong (against industry standards)? Its pointless talking with you. You don't listen.

The labelling on that Bourn's 4-port resistor seems to indicate its 500 micro-ohms. The OP is using 1 milli-ohm. At 1 milli-ohm a large SMD resistor with good layout can match the performance of these 4 terminal resistors down to 0.1%. At 100 micro-ohms this become hard, and a well designed 4 port resistor has big advantages. There is, obviously, a sliding scale in between. A well laid out SMD resistor works well enough, that now people like Analog Devices (and me - I came to the same conclusions as AD, doing the same research for the same reasons, but they published their app. note first) have published information on the most successful layouts for high volume manufacture, resistor makers widely endorse those ideas. There is no magic in spending extra for a 4 port resistor when a little care in design can match it with smaller cheaper parts.

[Alex Wolf]

"real metrology grade"
says 1% on top
    

Please, can you just go away now, we are dying out of laughter. I think you have totally lost all credibility by now.

Oh come on man, what's wrong with you? Can you just stop acting so childish, trying to prove to everyone that you are absolutely right and the superiority of your design, which is objectively wrong (against industry standards)? I’m not even talking to you, but you continue to try to start a flame with arguments to the individual. Just stop it, it's not your personal blog about how cool you are in engineering.

Oh come on man, what's wrong with you? Can you just stop acting so childish, trying to prove to everyone that you are absolutely right and the superiority of your design, which is objectively wrong (against industry standards)? Its pointless talking with you. You don't listen.

The labelling on that Bourn's 4-port resistor seems to indicate its 500 micro-ohms. The OP is using 1 milli-ohm. At 1 milli-ohm a large SMD resistor with good layout can match the performance of these 4 terminal resistors down to 0.1%. At 100 micro-ohms this become hard, and a well designed 4 port resistor has big advantages. There is, obviously, a sliding scale in between. A well laid out SMD resistor works well enough, that now people like Analog Devices (and me - I came to the same conclusions as AD, doing the same research for the same reasons, but they published their app. note first) have published information on the most successful layouts for high volume manufacture, resistor makers widely endorse those ideas. There is no magic in spending extra for a 4 port resistor when a little care in design can match it with smaller cheaper parts.

Oh great, another one with a high ego and low qualifications. I hear you perfectly well, I just ignored you in order to avoid a useless argument, so as not to hurt your feelings and provoke a conflict. But it seems that you are looking for conflict as a way of self-affirmation, and perceive my kindness as weakness, as an easy target for your attacks. Therefore, I simply have no other choice but to expose you to your own mistakes in order to shame and stop you.

First of all, you both have no idea what a "real" current shunt is by industry standards. No, it's not just a low resistance resistor. Let me enlighten you: this is a very specially designed resistor specifically and exclusively for use in measurement circuits and instruments (a metrology grade "resistor" = a “real” current shunt). Its main difference is greatly increased power dissipation to keep temperature rise low and minimize the significance of TCR. You are mixing together the concepts of current limiting for overload protection, where high accuracy is not needed, and current measurement for regulation, where high accuracy is needed.

You also have no idea that TCR is a curve that is relatively straight only in a narrow range that ends much earlier than the operating temperature of your standard 2512 resistors at 1 W or more. The manufacturer doesn't even bother to give this information to you for regular resistors because they were never supposed to be used for accurate current measurement. The manufacturer is counting on you to be skilled enough in these matters to know this on your own and understand this from the list of applications.





Is your ego big enough to argue with the manufacturer?

And are you really going to pretend to be a fool who doesn’t know the definition of the word “example”? From the photo I provided, you can clearly see the markings, which you can use to find the datasheet and select exactly the resistance value you need. As for the value of resistance: in itself, it has nothing to do with the accuracy of measurements, unless you bring into the context the operating current and power dissipation, that is, the operating temperature. Moreover, a CRE2512 resistor will never match in performance to a proper current shunt like the CSS4J-4026 series. Why, you ask? See above: due to the ability to dissipate much more heat and operate at much lower temperatures, reducing the significance of TCR to a minimum. And no layout will help change this, because the hottest point is in the center of the resistor, and the thermal conductivity of its tiny leads is too low to remove heat quickly enough for any accurate measurements. It's funny that Siwastaja correctly noted earlier that for accurate measurements you need to derate the power rating (by about 10 times – note from me), but then he ignored himself and continued to argue with argumentum ad hominem.

