Intermittent voltage recovery after discharge in X7R capacitors

Direct answer

Yes. A residual voltage reappearing seconds after a hard discharge is the textbook signature of dielectric absorption (soakage), and X7R is one of the worst common dielectrics for it. Shorting the terminals removes the free surface charge, but dipoles deep in the Class 2 dielectric keep relaxing afterward and re-establish a measurable voltage once the short is opened. Recovery of a few percent of the charge voltage, persisting from seconds to minutes, is entirely consistent with your observation.

What is confirmed vs. likely

Confirmed behavior: Class 2 (X7R/X5R/Y5V) ceramics exhibit dielectric absorption typically in the 1–10% range, and the magnitude grows with temperature, DC bias history, and aging. So the 25°C datasheet figure, if one is even published, is not your worst case.

Likely, but worth verifying in your circuit: whether the recovered voltage actually corrupts the next sample depends on the ADC driver's input impedance and your sampling interval. A high-impedance input lets the recovery voltage climb to nearly its full open-circuit value; a low-impedance or actively driven input partially re-absorbs it. If your driver input is high impedance and you sample within seconds of discharge, treat the recovery as a real error source.

Fixing the discharge protocol

A brief low-impedance short is not enough, because absorption is a time-domain effect, not a charge-removal problem. Practical adjustments:

• Extend the discharge time. Hold the discharge path closed for much longer than the RC time constant — tens to hundreds of milliseconds, not microseconds — so more of the relaxing dipole charge is drained while the path is still closed.
• Use a controlled bleeder. A resistor sized so the capacitor decays through several time constants before sampling guarantees the terminal voltage is below your ADC's threshold. The tradeoff is sampling rate: pick R so 5–7 time constants fit inside your inter-sample gap.
• Verify at the pin, not at the switch. Measure the voltage at the ADC driver input after the discharge interval with a high-impedance probe. If it is below your acceptable error band, the protocol is sufficient regardless of the open-circuit recovery.

Substituting the dielectric

Yes — this is the cleanest fix. C0G/NP0 ceramics and good film capacitors (polypropylene, polystyrene) have dielectric absorption below roughly 0.1%, often far less, which makes the recovery negligible at sample-and-hold accuracy levels. The cost is volumetric efficiency: a C0G or film part of the same capacitance and voltage rating will be physically larger and possibly unavailable in the value you need. For a hold capacitor in a precision front end, that trade is usually correct.

Vendor-independent verification

There is a standardized method: IEC 60384-1 defines a dielectric-absorption test — charge to a specified voltage for a defined time, discharge through a short for a defined time, then measure the recovered terminal voltage after a set recovery interval (seconds to minutes). You can reproduce a practical version on the bench:

1. Charge the capacitor to your actual operating voltage (e.g., 10 V) for one minute.
2. Short it to ground for a fixed interval (e.g., 10 s).
3. Open the short and measure the terminal voltage at fixed intervals (1 s, 10 s, 60 s) with a high-impedance meter or scope probe.
4. Express the recovered voltage as a percentage of the charge voltage.

Repeat at elevated temperature (e.g., 85°C) and after the part has seen DC bias aging, since both increase absorption in X7R. Then run the identical test on a C0G or film substitute; the residual should be at or below your meter's noise floor. Comparing parts from multiple vendors under your own fixed conditions is more reliable than comparing datasheet numbers, because vendors use inconsistent test conditions or omit the figure entirely.

One caveat

These absorption percentages and time constants are well-established qualitative behavior, but your exact worst-case number depends on the specific part, bias history, and temperature. Treat any single datasheet figure as a starting point and confirm with the bench test above before committing the design.