Author Topic: Shunt capacitance of 1206 SMD resistors; Jeroen Belleman - December 2010  (Read 5950 times)

0 Members and 1 Guest are viewing this topic.

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
Some time ago I had come across an paper written by Jeroen Belleman where he was studying the parasitic shunt capacitance of some 1206 resistors.  The original paper may be found here:

http://jeroen.web.cern.ch/jeroen/resistor/shuntC.html

He notes:
Quote
At the low-frequency end, it also curves up, for reasons I do not currently understand. If I insert a few dB of attenuation at both ends of my test jig, the effect goes away.

I had been wanting to repeat this test to see if I could replicate that curve up.  Shown are my 1206 test resistors plus an added 50 ohm part.   Sadly my PNA can't run below 300kHz.   What I have done is used my HP3589A to collect the data below 300kHz and then stitched the two files together.   Their setup starts to curve up below 100kHz so I thought I would collect data down to 3kHz. 

The source was set to 0dB and the RBW to 1.2kHz (thought about lowering it and using the log sweeps).   I have no additional attenuators inserted and see no signs of this upwards curve.   

****
Change scale of my graph to allow easier viewing.
« Last Edit: July 15, 2021, 12:11:56 am by joeqsmith »
 
The following users thanked this post: croma641

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
I never realized when the HP3589A starts a sweep, it causes a small disturbance in the data.
Operator error..  ::)   

Shown with nothing inserted in the test jig (open) to look at the floor with a RBW of 150Hz.  I would like to see about 10dB of headroom but it should be fine. 

It will be interesting to see if the METAS software can hand this much data and with it being stitched together.   

Note that I had started out with a cable that was loaded up with Ferrite where it it shown without.   As expected, I saw no difference and plan to run it as shown (without the Ferrites).
« Last Edit: July 14, 2021, 10:00:36 pm by joeqsmith »
 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
Shown is the combined data, sweeping from 900Hz to 6GHz with roughly 27,000 data points segmented quasi log sweep.    A fairly decent test for both programs.   

All the Ferrite was removed when collecting this data.  Once again, there are no signs that odd amplitude change at the < 100kHz.
 
The following users thanked this post: croma641

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
I converted the original data to Touchstone format and overlaid it with the data I collected.   

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21567
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Beats me; does seem like an instrument error or something.  Have you contacted Jeroen?  He's... well at CERN still AFAIK, and also I think checks sci.electronics.design sometimes.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
Beats me; does seem like an instrument error or something.  Have you contacted Jeroen?  He's... well at CERN still AFAIK, and also I think checks sci.electronics.design sometimes.

Tim

I wrote Jeroen to see if he had any additional clues.  They may attempt to repeat their test.  I have provided them with a link to this thread as well.   

It seems strange that the problem would be below 100kHz.  I'm actually surprised how well the two tests tracked.
 
The following users thanked this post: croma641

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16512
  • Country: us
  • DavidH
It looks to me like hook in either the resistor substrate or test fixture substrate.  It is in the correct frequency range.
 
The following users thanked this post: MegaVolt

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21567
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
I thought of that, but that doesn't explain it being independent of value.  It's got to be systematic.

However, it probably explains the discrepancy between ideal RC and the slightly softer curve measured!

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 16512
  • Country: us
  • DavidH
Hook can be part of the resistive element itself.  Some resistive materials display dielectric absorption.

I can think of tests which might resolve if it is hook or not, but they depend on finding a bad part or test fixture first.

My prime suspect is hook simply because whatever it is is in the same frequency range.  Is that a coincidence?
« Last Edit: July 16, 2021, 06:54:19 pm by David Hess »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21567
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
It's, kind of coincidence, kind of not, I guess?  Probes are some megs and some pF, equivalent to the diagonal asymptote being further to the left (more C), and a bit lower down (more R).  So, it's directly applicable when the components are crossing in that region -- but there's no reason to expect that time constant should be independent of value, indeed it should be proportional.  Which means, that would indeed explain the soft corner; but not the tick up on the left.  (And you can make probes with whatever resistances, and the hook scales proportionally, just as the soft corner does here.)