If you don’t believe me, then teardown any large-scale device that is designed to accurately measure current, at least your multimeter, and you will see confirmation of my words right there (and you will never ever find there your CRE2512-resistor-like-design, if it’s not a Wun Hung Lo product). The OP expressed concern about the accuracy of the measurements, so it can be assumed that he needs it. Then he asked to name possible alternatives for this. So, this is exactly the alternative according to industry standards, that he needs. Why do you so mad about this? Maybe because you made the same mistake as the OP earlier in your designs, and now you don’t want to admit it even to yourself? It's none of my business and I really didn't care until you started dragging me into it.

Finally, regarding the 4 terminals. First of all, this hints to you that you have the proper current shunt for really accurate measurements. Secondly, even Siwastaja himself encountered and admitted a problem with two-terminal resistors, which he mentioned above:

Exactly, and the current distribution is hard to perfectly control even within the footprint of a single resistor, it depends how the part gets aligned during pick&place and how it moves during reflow.

And guess why? Simple: because it's the wrong design for accurate current measurements. PCB assembly machines were never intended to work with such high precision, unlike CNCs, which make proper 4-terminal current shunts that you can't screw up with a misaligned installation.

Any sufficiently advanced technology is indistinguishable from magic, especially if you don’t know, how it works.

[magic]

LMAO

[langwadt]

Awesome, the outermost resistor that you are sensing has a significant fraction of mΩ more in series with it than the central resistors and the exact value of "significant fraction of mΩ" has a significant thermal coefficient too.

I doubt it would work well.

yeh, could atleast turn things 90 degrees so each resistor has the same length of trace in series

[nali]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

@Alex Wolf large components only help with heat dissipation if there's i)space ii)budget and iii)a path for that heat to go, none of which we really know within the context of this thread.

[Alex Wolf]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

It’s funny that after three pages, no one noticed it before, including me. Well noted. 

@Alex Wolf large components only help with heat dissipation if there's i)space ii)budget and iii)a path for that heat to go, none of which we really know within the context of this thread.

Yes this is true. Moreover, we don’t even know what kind of precision the OP needs. I simply made educated guesses and offered a reliable and proven option that will definitely work for high accuracy, always works. And then this whole pointless argument began. Of course, depending on the specific requirements and limitations of the project, one or the other method will be optimal.

One addition: when it comes to larger pure metal resistors made of copper alloys with high thermal conductivity, (all other things being equal) they themselves are better at dissipating heat to the air, as well as better distributing heat throughout their bigger volume, reducing the final operating temperature.

[langwadt]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

It’s funny that after three pages, no one noticed it before, including me. Well noted. 

look at the picture of how it is soldered, the thermal reliefs doesn't matter

[Alex Wolf]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

It’s funny that after three pages, no one noticed it before, including me. Well noted. 

look at the picture of how it is soldered, the thermal reliefs doesn't matter

Yes, this is also true, with one caveat: it's definitely not the root cause of getting hot (the root cause is too much power for its size), but it definitely contributes significantly to the increase of operating temperature. The fact is that the thermal conductivity of copper is almost an order of magnitude higher than the thermal conductivity of typical solder.

[langwadt]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

It’s funny that after three pages, no one noticed it before, including me. Well noted. 

look at the picture of how it is soldered, the thermal reliefs doesn't matter

Yes, this is also true, with one caveat: it's definitely not the root cause of getting hot (the root cause is too much power for its size), but it definitely contributes significantly to the increase of operating temperature. The fact is that the thermal conductivity of copper is almost an order of magnitude higher than the thermal conductivity of typical solder.

the solder is orders of magnitude thicker than the copper

[trobbins]

Geez, no one seems to want to follow the heat flow path.

I'd estimate that >90% flows through the terminal legs and into the terminal blocks, especially if 4mm2 cable was being terminated at the terminals.  What the OP hasn't elaborated on was what wire was used during qualitative testing.

[Alex Wolf]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

It’s funny that after three pages, no one noticed it before, including me. Well noted. 

look at the picture of how it is soldered, the thermal reliefs doesn't matter

Yes, this is also true, with one caveat: it's definitely not the root cause of getting hot (the root cause is too much power for its size), but it definitely contributes significantly to the increase of operating temperature. The fact is that the thermal conductivity of copper is almost an order of magnitude higher than the thermal conductivity of typical solder.

the solder is orders of magnitude thicker than the copper

Which leads us to the conclusion that if the contact pads were solid, the total heat removal would be twice as good.

