Measuring Voltage Reference Thermal Hysteresis After a Temperature Excursion

A voltage reference can return to its original ambient temperature without returning to its original voltage. This is thermal hysteresis, a history-dependent shift that a temperature-coefficient calculation does not capture. It matters when a board is calibrated in one condition, experiences shipping or service temperature excursions, and later resumes precision measurements.

Compare equal endpoint conditions

Analog Devices describes package and board stress effects in Design Note 229. Build a test that measures the same reference at the same endpoint temperature before and after an excursion. Comparing a cold voltage directly with a hot voltage measures temperature response, not hysteresis.

For example, establish a stable baseline near the calibration temperature, run the board through its intended nonoperating temperature excursion within every component’s limits, and then remeasure after returning to the baseline condition. The resulting difference must be separated from meter drift, changed loading, and incomplete thermal equilibrium. A nearby sensor indicates board temperature but does not prove the reference die has settled.

Control the history

Record whether power remained on, the excursion endpoints, dwell times, ramp conditions, and the elapsed time after return. Use an uncycled comparison board or stable measurement reference where practical. Alternate measurements between the device under test and that reference to reveal drift in the measurement chain.

• Fix the reference input voltage and output load.
• Use the same board mounting and connector arrangement for both endpoint readings.
• Measure more than one device when estimating population behavior.
• Preserve individual before-and-after pairs rather than averaging unrelated boards.
• Report repeat-cycle results separately from the first excursion.

Translate the shift into system error

A hypothetical 20-ppm reference shift produces about a 20-ppm scale change in a converter whose transfer function is directly proportional to that reference. The sign and effect depend on whether the system is generating or measuring voltage and whether it is ratiometric. Recalibration may remove the current offset in scale, but it does not guarantee immunity to the next excursion.

Consider package, board strain, thermal environment, and calibration policy together. Keep this result separate from reference startup settling and from the instantaneous ADC reference drive requirement. A useful release record states the allowed post-excursion error and the waiting condition under which that error was measured.