Using Models to Predict Capacitor Leakage Current

[GEOelectronics]

"- a repairman usually does not keep the defective parts to avoid mixing them with the good ones.
"

Again I agree 100%. Most, nearly all, of the testing of components I do is to see if they are suitable to go into the spare parts inventory. These can be brand new or surplus. Like a dog chasing his tail, I learned that "new" parts can be bad.

One time I bought a bunch of 7 segment LEDs from a local surplus electronics joint. Several were bad so next trip I returned them, which they gladly replaced. On the way out the door I noticed they threw the bad parts right back in the bin with the new ones!

Geo>K0FF

[joeqsmith]

I have picked out four low voltage (less than 100V) and two higher voltage caps.   I plan to start with two each,  just to get a feel for if I can get any meaningful data at all.   Most of my test equipment is very old.  My electrometer does not support GPIB or any other interface.       

Well, on the plus side it looks like placing the parts into a metal can and such, I can seem to get some sort of reading with the electrometer.   I wonder if Keysight would donate that new electrometer.     Now that would be the ticket. 

Here is another article that may be of interest.  The low voltage knee caught my attention.

https://www.avx.com/docs/techinfo/Low_Leakage_Current_Aspect_Designing_Tantalum_Niobium_Oxide_Capacitors.pdf

[joeqsmith]

First attempt at some sort of capacitor test fixture.  I have been testing with a NOS polycarb.  Using a single 9V transistor battery, it's no problem measure the current.   

[joeqsmith]

The part I am using is a 2uF 50V metallized polycarbonate.  The datasheets for this part state the IR is 25,000Meg X uF at the rated voltage after two minutes.  So basically it better be below 1nA at 50V.  I wiped off the part with iso and let it dry.  Was careful not to touch the body when installing it and then sealed the test fixture.   

I manually took three datapoints, allowing the part to settle for 5 minutes between each levels.   At 0V I measure 90fA.  At 25V, I measure 33.2pA.  If it's linear at 50V I was expecting to see around 70pA.   What I measured was 303.6pA.   Sure it's a crude measurement but it's not even in the ballpark of what I would expect from the simple resistive linear model.   Again, this is a new old stock part.  I have not stressed the part in any way.   

Even at 2 minutes into the test, the part was well below the manufacture's number.   It will be interesting to see what others come up with.

[ChunkyPastaSauce]

Lot of the charts I just checked have log-lin scale.

[GEOelectronics]

That fixture looks great. Is that Teflon sheet? Who supplies Teflon sheet? Thanks

Is that an MHV connector? does he cable have a BNC connector or did you modify the MHV connector to take both types with my hack? Thanks

Geo>K0FF

[joeqsmith]

That fixture looks great. Is that Teflon sheet? Who supplies Teflon sheet? Thanks

Is that an MHV connector? does he cable have a BNC connector or did you modify the MHV connector to take both types with my hack? Thanks

Geo>K0FF

Yes, its Teflon.  There was too much resistance to the case without it.  I imagine there are many places to purchase Teflon.  Google may be your best bet.   I get most of my hardware like this from McMaster  -Carr.

https://www.mcmaster.com/#teflon-(made-with-ptfe)/=1d0drid

That is an SHV connector.  It has not been modified in any way. 

Lot of the charts I just checked have log-lin scale.

There are several texts that talk about the leakage as a simple resistor.  Many will show an electrical model with a simple resistor as well.  It's easy to see how someone would come to the conclusion it is linear and that any problems (damage to a capacitor) with this part of the model would simply scale. 

[all_repair]

Even ignoring damaged parts which will suffer all kinds of weird breakdown effects, leakage tests need to be done at the capacitor's rated voltage or higher; low voltage tests will only reveal gross anomalies.

The same applies to semiconductors and especially high voltage ones.  A curve tracer is especially useful for detecting faulty transistors which operate at low voltages with mysterious behaviors that become readily apparent at high voltages.

I have come across aged capacitors that the internal material has changed characteristics, i.e. insulating material is no more insulating, and vice versa.  If the capacitors have experienced spike, the internal physical structure is deformed.  I came across a capacitor that only started to short at 300V, and from the sound, it appeared that something inside started to bend at that voltage.  Different voltage is like a differrent dimension when parts have failed.  For new and unused parts, LV testing has its value I think.

[PA4TIM]

This is my box, the innerbox is double sided pcb with a layer of  (i think)  polystyrene.The copper is connected to a banana-bus. I use this for all kinda things.The leakage is 1x10¹¹ ohm with the switchoff and 4,5x10¹⁰ with everything set up for a leakage test.(I have not yet cleaned it) For the  real tests I place a teflon disc inside as a platform.

