Crystal reference aging - how much drift over 10 years?

Crystal references age, and 10-year drift specs are often hand-waved. Has anyone tracked reference drift over a full product lifetime? How much did you have to budget for?

Crystal aging is real, but the "10-year drift" number on a datasheet is often a worst-case envelope, not what you will measure on one part in one oven.

Practical framing on the bench:
1. Separate mechanisms: frequency vs time has an early stabilization (days-weeks after power-up / mount stress), then a slower aging slope. Quoting only the end-of-life ppm without saying whether the part was pre-aged or how long it had been powered is why two labs disagree.
2. Temperature first: many "aging" complaints are uncorrected tempco or hysteresis around a poorly controlled oven / XO board. Log frequency vs case temperature (or oven setpoint) before blaming aging.
3. Stress / mount: epoxy, solder, and board flex shift the resonator; a part that was reflowed hard can look "aged" in the first weeks. If you can, compare a virgin socketed sample vs a soldered one from the same lot.
4. For a 10-year budget: use the vendor aging slope after the stabilization window (often ppb/day, then integrate), add temp, supply, and load, then leave margin for radiation / shock if that is in the use case. Blindly taking "+/-X ppm / 10 years" as a Gaussian yearly rate usually overstates year-1 and understates the early transient.
5. Field check: a quiet secondary reference (even a GPSDO when available) and a time-stamped beat note or phase-meter log beats a one-shot counter reading.

If you share the cut (AT/SC), package, whether it is OCXO/TCXO/XO, and the temp you hold it at, people can usually say whether your observed slope is in-family or a mount/temp artifact.

— rfforum 编辑部 / Editorial Team