Self-Discharge Rate of Various Capacitors

Using a high-impedance input described on the previous page, I tested a variety of capacitor chemistries, values, and ages. The ultimate goal is to discover how quickly a capacitor will lose power on its own, so that I can calculate the capacitance needed to power a circuit overnight.

Capacitors tested

Capacitors tested.

  1. Polyester film 10 nF
  2. Ceramic disc 10 nF
  3. Mica 10 nF
  4. Multilayer ceramic 0.1 µF
  5. Stacked film 0.33 µF
  6. Multilayer ceramic 1 µF
  7. Tantalum 1 µF
  8. Aluminum electrolytic 1 µF
  9. Metalized polyester 1 µF
  10. Aluminum electrolytic 10 µF
  11. Aluminum electrolytic 220 µF
  12. EDLC (Electric Double-Layer Capacitor) / ultracapacitor 1 µF
  13. EDLC (Electric Double-Layer Capacitor) / ultracapacitor 400 µF

Each capacitor was charged to 2.5 volts using an MCP1700 voltage regulator, disconnected from the voltage source, and then measured over a period of 25 minutes using a high-impedance MCP6S22.

My expectation was that each type of capacitor chemistry (plastic, ceramic, electrolytic) would lump together, with only slight differences due to capacitance value.

Self discharge of various capacitors

Self discharge of various capacitors (flawed test).

Something is wrong. Generally speaking, the higher capacitance capacitors appear to discharge more slowly. This shows that the overall test circuit is drawing sufficient current to negatively alter the results. Why? I thought this was a 10 trillion ohm input?

Close inspection of the PCB shows flux residue from soldering.

Flux residue fibers and other contaminants

Flux residue, fibers, and other contaminants.

This is “no-clean” flux, which means it will not harm the board or electronics if left in place. However, that doesn’t mean that the flux has no effect. Although the flux is electrically-resistant enough for most circuits, it allows sufficient current to flow to alter the results of this highly sensitive test. Furthermore, flux is sticky enough to trap hairs, fibers, and other particles that may also conduct electricity or alter electrical characteristics.

There are other things that allow teeny tiny amounts of current to flow. The printed circuit board GR10 substrate, DIP socket, and header connector insulator are highly resistant, but still allow enough electricity to flow that they drain low capacitance capacitors. The only way to avoid those sources of leakage is to lift the input pins out of the socket and connect directly.

Lifting DIP pins for high impedance

Lifting DIP pins for high impedance.

Let’s see if cleaning the flux and connecting directly to the DIP pins makes a difference.

Effect of test circuit changes on 10 nF capacitor discharge measurements

Effect of test circuit changes on 10 nF capacitor discharge measurements.

Measuring the same 10 nF capacitor under the three different conditions shows that the time to discharge between 2.5 V and 2.3 V took 155 seconds, 293 seconds, and 1041 seconds respectively. So, cleaning the PCB or connecting directly to the input pins does make a difference.

A little bit of math calculates the total resistance (including self-discharge) under those conditions went from 185 billion ohms, to 350 billion ohms, to 1,250 billion ohms. Solving for parallel resistance indicates that the flux residue had a net effective resistance of 392 billion ohms and everything else (connectors, dip socket, PCB) has a net effective resistance of 486 billion ohms, with respect to ground. That’s two orders of magnitude more conductive than the input pin of the measurement chip.


Could the difference between measured voltages be due to other factors, such as changes in the quality of the capacitor between tests, electrical disturbances, or the specific pin on the chip that is used for measurement? I tested two 10 nF ceramic disc capacitors from the same manufacturer and same batch at the same time. The tests were repeated three times, switching the chip pins after each test.

Electrical noise repeated tests and pin swapping

Electrical noise, repeated tests, and pin swapping.

All six lines should be overlapping. I notice three things:

When you’re dealing with such large resistances, such small capacitance, and such high impedance, then the slightest electrical interaction can alter the measurement results. In other words, even my best effort at creating a test setup is too unreliable to take these results seriously.

I’m going to go ahead and show the rest of my results. However, you need to keep in mind that the apparatus or test method has a problem with accuracy and repeatability.