I did a quick test on a ceramic doorknob capacitor and 1000V mica cap. Also an old bumblebee (the biggest on the pcb) tested from 10 to 1000V. I measured the current with a benchmeter. from 10 to 50V the current was in the pA's. This meter has only 1nA resolution. The SMU is a HP-4329. Leakage was not lineair. Calculating current from the reading gave:
10----25---50---100---250--500---1000V,
0.5---1.5---3-------9-----33---79----200nA mica
0.25-0.6---0.9-----3---- 17---44----117nA doorknob
0.5--2.8--8.9--28.8-------------------this one in uA !!, Volt breakdown starts around  100V, current slowly rises and the meter jumps a bit so now and the,. at 250V the meter pegs in the lowest resistance range.

I have not looked up the allowed leakage or max voltage of  bumble bees. But it will cause real problems over 100V.  Testing this at 25V only would not be a good idea. I think the 2.8 uA is not a bad value for 25V volt.   

[PA4TIM]

I have come across aged capacitors that the internal material has changed characteristics, i.e. insulating material is no more insulating, and vice versa.  If the capacitors have experienced spike, the internal physical structure is deformed.  I came across a capacitor that only started to short at 300V, and from the sound, it appeared that something inside started to bend at that voltage.  Different voltage is like a differrent dimension when parts have failed.  For new and unused parts, LV testing has its value I think.

You are right, The higher the voltage, the higher the mechanical force on the "plates" with probably more risk to breakdown. 

[bugi]

That fixture looks great. Is that Teflon sheet? Who supplies Teflon sheet? Thanks

I have bought thin teflon sheets from ebay/amazon searches, though what it really is, quality, and in what shape it appears, e.g. rolled vs. used as extra filling in the package might be pure luck, depending on the seller.

For more exotic pieces, like thicker plates, sticks, cylinders, etc., I found a german seller (polyplaste) which apparently sells production leftovers on ebay.de. Took a while to find that one since I don't understand the language

[joeqsmith]

I did a quick test on a ceramic doorknob capacitor and 1000V mica cap. Also an old bumblebee (the biggest on the pcb) tested from 10 to 1000V. I measured the current with a benchmeter. from 10 to 50V the current was in the pA's. This meter has only 1nA resolution. The SMU is a HP-4329. Leakage was not lineair. Calculating current from the reading gave:
10----25---50---100---250--500---1000V,
0.5---1.5---3-------9-----33---79----200nA mica
0.25-0.6---0.9-----3---- 17---44----117nA doorknob
0.5--2.8--8.9--28.8-------------------this one in uA !!, Volt breakdown starts around  100V, current slowly rises and the meter jumps a bit so now and the,. at 250V the meter pegs in the lowest resistance range.

I have not looked up the allowed leakage or max voltage of  bumble bees. But it will cause real problems over 100V.  Testing this at 25V only would not be a good idea. I think the 2.8 uA is not a bad value for 25V volt.

That's what I was afraid of.  I tried to look up your power supply but could not find any information on it.   So far I am still testing at low voltages.   I tried an AVX tant,  TAP157K016 150uF 16V.  These have a DCL or 19.2uA max.

0v  -  10fA
2v  -  39nA
4v  -  77nA
8V  -  188nA
16v -  470nA

I then tried an old ILL Cap.  alum elec RLS series.  These were a low leakage part.  It's a 68uF 35V.   Datasheet shows 0.002CV or 3uA, which ever is greater.  So 4.76uA at 35V. 

2V  - 17fA  (I checked to make sure the thing was connected)
4V  -  50fA
8V  -  22nA
16V  -  32A
35V  -  124nA

Again, these are new old stock parts and have not been stressed.   I have a few HV caps I plan to try next.  These parts are all used pulls.   

[David Hess]

The photograph below shows eight NOS (new old stock) 25 volt 15 microfarad epoxy dipped solid tantalum capacitors of indeterminate heritage being burned in and tested for leakage.  I just hack something together as required although I would not mind having a handheld capacitor tester which can make measurements and do the same thing as I described earlier.

3.1 microamps divided eight ways is 388 nanoamps.  Typical similar capacitors are rated to have a leakage of less than about 3 microamps.

Note that the power supply is configured for a very low limit current to prevent damage to the current meter in the event that one of the capacitors shorts.  This should be done externally with say a resistor if the power supply has excessive output capacitance.

[PA4TIM]

Joe, the SMU: http://www.pa4tim.nl/?p=2269

[bugi]

What would be interesting is that even if it is non-linear, would a change in the high voltage range (i.e. increased leakage at operating and/or rated voltage) also show as an increased leakage in the lower voltages, perhaps even proportionally increased? (Checking this would of course require deliberately worsening a previously measured good cap.)

Also, would multiple caps of the same type but with slightly different leakages have curves of same shapes but proportionally different leakage levels over the voltage spans? (Or, the counter-point, could they e.g. have similar leakages at low voltages, but have proportionally larger variation at high voltages, or other way around, similar high voltage leakages but larger differences at low voltages.)

