0805 Kevlin sense connections

On Monday, January 18, 2021 at 7:51:46 PM UTC-5, snipped-for-privacy@yahoo.com wrote :

======

I don't have a way to view your ascii art. It's one of the casualties of t he Google groups "improved" user interface. They rip out multiple spaces.

Since I can't see what you are talking about, I didn't read the details of your attempt, but that is entirely practical. But the mention of "corners" concerns me. If the traces feeding current to the 0805 resistor are from the far ends of the pads, you can be confident there will be a relatively c onsistent connection to the resistor ends at the near edges of the pads and so a very good kelvin connection. But if you are talking about picking of f the voltage at the corner of the pads that may be leaving a bit of accura cy on the table. I would have my sense traces run between the pads and att ach at the center of the inner edges to minimize any impact of the solder j oint being unsymmetrical.

One web site says the pads for an 0805 resistor have 47 mils between the in ner edges. That's wide enough to run 9 mil track/space. Wikipedia shows what I'm talking about.

With the current flowing to the part from the long edges, the near edges ha ve a nearly kelvin connection to the component.

te:

=======

s.

the Google groups "improved" user interface. They rip out multiple spaces.

f your attempt, but that is entirely practical. But the mention of "corners " concerns me. If the traces feeding current to the 0805 resistor are from the far ends of the pads, you can be confident there will be a relatively c onsistent connection to the resistor ends at the near edges of the pads and so a very good kelvin connection. But if you are talking about picking off the voltage at the corner of the pads that may be leaving a bit of accurac y on the table. I would have my sense traces run between the pads and attac h at the center of the inner edges to minimize any impact of the solder joi nt being unsymmetrical.

inner edges. That's wide enough to run 9 mil track/space. Wikipedia shows w hat I'm talking about.

have a nearly kelvin connection to the component.

I nosed around a bit more and found another approach that helps to exclude the effect of the solder joint, but I'm not sure it would be easy to use on an 0805 part.

They split the pad into three "zones" with the current flowing on the two w ide ones along the edges. Then a trace wide pad is used to contact the res istor caps in the center with a separate connection. Unless the solder joi nt gets very ugly, this should work well.

How about

Cheers

Phil Hobbs

I considered that. I also considered one Kelvin trace to one pad-center and the other to the other pad's corner. (One ideal, one compromise connection.)

The layout guy rejected two inner connections as a solder-shorting hazard, but it doesn't look bad to me.

The inner gap is 0.023", so you could slip a 0.008" trace in there with 0.008" air gaps. I just now managed to squeeze in a 0.006" trace down the center to each pad with your 45 deg diagonal jog just before reaching the part's center line.

But the real revelation to me was, it doesn't matter. Being able to probe the potential gradients on the copper past the solder fillet was revealing. To within my 50 microohm measurement resolution, corners were just as good as connecting to the dead center of the pads.

Cheers, James Arthur

Oops. 5uV / 0.5A = 10 micro ohm equivalent resolution. Doh!

--James

That one above looks about as good as you're going to get. It should be fine with 0805's and 6 mil traces. Our 0805s have 30 mils between the pads, which just works at 6/6. Tell your layout person to man up.

This worked very well:

The resistors are 39 mohm 2512's to get about 10 mohms net. We calibrate around that. More resistors helps reduce the solder resistance inherent in a 2-wire resistor, and I expect 15 amps or so.

To do much better, you'd need a proper 4-lead resistor.

On Monday, January 18, 2021 at 9:51:05 PM UTC-5, snipped-for-privacy@yahoo.com wrote :

, but

.008"

g. To

Did you take into account soldering variations? Not every solder joint tha t is adequate for electrical purposes will be completely uniform potentiall y giving a higher or lower voltage to a corner connection than a center con nection. This conversation made me take a look at the power supply board o n the project I'm working on and they guy used every variation of Kelvin co nnection you can think of. One resistor uses a good Kelvin connection to o ne pad and the "ground sense" connection goes directly to the ground via an d not the pad at all!

I expect this is because he is thinking the importance of the Kelvin connec tion is related to the current flowing in the connection, but in reality it is related to the resistance of the current sense element. The ratio of t he current resistor to the parasitic resistance in the connections is what determines the need for a Kelvin connection. In this case, it is the tinie st sense resistor on the board at 15 milliohms.

In the process of reviewing the design for the current sense resistors I fo und he has the power FETs wired in backwards.

I've seen people recommend trifurcating the inner aspect of the pad and connecting the Kelvin sense to the center trifurcation. I'm skeptical that does much more than assure a connection to the center of the resistor's termination, which you got anyhow by just running the Kelvin sense to ... the center of the pad's inner aspect.

The lower connection (to pins 1 of R17-20) connects to that pad-bus just north of the solder fillet, but at a position where I still measured noticeable voltage gradient in my pads running at 1/2A. Current flow at that position is still non-zero.

Connecting at the top corner past all the current flows, just as the pins 2 Kelvin goes to the corner above it, would avoid any trace-drops in that area.

You might arguably have wanted to tap off early like you did, to compensate for the drop across the component's end termination from solder fillet to the top of the part, though. Doesn't make all that terribly much difference for a 39 m-ohm shunt.

(I really like the sense resistors with resistive film top and bottom. Makes 'em tougher.)

The gap between your 2512s' pads is pretty small. You could widen that up and sneak in connections to the center of each bussed shunt connection, and add some tiny heat sink fins to make up for the lost copper pour. :-)

Cheers, James

Solid metal is even tougher!

Unless you buy manganin or zerenin resistors, or some Vishay or Susumu sort of thing, the details of the Kelvin connection are lost in the resistor errors.

PCB traces are about 500 uohms per square, so a Kelvin connection often makes sense. But don't get crazy about the details. Via in pad is I guess ideal.

When we were in the NMR business, we made our own 6-wire shunts and epoxied them into temperature-controlled aluminum blocks. That's a whole nother story.

That would probably be correct if the solder wicking was always uniform. But when it isn't, the trifurcated pad will most likely give a better connection than any other.

The end caps are made of plated nickel, iirc, which has a huge range of conductivities depending sensitively on conditions and composition--I've seen numbers quoted ranging from about 7 uohm-cm to 110 uohm-cm. (Copper comes in at about 1.7 uohm-cm, and solder somewhere between about 8 and 40 depending on composition.)

So it isn't instantly obvious that the end caps are going to be much more accurately equipotential than the pad if there are wetting or solder-volume issues.

Cheers

Phil Hobbs

rote:

n--

cal

. But when it isn't, the trifurcated pad will most likely give a better con nection than any other.

Of course it is a small order effect. I don't think it is a matter of resi stivity. It would be simple geometry. But maybe this not a real thing. I didn't sit down and plot our potential lines. I wouldn't go to such measu res myself. I can't imagine connecting at the inner edge of the pad isn't good enough. But then it all depends on the value of the sense resistor. Someone was trying to tell me it had to do with the current, but it doesn't really. How much impact the connection makes on the measurement (the need for the Kelvin connection) depends only on the resistance of the joint com pared to the value of the sense resistor.

Imagine some software that monitors the PCB base temperature and the current-squared history of the resistor. It runs a thermal model to compensate TCR and thermoelectrics, with suitable cal factors.

The fantasy does have to include well-defined and stable Peltier/Seebeck coefficients.

Once you put your blob of solder on the board and the component it ought to have fairly stable properties, but you'll probably have to measure them separately for each board and sorting one from the other might be tricky.