[langwadt]

Going back to the OP - He's made some efforts with polygons and vias to thermally connect top & bottom but then uses a footprint with thermal relief to mount the actual sense resistor. No wonder it's getting hot!

It’s funny that after three pages, no one noticed it before, including me. Well noted. 

look at the picture of how it is soldered, the thermal reliefs doesn't matter

Yes, this is also true, with one caveat: it's definitely not the root cause of getting hot (the root cause is too much power for its size), but it definitely contributes significantly to the increase of operating temperature. The fact is that the thermal conductivity of copper is almost an order of magnitude higher than the thermal conductivity of typical solder.

the solder is orders of magnitude thicker than the copper

Which leads us to the conclusion that if the contact pads were solid, the total heat removal would be twice as good.

do you really believe that?

[magic]

Of course it wouldn't; tin is ~6x less conductive than copper and easily 30x thicker than 35μ in the vicinity of the pin, so ~17% goes through copper. And OP's pattern still left maybe 33% of copper in place, so only 12% is lost.

The connector and its long pins are the bottleneck anyway.

[Alex Wolf]

do you really believe that?

Not really. This is a very rough approximation, more logic than mathematics. But yes, definitely, the larger the terms that make up the final value of thermal conductivity (to which both copper and solder contribute, and copper is approximately an order of magnitude greater), the higher the value of the total thermal conductivity. The exact value depends on the exact thickness of the copper and solder. But I think you already know that.

Edit: Clarified the wording.

[coppice]

(Attachment Link)
(Attachment Link)
(Attachment Link)

Is your ego big enough to argue with the manufacturer?

One resistor is considered suitable for power supplies and motor control. Low resistance values in those products basically means current measurement.

One resistor is considered good for current measurement. That will be applications like power supplies and motor control.

What's you point?

[nctnico]

Geez, no one seems to want to follow the heat flow path.

I'd estimate that >90% flows through the terminal legs and into the terminal blocks, especially if 4mm2 cable was being terminated at the terminals.

This has been discussed at length and the conclusion is that the terminal blocks the OP has used are not suitable at all and are very likely the cause of heating the resistor instead of cooling it.

[Alex Wolf]

One resistor is considered suitable for power supplies and motor control. Low resistance values in those products basically means current measurement.

One resistor is considered good for current measurement. That will be applications like power supplies and motor control.

What's you point?

Do you really not understand the difference, especially after I've already pointed it out twice, or is this just an attempt to suck another argument out of thin air to continue that pointless argument?

You are mixing together the concepts of current limiting for overload protection, where high accuracy is not needed, and current measurement for regulation, where high accuracy is needed.

Typical power supply design usually doesn't include any accurate current sensing requirements, since there is a good margin between the overload protection level and the power supply's destruction current (and usually this margin is an order of magnitude greater than the accuracy error).

There are bench power supplies that actually require accurate current measurement to accurately regulate the current, but this is a special case that requires a resistor of a higher grade (which is usually called just a current shunt, since this is its the one and only special purpose).

Yes, you can use a higher grade resistor for less demanding tasks, but not vice versa if you care about accuracy. And if you don't really care, our argument becomes even more pointless. 😐

[coppice]

One resistor is considered suitable for power supplies and motor control. Low resistance values in those products basically means current measurement.

One resistor is considered good for current measurement. That will be applications like power supplies and motor control.

What's you point?

Do you really not understand the difference, especially after I've already pointed it out twice, or is this just an attempt to suck another argument out of thin air to continue that pointless argument?

You are mixing together the concepts of current limiting for overload protection, where high accuracy is not needed, and current measurement for regulation, where high accuracy is needed.

Typical power supply design usually doesn't include any accurate current sensing requirements, since there is a good margin between the overload protection level and the power supply's destruction current (and usually this margin is an order of magnitude greater than the accuracy error).

There are bench power supplies that actually require accurate current measurement to accurately regulate the current, but this is a special case that requires a resistor of a higher grade (which is usually called just a current shunt, since this is its the one and only special purpose).

Yes, you can use a higher grade resistor for less demanding tasks, but not vice versa if you care about accuracy. And if you don't really care, our argument becomes even more pointless. 😐

You don't really argue. You're preferred move seems to be the ad hominem attack, from a position of complete ignorance of the other party's capabilities.