If the curve shapes stay the same (with just proportional changes), it would allow some sort of testing at low voltage, _if_ the leakage "curve" for the component being checked is known. (Probably still not 100% certainty, but perhaps better than fixed good vs. bad threshold at low voltage. And of course, would not test breakdown.)

[joeqsmith]

I had recently repaired an old radio (link below) that is roughly 80 years old.   

https://www.eevblog.com/forum/repair/old-console-radio-repair/

One of the caps was a 0.02uF 200V made by Cornell Dubilier.   Because of the low voltage, I used used the bench meters to directly read both the voltage and current. 

[oldway]

Topic very interesting but I think you make a fundamental mistake in the test procedure.

You forget an important point.
The curve that you take with your measurements is only valid if the breakdown voltage of the capacitor is higher than the rated voltage.

Then, modelling is useless if you don't know the breakdown voltage of DUT.
Breakdown voltage is an essential information in the modelling of the condensator under test.
For this reason, to calculate the leakage current at rated voltage from a low voltage test is only valid with a good condensator with breakdown voltage higher than the rated voltage.

You test the capacitors which have no problem of breakdown voltage too low and therefore you do not notice the phenomena that occur if the voltage applied exceeds the breakdown voltage.

When the breakdown voltage is reached and exceeded, the leakage current increases exponentially and if there is no current limitation, there is an arc in the capacitor that causes a catastrophic failure.

So you should do your tests until you reach the breakdown voltage even if it is higher than the rated voltage.

This is one of the reasons why ALL manufacturers always specify the leakage current at rated voltage and not at low voltage because this test ensures that the breakdown voltage is higher than the rated voltage.

A faulty capacitor is characterized by a low breakdown voltage lower than the nominal voltage and your tests on capacitors that are not defective can not measure the breakdown voltage since a good capacitor has a breakdown voltage greater than the rated voltage.

It's obvious that a low voltage test does not let you mesure, nor even guess, what's the breakdown voltage an thus is not able to clearly and surely detect  if a condensator is good or not.

[BravoV]

CMIIW, so the only valid test is the "destructive" one ? In order to know the exact breakdown voltage.

Breakdown means once this condition reached, the CUT is no longer useable right ?

[bugi]

Topic very interesting but I think you make a fundamental mistake in the test procedure.

You forget an important point.
The curve that you take with your measurements is only valid if the breakdown voltage of the capacitor is higher than the rated voltage.
...
It's obvious that a low voltage test does not let you mesure, nor even guess, what's the breakdown voltage an thus is not able to clearly and surely detect  if a condensator is good or not.

One does not need to know or find out the breakdown to model a limited range, especially if the goal is not to decide whether the cap is good or not. Note the OP's / topic's limited goal to just find more measurement results on the non-linear leakage behavior vs. some ideas that leakage would be caused by a simple (ideal) resistor with linear response to voltage. (And there is no demand on the voltage range of the measurements, though the larger the better).

Ignoring the breakdown will just result in a model that will not be as useful above the measured range (extrapolation of a model is always a risky thing). Even if the breakdown (or anything else for that matter) would start to affect leakage well below its voltage (or at any range), a simple model vs. measurements might reveal how the model starts to drift at higher voltages (or another range), leading naturally to limit models usable range. Just like every model should list the limitations. (E.g. "Note that this simple model is only valid upto 85% of breakdown voltage, whatever that voltage happens to be.")

Of course, it would make the model better to also study the behavior near and up to breakdown, so as to know how it would affect the model, and especially, how far below the breakdown it has an effect on the leakage current. This is a sort of win-win possibility: if the breakdown affects leakage only near the breakdown voltage, then the bad side is that it can not be predicted beforehand, but the good side is that the simpler model is valid for larger range.  On the other hand, if the breakdown affects leakage further below the breakdown voltage, the bad side is the simpler model has lesser useful range, but it may also become possible to predict the breakdown beforehand.  It would then depend on how the breakdown affects the leakage, if or how well it could be predicted (with a better model that includes breakdown effects).

EDIT: Personally I'd be happy just for a model that handles the non-linearity below breakdown effects better than the typical "parallel ideal resistor".

[PA4TIM]

I  agree with Bugi, breakdown is a failure and for most important to find but not predictable. They can go totally short or break down just under the VW. The only way to find one is measuring upto the WV.  It is not part of the model. Like a transistor, stating max current at 100A during breakdown in the datasheet is not very useful.

