Biased long-term drift testing was performed on the automotive-grade RES11A-Q1 using a temperature-controlled oil bath. The devices under test were first soldered to the bias boards using a reflow oven, per J-STD-020E, then were cleaned in an ultrasonic bath. The boards underwent an additional bake step before placement in the oil bath. The ambient bath temperature was fixed at 46°C and each divider was biased to a fixed voltage of 15V. Devices were allowed to soak for 45 minutes, to achieve thermal equilibrium, before measurements commenced.

An integration circuit was used to measure current through each divider, and thus identify any changes in end-to-end resistance. A delta measurement between the midpoint pins of the dividers provided a high-resolution measurement of divider-to-divider shifts for the same DUT, allowing approximation of shifts in t_M. Additional measurements between the midpoints and ground were performed, to be used in conjunction with the previous measurements for calculation of shifts in t_Dx. The bias voltage was also measured for each channel. These measurements were implemented using multiplexed digital multimeters (DMMs).

Measurements were recorded at intervals ranging from 15 minutes to 24 hours, with run durations from 1000hrs to 3600hrs. In some instances, measurement data collection was interrupted due to external factors, such as mandatory system updates to the computer used for the measurements. In these instances, the bias voltage and temperature control were not disturbed due to the use of uninterruptible power supply (UPS) backups for the biasing circuitry; there was simply an absence of recorded data from the DMMs. Plots presented are normalized to the respective run mean.

Accurate measurement of errors in the single-digit ppm range requires careful consideration of circuit parasitics. Noise sources can easily contaminate or dominate measurements. While external filters at the measurement nodes assist with high-frequency noise, the thermal noise of the filter resistors contributes low-frequency noise, so a series of design and architectural tradeoffs are required. Ongoing development of biasing boards, procedures, and facilities continues to increase the effective resolution of these measurements.