Dielectric Soakage Memo 

WHAT IS THIS MEMO ABOUT?

             You asked me to characterize a handful of 871.0714.972 polyester capacitors with regard to dielectric absorption (AKA “capacitor memory”, AKA “soakage”). 

 WHAT IS DIELECTRIC ABSORPTION?

This is an error term due to polarization of dipoles at dielectric interfaces.  Messrs. Horowitz and Hill (authors of The Art Of Electronics) claim that dielectric absorption is not understood. 

 In the hierarchy of a capacitor’s undesirable features, this property generally is fourth or fifth (behind tempco, leakage, ESR, ESL etc.).  When the application involves fast sample and hold circuits or dual slope A/Ds, dielectric absorption jumps right to the top of the list of things that you do not want.

 HOW DOES DIELECTRIC ABSORPTION HURT US? 

The research that you asked me to do on this component has to do with the dual slope A/D in the model 787 instruments.  I will place this discussion in that context. 

Electronics hobbyists first discover dielectric absorption when capacitors that they have carefully discharged bite them anyway.  This is particularly true for vacuum tube guys (a fixed percentage of lots of volts is STILL lots of volts).   The model of dielectric absorption is a parallel combination of series RC networks with the “real” capacitor. While one can short out the “real” capacitor, the parallel networks discharge at their own rates (determined in each case by the magnitude of the R and the C).  When the short is removed, the charges on the parasitic elements move back to the “real” capacitor.  This works in both directions.  If one starts with a discharged state, all of the parasitic elements start in the discharged state.  Upon charging the capacitor, the parasitics pull charge from the “real” capacitor. 

All of this is of import in that we expect the movement of charge out of C109 to be related only to the potentiostat’s sink current.  Any other charge movement is an error term.  Combine this with the fact that errors caused by dielectric absorption are related the ratio of charge time to discharge time and our problem becomes evident given that charge and discharge time in a dual slope A/D directly affects the conversion result.  This ratio in our case is big.  The instrument has set for hours (days, weeks?) with no power applied to the A/D.  This assures that all of the parasitic networks in the model for dielectric absorption are completely discharged.  We then charge the capacitor in a period measured in milliseconds.  All of those parasitics try to “pull” the voltage on the capacitor back to ground.  If this happens during the auto-zero time, the resulting conversion reads low (we subtracted too big a number when calculating corrected counts).  When this occurs during an electrode current conversion time, the resulting conversion is too high.  The sequencing of the A/D is in the hands of the bits guys.  I do not remember what order things happen in.

 WHAT CAN WE DO ABOUT DIELECTRIC ABSORPTION?

             We can choose a good capacitor.  Mica, high dielectric constant ceramics, polycarbonate and polysulfane are among the worst capacitors for soakage.  Teflon, polystyrene and polyester are (in that order) among the best.  I have no data to this effect, but I would guess that an air capacitor would probably be the best.  Of course a double plate air capacitor of the value needed would probably be as big as Rhode Island.  The 871.0714.972 has a polyester dielectric.  A polystyrene dielectric might improve the situation.  I believe they are more costly.  Teflon is lots more costly and would require a capacitor as big as my fist.  As with all of our “high value” product, the trick is to be good enough.

             What we do do is fire the A/D a couple of times.  This lets the capacitor “see” the one volt bias that capacitor will see (on average) during the ramp up and ramp down times.  The “on average” part of that statement is problematic.  The dual slope counts on changing capacitor voltages in clearly identified ways.  Soakage tries to put things back the way they were way in the past.

 HOW DID I MEASURE THE CAPACITORS?

             I stole and augmented a circuit from Mr. Pease.  Basically, this circuit buffers the capacitor’s voltage with a leakage compensated circuit so that one can observe soakage without influencing it by measuring it.  One starts by adjusting the leakage circuit for a zero dv/dt (this implies that the leakage from the amplifier precisely matches the leakage of the capacitor).  I charged the capacitor to 10.0 volts for one minute, shorted them (through a 5 ohm resistor) for six seconds and watched the soakage voltage for one minute.  All timing was done by me, my technician and a stop watch.  The output was read by the LeCroy oscilloscope and plotted.  I looked at the ten capacitors that you gave me and at some others (to provide some perspective)   

WHAT DID THE MEASUREMENTS SHOW?

            The plots that I did are available in my Lab Notebook.  Here’s my interpretation of the data:

� measured in each polarity

‚ I would call this to be at the limit of my ability to measure.

             I don’t see anything that I could call anomalous behavior among the samples of the 021.0014.012 that you gave me.  Anyway, you at least have some characterized samples now ….

 HAVE I QUANTIFIED THIS MEASUREMENT’S EFFECT ON OUR PRODUCT?

            No.

WHAT CAN WE DO TO IMPROVE THE SITUATION?

            Compensation is out of the question ($).  We could look at a polystyrene (probably $).  I doubt that the time is right for this from a business standpoint.  Would it make sense to measure some of the caps from problematic instruments?  Or is that what I did?

HERE DID I GET MY INFORMATION

1.      Horowitz and Hill, The Art of Electronics, second edition

2.      Robert Pease, Understand capacitor soakage to optimize analog systems, EDN October 13, 1982

Back