Kelvin SMD connection

If it was mine, I apologize. Although, I think my customer would have complained before now.

I think the requirement of contacting a "non-current" carrying part of the resistor termination is a specious requirement. The little trace may not be contacting a "non-current" carrying part of the resistor termination, but the part it does contact has to contact a current carrying part somewhere. So how does this reduce the error? It just means you are measuring the voltage of an unknown location on the resistor termination.

I think I would bring the trace to the pad from the center of the chip to contact the pad from the inner portion. I expect that would provide the least error from the pad and solder while preserving the assembly characteristics.

Better yet would be to use a resistor package that is designed for a Kelvin termination.

WOW, that's the worst layout I've ever seen. With predictable results. For shame it was found in an LT datasheet.

Some other article (which I believe was even from LT, come to think of it) goes through different connect methods. The one where fine traces enter the inside (underneath, interior, facing) edges of the pads (with the high current path connecting to the ends) was found to be the best for typical resistors.

(Another "option" that may come to mind is via-in-pad, but that's bad for soldering, and actually not as well controlled.)

As Rick says... if you're so desperate for precision, buy the four terminal part. It'll have far better stability than any two terminal part, besides (for one project, I spec'd Vishay Foil shunts, that are milliohms, 0.1% and 15ppm/C).

Tim

-- Seven Transistor Labs Electrical Engineering Consultation Website:

Not the best parts that they sell - 0.1% and 15ppm/C are pretty much standard E96 thin film resistor levels. Farnell used to sell a variety of non-Vishay parts that do that well - they seem to have fixed on Panasonic at the moment. Vishay can do better.

Milliohm shunts are more specialised, and pretty much have to be four terminal to work - even so 0.1% and 15ppm/C don't sound exceptional.

I don't know who came before. Apparently it's been through several sets of hands.

No worries--it's possible purchasing substituted a narrower-than-intended sense resistor. Now that no one has their own inventory or assembly, that happens a lot.

Cheers, James Arthur

Isn't what Kelvin is for?

That's what I usually do.

| | .---' '---. | | | | | | | | '-----..-----' ||

-------'|

--------'

--------.

-------.| || .-----''-----. | | | | | | | | '---. .---' | |

Why spend for more part than you need? Our job is to do more with less.

Cheers, James Arthur

It's not bad theoretically, just an implementation problem. Electrically fine.

I wonder where the vendors measure the resistance of the SMD parts they sell? That would determine the optimal sense-point.

Why go exotic when ordinary is good enough, used properly? Save the good stuff for the people who really need it.

It's just a few switchers and a handful of sense resistors,

Cheers, James Arthur

I've done things like that.

R4 is a neat IRC all-metal 1206 shunt resistor, 10 mohms. The board is a tiny 5 kilowatt laser driver.

A pickoff via in the middle of each big pad is another good way to do this. Magnetic coupling can be a problem, both 60 Hz and self-induced, and layout tricks can help there. Thermal symmetry can matter, too, to avoid thermoelectric transients.

Yep. The main thing is avoiding drop in the feeding trace, the rest is lagniappe.

Is via-in-pad cool? I thought that was a soldering no-no. I s'pose a tiny via wouldn't suck too much solder...

Just tapping off the inner faces of the footprint pads would seem pretty rugged, and gets a good sample of the voltage at the trace-resistor interface. I'll probably stick with that.

Cheers, James Arthur

Make the main pad narrower and add another Kelvin pad on the right side to make it symmetrical. ...Jim Thompson

It doesn't seem to be a problem. One small via in a huge pad won't slurp much solder. We pepper IC power pads with vias, to heat sink the chips, and that works fine too.

Symmetry is a good idea, but that also guarantees not using the resistor's full end termination width for the hi-current path. That might not matter. Two more effects are taking more room, and reducing heat-sinking--the latter might matter. In my current project one resistor dissipates 1W, worst-case.

I think John's via-in-pad on the inner aspect of each pad is probably easiest and best, or making the Kelvin connection there with a trace.

~ ~ | | .---' '---. | | | | | | | | '-----..-----' ||

-------'|

--------'

--------.

-------.| || .-----''-----. | | | | | | | | '---. .---' | | ~ ~

That's what I've always done. Works when the "pads" are just non-masked bits of copper pour for heat-sink, too.

Someone mentioned that one of the IC vendors has an app note with tests of the various Kelvin sensing layout schemes. That would be interesting reading.

Cheers, James Arthur

Trouble with a trace tap is you see the solder joint resistance which might be significant... I don't know. ...Jim Thompson

That's good, but maybe tighten up the pickoffs some...

That's symmetric. The current is nearly zero at the pickoff points. No vias, unless you want to add some for cooling.

The magnetic pickup area is near zero. The thermoelectrics should be good.

It's really easy to pick up 60 Hz from stray mag fields, when you're interested in PPMs of 10 millivolts or so.

Does the narrower main pad also reduce the current capability of the shunt? Reduce power dissipation? Unless the two pads together are the same width as the resistor, the resistor is going to float around (doing who knows what to the current/power capability).

Me too.

Indeed. More loses the contract so more soon becomes zero.

Via in pad is no big deal, particularly on pads this big. For smaller pads (.5mm pitch BGAs and such), they have to be filled. I can't see that there would be any problems with that big glop of paste.

It's a lot cheaper, too.

I don't know how it is if you're using spendy tools, but if you're using budget tools it's hard to convince the tool that you want a fat trace on part of a net (motor to resistor) and a thin one on another part (resistor to op-amp).

This sort of connection lets the tool manage trace widths, and reminds whoever (or whatever) that routes the board to actually make a Kelvin connection to the resistor.

Assuming an unlimited component budget, yes. The choices are much broader and less expensive if you don't go there.

With a mormal footprint, you can get ~ kelvin by tapping off between the pads, if you flip the resistor so that it's film element is face-down.

Most auto insertion tape configurations will prevent this, however.

RL