Are there any downsides to using smd high current shunts?

[teslinius]

I am designing a circuit that needs to be able to measure currents up to 25A, i'm just using a differential amplifier to amplifiy the voltage drop from the shunt, firstly i wanted to just a copper wire but i realized that it has way too low resistance, so i found some 2512 8 mohm 3w current shunts, i am planning to put 2 in parallel that i get 4 mohm resistance which will generate around 2.5w of heat which should heat them up too much right?, plus i will put copper plane and thermal vias around them also they are right next to the fan inlet so it's sucking heat away from them, are there any downsides that i don't know of?

link to shunts:
https://www.lcsc.com/product-detail/span-style-background-color-ff0-Current-span-Sense-Resistors-span-style-background-color-ff0-Shunt-span-Resistors_SART-Nanjing-Sart-Tech-SMA25A3FR008T_C2692031.html

[TimFox]

For high-current measurements, you should use a 4-terminal shunt resistor (which are available in SMT).
This is necessary to avoid errors due to the voltage drop across the PCB traces.
However, you cannot get the proper 4-terminal connection with two such resistors in parallel.
Check the datasheet for the resistor for thermal ratings.
A copper wire is ordinarily not a good shunt resistor, since copper has a higher temperature co-efficient than good resistor alloys.
The same vendor has a better part, although it costs about $0.70 US, or $0.46 US at qty 100:  4 m at nominal 2.5 W, with other lower resistances (and correspondingly less power at 25 A) at somewhat higher power.  Perhaps you can increase the gain on your differential amplifier.
https://www.lcsc.com/product-detail/Current-Sense-Resistors-Shunt-Resistors_Isabellenhuette-BVS-A-R004-1-0_C467243.html
https://datasheet.lcsc.com/lcsc/2105241238_Isabellenhuette-BVS-A-R004-1-0_C467243.pdf

[radiolistener]

are there any downsides that i don't know of?

Yes, it has much less peak current limit. For example, with usuall shunt you can put a pulse with 10-100 times more power than it's specification and it will survive. But such SMD resistor will be damaged even if power will be exceeded for a little value.

[perieanuo]

I am designing a circuit that needs to be able to measure currents up to 25A, i'm just using a differential amplifier to amplifiy the voltage drop from the shunt, firstly i wanted to just a copper wire but i realized that it has way too low resistance, so i found some 2512 8 mohm 3w current shunts, i am planning to put 2 in parallel that i get 4 mohm resistance which will generate around 2.5w of heat which should heat them up too much right?, plus i will put copper plane and thermal vias around them also they are right next to the fan inlet so it's sucking heat away from them, are there any downsides that i don't know of?

link to shunts:
https://www.lcsc.com/product-detail/span-style-background-color-ff0-Current-span-Sense-Resistors-span-style-background-color-ff0-Shunt-span-Resistors_SART-Nanjing-Sart-Tech-SMA25A3FR008T_C2692031.html

well i saw this 'technique' applied with success in some motor controllers for example.
yes, it works, but that's for you to decide if the error caused by not using 4-wire connection is ok for your client. for our clients at the time, it was fine if we had some little error in measured phase motor crt
evacuating heat depends on some facors: upto 25A, but how much 25A charge you expect? you see, that's for you to answer, as we know no schematic like this one will work with 100% charging time, so the loads you measure tells you if that peak causing overheat is acceptable. for most cases it can work if resistor power is doubled for example (to be clear, for x Watts dissipation your resistor is rated 2x watts). so some "2" security coef i use regularly works for this case. in fact depends of how secure your app is, you didn't specify. if it's medical app it's one thing, if it's measuring some DC motor current for moving some conveyor working usually at 60-80% speed, it's another case.
anyway, good practice (almost not-existing those days, use only one resistor 2 or 4-wire, choosing a security coef for his wattage -dunno how it's called in english sorry- regarding into your application specificity).
why i say those practiced are gone? when i was 'engineering' in some obscure R&D company in France i had some current motor protection boards returned from some client wanted more current limit (like triple) and the solution my boss imposed wad not changing the sense resistor with new calculated one but mounting another 2 resistors on top of the original one, same value>>triple crt. the rest of the pcb and layout were ok for the new current limit.
it worked? yes it worked. done for at least 200 pcb's. zero failed boards returned in our custody. end of story

[Siwastaja]

Paralleling shunt resistors makes Kelvin layout difficult or impossible and decreases accuracy. Use single device if at all possible.

