LCR Cal Test PCB

[KungFuJosh]

@Josh:

1mm drill hole..

Will 1.5mm still work for you? I don't mind changing it, do you have the sheet for the pins?

[mawyatt]

It's mostly going to be the same as V4 board pictured above. The V4 board has a 4mm SMT pad on top and bottom of the board running to each test point, with the bottom traces connecting via a hidden via near each component. The issue was only making the physical connection with the kelvin clips more reliable.

Instead, for the new version, I'll keep everything top side in parallel and eliminate the extra vias. As you can see from the test results, accuracy was great (at least for my needs). I'm not trying to get anywhere near metrology craziness. But considering the ST42 was direct to component, and the ST2832 was via the test points, that's certainly close enough for me. If any of the disagreements were related to the connection type, that would most likely be the fault of the inconsistency with the mechanical connection on the top/bottom pads that required solder to make the connection work.

I'll still be adding the open/short points as well for calibration before testing.

If you have any opinion about the above, please share. 😉

For the Kelvin Clips to mount directly to thru holes Force and Sense, you want the holes 90 degrees to the DUT and on the outside of DUT lengthwise. You need enough spacing between the Hcur Force Hpot Sense on one Kelvin Clip to the Lcur Force and Lpot Sense on the other Kelvin Clip. These Clips need to be mounted with thin side 90 degrees to the DUT length, otherwise the handles will interfere. The Kelvin Clips we have are ~18mm thick across the handles.

Make a mockup out of cardboard and you'll see what we mean

Best

[KungFuJosh]

For the Kelvin Clips to mount directly to thru holes Force and Sense, you want the holes 90 degrees to the DUT and on the outside of DUT lengthwise. You need enough spacing between the Hcur Force Hpot Sense on one Kelvin Clip to the Lcur Force and Lpot Sense on the other Kelvin Clip. These Clips need to be mounted with thin side 90 degrees to the DUT length, otherwise the handles will interfere. The Kelvin Clips we have are ~18mm thick across the handles.

Make a mockup out of cardboard and you'll see what we mean

Best

Thank you, I totally understand regarding the spacing issue. It will be most important for the low value components that get independent connections for both sides. It will be less of an issue for the rest, but I'll keep it in mind for the placement.

OTOH, we could just connect with one kelvin clip on each side of the PCB. No space issue that way! 😉

[KungFuJosh]

Here's a tentative layout with the new connectors. I haven't run any traces yet. I think angling the kelvin clips a little bit should remove any physical limitations given the space. Currently the board is 150mm x 120mm, and I'd prefer not to go any larger.

Thoughts?



Thanks,
Josh

[mawyatt]

The "Open & Short Cals" must include the traces and DUT pads to be accurate.

For example, the 1, 10, 100pF must have an accurate "Open Cal" to remove all the parasitic trace & pad capacitance to accurately resolve the actual DUT capacitance.

For low value resistors like 1, 10, 100 ohms, these require an accurate "Short Cal" to remove the parasite trace and pad resistance, same for low valued inductances like 10, 100, 1000nH.

The "Split Kelvin Holes" are for the low R, L & high valued Cs, maybe consider going to 100 ohms, 1uH, and above 10nF.

No need for the "Split Kelvin Holes" for the low valued Cs (use 2 Kelvin Hole Open Cal here), you want this for the high value Cs where ESR and ESL are important, like above 10nF.

Move the Kelvin Holes closer to the DUT on the Non-Split Kelvin fixtures since the traces are not compensated, keep traces as short as possible.

Since you have plenty of space, maybe locate Open and Short Cal near the high and low Impedance valued components on each side by R, L, and Cs.

Also, why the small thru holes on the Kelvin Clip Thur Hole tops? Those were for the terminals if used.

Avoid thru holes if possible, they add unwanted additional resistance and inductance to the fixture parasitics.

For low Z components try and "carry" the Kelvin connections right to the DUT Pad with low trace resistance for the Force trace, Sense trace shouldn't matter, then for the high Z components "carry" a single short length trace from the Kelvin Thru Hole to the DUT Pad and minimize parasitic capacitance.