I cleaned the "farraday" box for these tests. (isolation was now 4x10¹⁰ ohm)

Cap : 1000uF-250V, Nippon Chemicon SMH . replaced because it leaked 900uA after 5 minutes. This is within specs (max 3mA for this one) but 99% from the caps I test leak less as 100 uA so I replaced it. The tester was set at 7,5 mA and 250V, the cap was 250V around 2 minutes. C was 891.1 uF, D= 0.04 at 100Hz. Datasheet states a max D of 0.2 for a worn cap.

I repeated the same  test at 25V volt (and max 7.5mA) for 5 minutes. Leakage was less as 5 uA (the tester was in the 100 uA full scale setting),

For more resolution and less noise I also used on my reformer/tester, set at 25V and max 750 uA, with a 6,5 digit Keithley 196 to measure the current. It was dancing around 100 nA. I did not timed it, just waited until it was stable. 

Line B, the same cap on the  HP SMU from 10 to 100V, Leakage at 250V was to high for this meter. It can source only few hundered uA. Every measurement was exact 5 minutes. Here the meter worked in the charge position and the Keithley measured the current. Not usable because the charge current was to low to get it charged in 5 minutes

Line C, also the HP + Keithly but now the HP in resistance measure mode and I waited until the current became constant. This results in much lower numbers under 100V but above the same problem, I did not liked the meter hammering in the corner so I did stopped at 100V.

Line D, the cap on my leakage tester on 25V, 50V, 100V, 250V. And after excact 5 minutes. Also here around 100V some problems. I have no clue why the current had a peak there. The real values in uA are 10x lower (so it fitted in the same graph.)

Line E is completely different, value is x100 in the graph. This is a 330 nF 400V Rifa miniprint tested from 0V to 350V on my Tek 576 in leakage mode, in 10V steps upto 100V, 20V steps upto 300V and then 350V.

The graphs are only useful for the trend, not the absolute values because the small number of measurements. This 5 minute tests are as interesting as seeing the gras grow. This would be the perfect job for a Keithley 2450 SMU, if I ever win the lottery I buy one.

Up to now all tests I did show a non lineair leakage.

[oldway]

CMIIW, so the only valid test is the "destructive" one ? In order to know the exact breakdown voltage.

Breakdown means once this condition reached, the CUT is no longer useable right ?

When you are testing new capacitors, you are only repeating the tests made by the quality control of the manufacturer...in this case the breakdown voltage should be above the rated voltage...It is not necessary to test the condensator above the rated voltage with the risk to damage it.

For used capacitors, the caracteristic of the curve of the variation of leakage current with variation of voltage is only valid up to the breakdown voltage that can be lower than the rated voltage if the capacitor is bad /damaged.

At breakdown voltage or above, the variation of leakage current varies much more than below.
In my opinion, variation is more or less the same as a high value resistor (linear or not ?) in parallel with one or two zener diodes... The voltage of those zener diodes is the breakdown voltage.

NB: Curve C (red) seems to show a breakdown voltage of 210V

[joeqsmith]

This 5 minute tests are as interesting as seeing the gras grow.

  Thanks for your time none the less. 

I have yet to break out the gloves but cleaning everything is important at these low voltages.   If you have a 1Tohm or so resistor, you could try and prove it's not something in your test setup.    I am continuing to manually run my tests as well, also due to lack of better test equipment. 

[joeqsmith]

10 pcs all from the same lot.  These are 0.01uF Z5U 1KV Cera-Mites.  These all started out the day brand new.  Using a 1V signal, the capacitance currently ranges from 8.9nF to 9.5.  D is between 0.005 and 0.006.  ESR from 88 to 108.   However, 8 of the 10 pcs have been vary carefully and methodically damaged  using various methods.   2 pcs remain as a control. 

I can't tell the the good from the bad with my RLC meter.   Time to see if my old equipment can detect which ones are damaged using only a 24V DC source.   I havn't tried it yet and plan to automate the test because as mentioned, this is like watching grass grow.

[oldway]

And, if possible, could you test them with an old HV tester like the Heathkit to see if you can tell the good from the bad with it ?

[joeqsmith]

First attempt at automating the testing.   I plan to sweep from 0.5 volts to 20 volts in 0.5 volt increments and will hold it at each level for 10 seconds.  The software will collect the data every half second.   

BASELINE1: Showing the current through a 20 Meg, 40 Meg, 100 Meg and 1T ohm resistors. One of the test capacitors is also shown.  The file format is the DUT#, capacitance, ESD ESR and dissipation factor.   So 20V/20M for example is 1uA which we can see is the green trace.   

BASELINE2: Showing the 1T ohm resistor only.   So 20V/1T ohm is 20nA.    I did not clean the parts and the effects can easily be seen.  It gives us some idea of what we can expect from the test fixture. 

The datasheets for these parts state the insulation resistance is 1000ohmF or 20,000Mohm.  There is no mention but I would assume the higher value, or 50nA at 1KV.  At 20V that's only a nA assuming it is linear.