If 2-3% accuracy is enough, a special 4-terminal resistor is not needed. The standard Kelvin layout (sense traces routed below the pad) as shown in numerous appnotes is OK.

There is no downside up to approx. 2W dissipation. If you can limit yourself to that power, definitely use SMT.

I would consider using a single 3mOhm part (1.9W) instead of two 8mOhm parts in parallel. Increase amplifier gain if necessary.

For really high-accuracy stuff, you'd need to compromise, increasing burden voltage to lower amplifier gain, use large through hole part with 4 terminals...

But single SMT part is OK up to some 2-3% accuracy and you can increase that by software calibration. Most of the error is fixed (not drifting) due to tiny details how solder flowed on the pads.

With 4-layer board with continuous ground plane below the shunt, like often used in power switcher designs, you don't need much derating, 2.5W from a 3W rated shunt is usually perfectly fine but I tend to limit myself to 2W is possible.. With 2-layer board especially with hot components nearby I'd consider derating the 3W part to 1.5W or even less. This is also a compromise in accuracy, you get more accuracy by running cooler due to lesser effect of temp coeff, but OTOH if you need to decrease burden voltage to make it run cooler, then you are increasing all other error sources. There is some sweet spot and that could be running 3W part at 2W.

2512 is a fine package for this purpose. Lot of room to route the Kelvin sense traces as well. I prefer 1210 in switchers whenever it's directly measuring the switching loop to due even less inductance. If it's measuring inductor current or even better, smoothed DC after capacitors, 2512 is more than fine inductance-wise.

[Kleinstein]

2 shunts in parallel is usually not a good idea - they would need to be very closely coupled. When in close proximity this would cause thermal derating. So 2x 3 W would no longer give you 6 W, more like 4 W.  Power ratings with SMD resistors are anyway often optimistic, valid as single part with lot of copper around. So one would real world more like derate the power anyway. In addition for shunt use, to get precision one would usually only use a small fraction of the power-rating.

Chances are a single 1 mOhm shunt could give better performance at 25 A.

A general problem with larger SMD form factors it, that they are sensitive to board stress and bending. Thermal and humidity related expansion of the PSB can also limit the accuracy.

[richard.cs]

Shunts in parallel with Kelvin connections is possible, make the Kelvin connections as normal to individual resistors and then resistively sum them with modest-value resistors (say 10 Ohms to 1 k). That ensures you get a good measure of the total current, separately you also need the careful layout to share the current evenly between shunts to get acceptable dissipation.

[Siwastaja]

Yes it's possible but usually not worth the hassle and extra cost in area and extra components. If 3W part in 2512 does not suffice, I'd say go for 4-terminal through hole part.

[Vovk_Z]

If there is a possibility to use through hole parts I would like to introduce OAR series (OAR1, OAR3, OAR5 - 1, 3 and 5 Watts rated respectively).

[connectTek]

https://www.allegromicro.com/en/products/sense/current-sensor-ics/zero-to-fifty-amp-integrated-conductor-sensor-ics

[teslinius]

Thanks for the replies, i have no problem using tht shunts just wanted to know if smd's can be used, this is just a hobby project an electronic load but i want to make it as accurate and professional as possible so i'll learn. Also one thing to note is that i will mostly use this load for currents up to 10A, up to 25A very rarely. I found these tht shunts which seem ok (https://lcsc.com/product-detail/Current-Sense-Resistors-Shunt-Resistors_Shenzhen-Yezhan-Elec-Y110-10-3-H_C695786.html), my only concern with lower resistance shunts is that i will have to increase opamp gain which also increases the risk that it will amplify noise from nearby components (serarated digital from analog circuit and watched very closely for that but you never know). A lot of you recommended me using a 4 terminal resistor but i couldn't find any and i'm not sure exactly how they work, will have a further look. It looks like the bk precision electronic load uses 2 shunts in parallel.