Best,

[mawyatt]

Josh,

Another way of thinking about the above is, the high Z components will experience higher voltages (low currents) during test and thus small shunt parasitic capacitance matters, whereas for low Z components will experience higher currents (low voltage) and small series parasitic resistance & inductance will have a direct influence.

This is why "carring" the 4 Port Kelvin Connection to the DUT contact point is important, it helps by sensing the DUT Voltage at the DUT point of contact thus partially removing the effects of parasitic fixture (and cable) series resistance and inductance.

We must also remember that our Lab Grade LCR meters measure complex Real and Imaginary DUT Voltage and Current, then compute the desired display DUT parameters and why getting the DUT thru current and voltage across DUT is paramount in achieving accurate and repeatable results

Best,

[Phil1977]

It´s really academic to discuss if smaller or larger values need 4W-wiring. There are good reasons to do both, but I´d be very satisfied with such a board that has 4W-connections only for low C, low L and low R - as depicted in the latest rendering. I think if you need more accuracy, you should take dedicated boards for each value and also avoid common return lines - but that contradicts the purpose of a handy and convenient quick test board.

Last but not least: Please make the penguin bigger again 

[KungFuJosh]

The "Open & Short Cals" must include the traces and DUT pads to be accurate.

They will, but I'll see if I can make them the same size/length as the low value stuff. I downloaded your doodles from your previous reply, they were definitely helpful in understanding the setup.

For example, the 1, 10, 100pF must have an accurate "Open Cal" to remove all the parasitic trace & pad capacitance to accurately resolve the actual DUT capacitance.

For low value resistors like 1, 10, 100 ohms, these require an accurate "Short Cal" to remove the parasite trace and pad resistance, same for low valued inductances like 10, 100, 1000nH.

I think this should be easy enough for those specific low value items. I know I could go much further with this and have separate connections for everything, at perfect lengths, with open/short cal for each section depending on pad sizes. However, that's not the purpose of this. This board is supposed to be a relatively quick way to test each range. I need to compromise between what's honestly very slightly better results and test speed. If I have to disconnect clips for every single measurement, that's time wasted. My impatience is far stronger than my desire for perfect accuracy. 😉

Based on my previous board, which is certainly a lesser design than the direction this is going, only 1Ω was off from the ST42 direct to DUT connection, and only by 5µΩ.



Capacitance was generally well agreed until 10µF and 100µF. Should I have those higher values separated from the string? It sounds like you're saying those high values need the separate connections for C, not the low values? Even 1pF or 10pF can be in the string? IIRC I added the separate points for 1pF and 10pF because of issues with them when in the string with the rest.



Besides that I need to reduce the distance between the DUTs and the terminals. No problem.

It´s really academic to discuss if smaller or larger values need 4W-wiring. There are good reasons to do both, but I´d be very satisfied with such a board that has 4W-connections only for low C, low L and low R - as depicted in the latest rendering. I think if you need more accuracy, you should take dedicated boards for each value and also avoid common return lines - but that contradicts the purpose of a handy and convenient quick test board.

You're right, of course, but I enjoy learning from Mike. If I'm understanding him correctly, it seems that I should have the separate connections for C at the higher values instead of lower values. If that's correct, it's not a big deal. As long as I don't need separate stuff on all of them.

Last but not least: Please make the penguin bigger again 

Absolutely! 🤣

Thanks,
Josh

[mawyatt]

Having separate Force (Hcur & Lcur) and Sense (Hpot & Lpot) connections for high value Cs (low Z) will allow better ESR and ESL evaluations. Also for low Cs (High Z) consider just single trace to DUT ends, this reduces parasitic capacitance.