Dielectric absorption, now that's interesting; I suppose it could even be something like substrate (does the fixture use FR-4?).  But again it seems unlikely that it should be so strong (it's a constant number of dB from the flatband level, and independent of attenuation level?), or that it should attack so suddenly (it looks roughly like another simple pole; absorption I think I'd expect a diffusion mechanism instead? -- not that we have much of the curve to really infer much about its overall shape).

Of the materials in the resistor itself, I would think that's very unlikely; the alumina substrate should be pretty clean, and the enamel is very thin.  But Idunno; is alumina known to have significant absorption at lower frequencies?  I suppose it could, if it's got like hydroxyls in the crystal, or adsorbed water or something.  Neither of which should really be all that significant given how it's made (fair purity, high density (very low porosity), high firing temperature), but it would be interesting if true.

It really just seems like AC coupling in the system... was an AC block accidentally left on, maybe?

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
It's too bad they didn't take the time to document the their experiment better.  Just a few pictures of the setup may have helped.    I doubt you would find 10 different valued resistors all having the same problem.   I would also expect that when the system was calibrated that nothing odd turned up which would rule out the cables.   Then again, maybe it wasn't checked.  I would suspect something with the fixture.   From their write up, I envisioned two SMAs soldered back to back with the part mounted between them, similar to this:

https://www.eevblog.com/forum/projects/20db-rf-attenuator-seeking-feedback-to-improve/msg2965430/#msg2965430

I doubt that would cause any problems assuming decent parts were used.  I would think they checked it but I have no idea.  Hopefully they will take the time to repeat the test and report back their findings.   

I'm sure a few may find it a bit strange that the part of their test I am interested in has nothing to do with the problem they were trying to solve.   :-DD   

Offline G0HZU

  • Super Contributor
  • ***
  • Posts: 3007
  • Country: gb
Stumbled across this on a google search. I think this is my first post for over two years on eevblog!

I think the issue with the 8753D VNA could be that is has internal bias tees on each port and the cold (VNA) side of the internal bias tee will have a series capacitor to protect the deeper innards of the VNA from any externally applied DC at the bias tee port at the back of the instrument. This means the mismatch uncertainty (due to uncorrected VNA port mismatch) could be significant down towards 30kHz if the operator doesn't calibrate the VNA correctly for a two port measurement. The alternative to a proper calibration is to fit decent quality attenuators inline although these will eat away the dynamic range of the VNA.

This mismatch uncertainty issue also applies over the full 6GHz range of the plots. Both the Jeroen and the joeqsmith plots look really poor considering a lab grade VNA has been used. If calibrated correctly I'd expect to see a smooth trace for a chip resistor right up to 6GHz. The wiggles in the response shouldn't be there. Ideally the VNA and the test fixture needs to be calibrated properly for a full two port measurement because the mismatch uncertainty (when measuring something like a series resistor of value 100R to 1Meg) will be very significant otherwise.
« Last Edit: July 18, 2021, 06:06:26 pm by G0HZU »
 

Offline G0HZU

  • Super Contributor
  • ***
  • Posts: 3007
  • Country: gb
As you have a decent (but old!) PNA VNA my advice would be to study up on how to calibrate the system when making measurements in a test fixture. I think the older PNAs can support two port fixture simulator models for example. To avoid the wiggly plots shown in this thread you really do need to cal/correct the system right up to the pads of the series resistor under test.
 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
I can't speak for Jeroen but in my setup, I have only normalized the data.   I haven't thought about how to calibrate with that fixture.  Sounds like you have some ideas how to pull it off.

Offline G0HZU

  • Super Contributor
  • ***
  • Posts: 3007
  • Country: gb
Yes, your plots do look like normalised measurements. I suspect Jeroen has done the same and this might explain the 'curve up' in his plots because the port match of the 8753D might be compromised slightly down at 30kHz and this could be partly due to the internal bias tee hardware.  It would explain why he saw the problem go away when attenuators were added inline. However, all we can do is guess at the problem.

I haven't used an 8753D in many years so I can't remember how good the port match is at 30kHz. At work all the 8753D and 8573ES VNAs were disposed of about 3 years ago so I can't do any tests on one. The 8753D datasheet suggests that the port match is still quite good at 30kHz but I think there are different versions of the 8753D VNA. I think there is a version that uses an external s-parameter test set for example.