Also i have another question for voltage measurement, so if i want precise voltage measurements i need a precise adc voltage reference, precise resistors and low temperature coeficient ones?, what's the difference between normal resistors and resistor networks, are the networks just designed to be more precise and have low temperature coeficient compared to normal resistors?

[Doctorandus_P]

I'd think that 4 terminal resistors are overrated.
They sure do have some advantage if you want an accurate absolute resistance value, but chances are your gadget has to be calibrated anyway, and that makes stability more important then absolute value.

The 4-wire Kelvin connections are mandatory though and you have to take temperature effects into consideration. For such SMT resistors it's common to put the high current path on the outside of the resistors, and take the voltage sensing feedback from the center of the insides of the pads. This is also shown in the "recommended land pattern" on page 2 of the datasheet. I do like the solid copper pads inside the connections of the resistor you've chosen. For temperature effects, imagine what happens if the resistor heats up because of high currents. Copper has a relatively high temperature coefficient, and will therefore change in resistance and therefore in voltage drop. When you connect the feedback lines to the insides of the pads, the amount of copper between the "resistance element" and the "feedback sense lines" is minimized. It's probably less the 0.1mm long, and over the full width of the resistor. This is a quite minimal resistance. There is not much that can be improved here with a 4 terminal resistor. 

I do have some doubts about this resistor though.
The datasheet of your resistor specifies a "load life" of only 1000 hours at rated load, and nothing about long term stability. 25*25*0.004 = 2.5W and this is also quite a lot of power for such a small SMT resistor.

Also i have another question for voltage measurement, so if i want precise voltage measurements i need a precise adc voltage reference, precise resistors and low temperature coeficient ones?

No, not really.
As Davey Jones has said many times in his video's, stability is much more important then absolute accuracy. Static errors can quite easily be calibrated out, while drift can not, because it changes over time.
Temperature coefficient is important, because a low temperature coefficient reduces changes with temperature. But it really does not matter if your voltage reverence is 5.0000V or 5.2314V, as long as it stays put at the value it was initially.
With calibration you can get the total a few orders of magnitude more accurate than the initial tolerances of the parts.

[Siwastaja]

Doctorandus is spot on,

Constant errors are really easy to calibrate out. Get a reference current meter you can trust, generate and measure current at two different levels, one near minimum, one near maximum, fit a first-order linear curve with offset and gain.

But it's worth spending some money in low tempco parts because it's getting tedious to compensate that out. It's possible, yes, you can place a thermistor near the shunt, measure the temperature, then calibrate the thing at 2..3, maybe 4 different temperatures and build a lookup table, therefore changing calibration values on the fly. But it's a lot of work, if you can get a better lower tempco part by paying $0.50 more.

Noise is concern if you need both bandwidth and accuracy. If this is not inside switching loop, then you don't need much bandwidth, and should use low-BW shunt amplifier to begin with, and you can add some filtering.

[Kleinstein]

Also i have another question for voltage measurement, so if i want precise voltage measurements i need a precise adc voltage reference, precise resistors and low temperature coeficient ones?, what's the difference between normal resistors and resistor networks, are the networks just designed to be more precise and have low temperature coeficient compared to normal resistors?

Resistor networks come in different flavors. There are just cheap ones, that are essentially the same as the cheap seprate resistors, just multiple resistors in a case, e.g. as pullup for 8 digital lines.

There are also precision ones that have relative good matching in the TC and this way get a stable divider ratio, even of the absolute resistor values is not that stable.  Quite often the resistor ratio also has relative low tolerance.

For an electronic load one often has a relative large (high power) shunt, as the set current is variable. So one also wants reasonable performace at a small fraction of the maximum currents.