One reason you may be seeing the discrepancy between the LCR and Tweezers for high value Cs is these are likely the high dielectric types which have a high voltage/temp/hystersis/age sensitivity unless they are NP0/C0G types. This sorta implies that the Tweezer/LCR excitation voltage needs to be the same as well as temp and the readings repeated back and forth between the two instruments with cap leads shorted in between the swaps. Getting repeatable results with high valued ceramic caps is almost a waste of time unless they are NP0/C0G, electrolytics are a little better but still difficult wrt repeatability.

For higher C values (up 10/100uF) you might consider Film types as even the cheap Mylar caps are much better wrt to repeatability, Polypropylene is better and Polystyrene are excellent.

Awhile back we considered making a somewhat small PCB with a few Cs and Rs, no Ls tho. This would serve for some quick checks but would be designed such that it could be considered for more serious verification. Even considered including a thermistor on PCB so temperature could be recorded. One of these PCBs could be sent around for folks to compare results with various instruments, which would be interesting indeed!!

If we ever get enough time and things calm down we'll consider rolling one of these, but for now it's on the back burner. Of course your project is heading in the same direction

Anyway, hope this helps.

Best,

[KungFuJosh]

Having separate Force (Hcur & Lcur) and Sense (Hpot & Lpot) connections for high value Cs (low Z) will allow better ESR and ESL evaluations. Also for low Cs (High Z) consider just single trace to DUT ends, this reduces parasitic capacitance.

One reason you may be seeing the discrepancy between the LCR and Tweezers for high value Cs is these are likely the high dielectric types which have a high voltage/temp/hystersis/age sensitivity unless they are NP0/C0G types. This sorta implies that the Tweezer/LCR excitation voltage needs to be the same as well as temp and the readings repeated back and forth between the two instruments with cap leads shorted in between the swaps. Getting repeatable results with high valued ceramic caps is almost a waste of time unless they are NP0/C0G, electrolytics are a little better but still difficult wrt repeatability.

For higher C values (up 10/100uF) you might consider Film types as even the cheap Mylar caps are much better wrt to repeatability, Polypropylene is better and Polystyrene are excellent.

Awhile back we considered making a somewhat small PCB with a few Cs and Rs, no Ls tho. This would serve for some quick checks but would be designed such that it could be considered for more serious verification. Even considered including a thermistor on PCB so temperature could be recorded. One of these PCBs could be sent around for folks to compare results with various instruments, which would be interesting indeed!!

If we ever get enough time and things calm down we'll consider rolling one of these, but for now it's on the back burner. Of course your project is heading in the same direction

Anyway, hope this helps.

Best,

Okay, so single traces for the two independent low value Cs. Split traces for the high value Cs with separate connections. I can do that, I'll get over the 3 extra clip movements. 😉

I'm using NP0/C0G for all the Cs that those are available for. I'm using X7R for the larger caps, 10% X6S for the 100µF, and a 20% tantalum for the 1mF. Switching to other options becomes difficult either due to space or cost or both. I will look at a couple alternative options for the 10µF & 100µF caps. I may keep them as they are and measure them off board to find the desired value rather than the stated nominal value before installing.

You make a good point regarding the test voltage. The tweezers test at 1Vpp and the ST2832 at 1Vrms. I may try those C values that disagreed with matched test voltages and see how they look.

Thanks as always,
Josh

[mawyatt]

Tantalums are usually not too bad, however the 100uF should be in X7R if possible, maybe parallel a few up for ~100uF, or just use a Tantalum, it should be better than the higher K dielectrics!!

Best,

[KungFuJosh]

Tantalums are usually not too bad, however the 100uF should be in X7R if possible, maybe parallel a few up for ~100uF, or just use a Tantalum, it should be better than the higher K dielectrics!!

How about this guy? F950J107KAAAQ2

[mawyatt]

Kyocera is a good brand, never had any problems with their components. You might consider a higher voltage rating like 10V, and/or a leaded capacitor.

Best,

[KungFuJosh]

Kyocera is a good brand, never had any problems with their components. You might consider a higher voltage rating like 10V, and/or a leaded capacitor.

Test voltages are typically between 0.354V to 2Vrms. Do you often test with higher voltages?