To get good plots up to 3GHz and hopefully to 6GHz you would obviously need to have a decent cal kit and you would also need to know how to correct for your test fixture. I suspect the older PNA can support a 2 port model of your test fixture and you can tell it to use this model internally. This means the VNA will spit out s-parameters that are corrected to the pins of the resistor. Your test fixture looks very big and also very long to me... not where I'd want to be starting from.
 

Offline G0HZU

  • Super Contributor
  • ***
  • Posts: 3007
  • Country: gb
I don't have a 1206 test fixture here but I did manage to find a 100k 1206 resistor and I knocked together a very crude 1206 sized 2 port fixture using SMA PCB connectors.
This is far from ideal but I tried measuring it in the 2 port fixture using a VNA with a homemade SMA cal kit and an Ecal kit.

The comparison is below. My ancient homebrew SMA kit does quite well here although I did exploit the 'unknown through' feature and the 'fixture simulator' feature in the VNA itself in order to get this level of performance. I've not cheated and applied smoothing on the trace data.

I'm not really that comfortable using a homebrew SMA cal kit above 2 or 3GHz but the results in this case are quite good all the way up to 6GHz. I did generate Cx and Lx corrections for the SMA cal kit some years ago and these obviously help a bit once up above 3GHz.

 

Offline rf-messkopf

  • Regular Contributor
  • *
  • Posts: 156
  • Country: de
  • Mario H.
    • Homepage
To get good plots up to 3GHz and hopefully to 6GHz you would obviously need to have a decent cal kit and you would also need to know how to correct for your test fixture. I suspect the older PNA can support a 2 port model of your test fixture and you can tell it to use this model internally. This means the VNA will spit out s-parameters that are corrected to the pins of the resistor. Your test fixture looks very big and also very long to me... not where I'd want to be starting from.

If you want your cal planes somewhere in a planar circuit, TRL calibration is a good way to go. You could include include the requisite cal standards on your test fixture pcb. The problem is that you need a tightly controlled board impedance as the impedance of the line standard determines your system impedance. Another problem is that TRL has a lover frequency limit, determined by the length of the line standard. There is however a variant of TRL, called TRM (thru-reflect-match), that extends TRL to low frequencies. Modern four-receiver VNAs have that implemented in firmware.

I did some experiments with that a while ago: https://www.mariohellmich.de/projects/trl-cal/img/trl-fixture.jpg. See also https://www.mariohellmich.de/projects/trl-cal/trl-cal.html.
 
The following users thanked this post: ON7CH

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
I have a set of standards that I can use if I need to calibrate the system.  If they are decent or not is subjective.  Poor Mario had to put up with me while I attempted to sort them out.   
For my home hobby use, good enough. 

Offline G0HZU

  • Super Contributor
  • ***
  • Posts: 3007
  • Country: gb
To get good plots up to 3GHz and hopefully to 6GHz you would obviously need to have a decent cal kit and you would also need to know how to correct for your test fixture. I suspect the older PNA can support a 2 port model of your test fixture and you can tell it to use this model internally. This means the VNA will spit out s-parameters that are corrected to the pins of the resistor. Your test fixture looks very big and also very long to me... not where I'd want to be starting from.

If you want your cal planes somewhere in a planar circuit, TRL calibration is a good way to go. You could include include the requisite cal standards on your test fixture pcb. The problem is that you need a tightly controlled board impedance as the impedance of the line standard determines your system impedance. Another problem is that TRL has a lover frequency limit, determined by the length of the line standard. There is however a variant of TRL, called TRM (thru-reflect-match), that extends TRL to low frequencies. Modern four-receiver VNAs have that implemented in firmware.

I did some experiments with that a while ago: https://www.mariohellmich.de/projects/trl-cal/img/trl-fixture.jpg. See also https://www.mariohellmich.de/projects/trl-cal/trl-cal.html.

Interesting stuff, thanks!