Do you not understand what a typical in system current sensing solution looks like? You write like someone in a metrology lab, rather than someone who designs equipment. The electronics connected to the shunt usually needs calibrating if high accuracy is needed, so the shunt might as well be calibrated too. What you need from the shunt is long term predictability, over the operating conditions the equipment will face. Why do shunts get so hot? Because people want to use a big resistance to simplify the electronics. How does it simplify the electronics? I two predominant ways. It helps with a limited CMRR and with the SNR. A lot of shunts now feed into a sigma-delta converter, which can be really good for the CMRR issue. A well designed, but cheap, switched cap front end sigma-delta converter usually has a worst case CMRR of 150dB or more. So, those are really helping to keep down the heat from the shunt. How much the SNR issue limits you depends on the application. If you need a very fast response, like cycle by cycle measurement in a switched mode system, you probably can't do much to smooth out the noise. In most of the applications where you need high accuracy you can do some smoothing. So, one good move is to look carefully at the electronics, and try to get the value of that shunt down to the minimum possible value. I see people with heat problems from 2 or 4 milli-ohm shunts where I could get them down to maybe 200 micro-ohms at no additional cost, and make their heat issue go away.

[Alex Wolf]

You don't really argue. You're preferred move seems to be the ad hominem attack, from a position of complete ignorance of the other party's capabilities.

Oh really? If you scroll up the topic, you can clearly see that I did not attack anyone, I was only defending myself from all the nonsense and personal attacks that were thrown at me, including from you, which I ignored and tried to nullify using humor until they cross the red line of arrogant rudeness. And your favorite type of attack is to repeat after me my well-founded accusations against the attackers, without supporting them with any justification. I've lost all interest in talking to you, it's hopeless in every sense. One piece of good advice in the end: if you don't have the capability to make an accurate measuring circuit (no matter for what reasons), you should not present your solution as an accurate measuring circuit, much less try to argue with those who know for sure that it is inaccurate (not even close to the worse meter on the market). Usually, people are annoyed by lies, especially blatant lies. Good luck, you'll need it.

[trobbins]

This has been discussed at length and the conclusion is that the terminal blocks the OP has used are not suitable at all and are very likely the cause of heating the resistor instead of cooling it.

Is that the conclusion based on the OP's qualitative 'gets excessively hot', and 'with a 30A current, it gets extremely hot to the point where I can't touch it for even a second', and 'But I wonder for 20A also the trace and the resistor are getting warm but not the screw connector body.'  Considering those vague observations of temperature rise, and no indication of what wire was terminating at the connector to act as imho the dominant heatsink, then I can't see how any conclusion could be made for a current up to the rating of the terminal block.  Sadly the OP has used a terminal block only rated for 30A, in an application that may require up to 45A, but again the OP has not provided any detail as to the time profile of current, as 30A, and even the 45A, could relate to short duration peaks - who knows!

[tszaboo]

You don't really argue. You're preferred move seems to be the ad hominem attack, from a position of complete ignorance of the other party's capabilities.

Oh really? If you scroll up the topic, you can clearly see that I did not attack anyone, I was only defending myself from all the nonsense and personal attacks that were thrown at me, including from you, which I ignored and tried to nullify using humor until they cross the red line of arrogant rudeness. And your favorite type of attack is to repeat after me my well-founded accusations against the attackers, without supporting them with any justification. I've lost all interest in talking to you, it's hopeless in every sense. One piece of good advice in the end: if you don't have the capability to make an accurate measuring circuit (no matter for what reasons), you should not present your solution as an accurate measuring circuit, much less try to argue with those who know for sure that it is inaccurate (not even close to the worse meter on the market). Usually, people are annoyed by lies, especially blatant lies. Good luck, you'll need it.

You are not wrong, those 4 terminal shunts are better than the usual 2512 and others.
But OP seems to not even own a thermocouple, so I question a little bit the attempt to make a serious measurement equipment.
I've used and designed in a lot of shunts during my career. Usually the selection is based on requirements, and we didn't see those, other than "it shall not get hot to the touch".

[temperance]

@ Alew Wolf

You also have no idea that TCR is a curve that is relatively straight only in a narrow range that ends much earlier than the operating temperature of your standard 2512 resistors at 1 W or more. The manufacturer doesn't even bother to give this information to you for regular resistors because they were never supposed to be used for accurate current measurement. The manufacturer is counting on you to be skilled enough in these matters to know this on your own and understand this from the list of applications

An other factor to consider is that the CTE of standard 2512 ceramic resistors doesn't really match with the CTE of FR4 material. Repeated thermal cycling will cause too much stress on the solder joints.