ETA: size is my only concern. If they fit at 10V, that's fine with me as long as they don't cost 10x as much. 😉

[KungFuJosh]

Here's an updated version of the board with the traces ran.

[KungFuJosh]

Here's a suggest BOM for V5. Many of the parts have been used previously, some are new. I didn't like some of the inductors, so I changed some of them out.

Everything is at least 10V except the 1mF cap which is 6.3V.

Feedback is welcome. Try not to cause too much work. 😉

Thanks,
Josh

[tautech]

My 2c worth, move all the value labels further away from the test points and/or increase the font size.

[KungFuJosh]

One reason you may be seeing the discrepancy between the LCR and Tweezers for high value Cs is these are likely the high dielectric types which have a high voltage/temp/hystersis/age sensitivity unless they are NP0/C0G types. This sorta implies that the Tweezer/LCR excitation voltage needs to be the same as well as temp and the readings repeated back and forth between the two instruments with cap leads shorted in between the swaps.

Correct as usual. The biggest discrepancy was the "100µF." I retested with 350mVrms on the ST2832, and 1Vpp on the ST42:

ST2832: 68.0038µF
ST42: 68.06µF

Much better agreement! Previously they were off by 4µF.

Thanks,
Josh

ETA: I suppose when I get V5 of the board, I should leave the test voltages matching.

[tautech]

My 2c worth, move all the value labels further away from the test points and/or increase the font size.

Oh and add device tolerances, 0.1% etc.

[KungFuJosh]

My 2c worth, move all the value labels further away from the test points and/or increase the font size.

No problem, old friend. 😉

Oh and add device tolerances, 0.1% etc.

That might be an issue if other people build them with other parts.

[Martin72]

Here's a suggest BOM for V5.

1% capacitors ? Very good, I'm a bit surprised, I would have expected more from mica capacitors.
I still have to catch up on the thread a bit, up to the point where it will be explain why a four-wire method is not executed for all values.
With the resistors this makes immediate sense, with a tolerance of 0.1% for a 10K resistor for example, a few micro ohms make no difference.
The comparatively high tolerances for the coils are unfortunately unavoidable if you don't want to build huge boards.
The core makes the tolerance, roughly speaking, "better" would be air coils, but these become huge with increasing values.

I also admire your work with KiCAD, I've tried it a few times but can't get to grips with it and am still stuck with EAGLE.

[tautech]

My 2c worth, move all the value labels further away from the test points and/or increase the font size.

No problem, old friend. 😉

Better. 
Would you consider shifting those that can be to above their respective devices and making them larger still ?

[KungFuJosh]

I still have to catch up on the thread a bit, up to the point where it will be explain why a four-wire method is not executed for all values.

The short answer is: laziness and impatience combined with the lack of evidence showing that it's needed.

I might be convinced to add them all, but I don't like the extra time it takes to single everything out (while testing), especially when any difference is negligible. Look at my updated reading above based on Mike's feedback to match the test voltages. The ST42 with a direct connection to the DUT, and the ST2832 using the non-isolated kelvin connections had excellent matching, and that kelvin connection is one of the longest in the section.

I also admire your work with KiCAD, I've tried it a few times but can't get to grips with it and am still stuck with EAGLE.

Thank you, KiCad us great. I learned it out of necessity. Years ago I was using some crappy proprietary PCB designer (PCBExpress), and refused to pay the cuckoo prices for Eagle when I outgrew that other one. Like anything though, it's easier to learn if you're not stubborn about whatever program you were previously used to. 😉 There are tons of youtube videos to help learn it also. Also some good plugins too.

[KungFuJosh]

Better. 
Would you consider shifting those that can be to above their respective devices and making them larger still ?

I don't like inconsistencies like that. Use one of your 500 pairs of reading glasses. 😉😉

[tautech]

Better. 
Would you consider shifting those that can be to above their respective devices and making them larger still ?

I don't like inconsistencies like that. Use one of your 500 pairs of reading glasses. 😉😉

PCB finished size ?