To demonstrate how easy it is to get wiggly data if the VNA isn't calibrated very well I tried putting together a simulation to demonstrate the window of uncertainty if the uncorrected VNA port VSWR is as poor as 1.3:1 across 30kHz to 6GHz. This tests a crude 100k resistor model that has 0.05pF parallel capacitance. This is shown below and it shows the area that any trace wiggle could reside within if the VNA ports aren't calibrated/corrected properly.


 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
No doubt it's easy to get wiggly lines.  No simulation required.   I suspect OP didn't feel the need to cal it to see the general trend they were looking for but you would need to ask them.   For me, that was certainly the case.   

There was a link above showing simple back to back SMAs for similar use.  For lower frequency use, I mount the parts to a PCB.  The two large fixtures make it convenient to change out parts but obviously there is a down side.   As I said, not normally a problem though.     

I have not heard anymore from Jeroen.  Maybe someone else will attempt to run it on their 8753D to see it if the problem was caused by normalization.   These seem fairly common now. 

When looking to replace my 1970s HP8754A, I thought about getting the E or ES model but ended up with the PNA.  For the most part I like it.   

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
As long as the PNA was warmed up, ran a cal and soldered a 1206 into one of the small fixtures and ran it. 

Fullcal3:  looking at the original data I collected, the data from CERN and tonight's sweep. 

Fullcal2:  tonight's data.  RBW was set to 200(?) 1kHz.   

I suspect the tail is the fixture.   Open and short standards are home made.  Loads are purchased and sorted.  Borrowed a couple of sets of standards to characterize them (where Mario helped out).   

***
Correct RBW
« Last Edit: July 24, 2021, 03:10:29 pm by joeqsmith »
 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
Like the original paper, I added included a simple RC for the 50fF//1Meg along with a 40fF and 60fF.   

Window0 is showing the original CERN data.

Window2 is showing the original CERN data along with the first data I collected.  Zoomed in 20MHz - 2GHz to better show the fit.  The 50fF set was turned off.   Granted, if I had spent the time to calibrate the system the data would indeed not have near the ripple.  But when you consider the time along with wear and tear on the standards,  I don't see the point if you were looking for a ballpark number.   I can only guess if that was what the OP had concluded.   

I wonder if a high quality, characterized jig had been used with a modern VNA that was fully calibrated, what sort of variance would you measure across resistor brands and lots.  Again, I'm surprised how close the data came out. 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
I attempted to build up a little better fixture to see what effects it has on the upward trend above 2GHz.  The PNA was calibrated w/ homemade standards and the RBW was set to 1kHz.   

Fullcal4_fixtureII_1Meg_1kRBW:  Comparing the new fixture with the previous and data from CERN.  Also shown is a perfect 50fF capacitor.

Fullcal4_fixtureII_1Meg_1kRBW_10MHz_6GHz_normalized: To better show the errors, I normalized the three datasets to the calculated 50fF and zoomed into the 10MHz-6GHz range. 

As suspected, the fixture does appear to be the cause of the upward trend.  Using homemade standards and jigs, I wouldn't put much stock in the data.   

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
Shown with the ideal 1Meg in parallel with a 60fF and 40fF capacitors.  Comparing the uncalibrated data (normalized only) from my quick change coaxial fixture, the data from CERN and the most recent data with the PNA fully calibrated using my coplanar fixture with the resistor soldered in place.   The three data sets are again normalized to an ideal 50fF capacitor.   

Looking at the normalized data from the quick change fixture, I can tell it's roughly 50fF as the original author stated.  Other's need to take it to another level before coming up with an answer.  There's a thread where a member was talking about cal standards and measuring SWR and I was using rough numbers as being good enough. 

https://www.eevblog.com/forum/testgear/nanovna-v24-sma-cal-load/msg3360916/#msg3360916
« Last Edit: July 24, 2021, 04:56:11 pm by joeqsmith »
 

Offline joeqsmithTopic starter

  • Super Contributor
  • ***
  • Posts: 11601
  • Country: us
For fun, I thought I would repeat the test using both the original NanoVNA and the V2Plus4.   I used the coplanar fixture and just normalized the data.    Compared with the normalized data using the quick change fixture and the data from CERN. 

Also shown, all four data sets normalized to a perfect 50fF with a sweep range of 20MHz to 1GHz.   Both low cost VNAs performed better than I would have expected.


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf