ADR1000 10 V Reference

Why a buried-Zener reference

The ADR1000 is not a normal reference chip. It puts a buried Zener, a temperature-sensing transistor and a heater on one die. External circuitry holds that die at a constant elevated temperature and controls the Zener current, so the reference sits in its own thermal environment regardless of what the room is doing.

That is the reason this class of part turns up inside calibrators and 8½-digit meters rather than ordinary power supplies. The figure of merit is not how close the raw output sits to a round number. It is how little that output moves over time and temperature.

Measured, not assumed

The board originally ran an LTZ1000 and was later changed to an ADR1000. The two are pin-compatible, but they do not sit at the same voltage, so the swap was not a drop-in. The entire 10 V scaling network had to be recalculated around the new device.

Rather than assume a nominal value, the board was left to stabilise and the gain network was calculated from the voltage it actually produced. These parts are stability references, not factory-trimmed sources; every individual device lands somewhere slightly different. The absolute value can be calibrated later. What matters is that it stays put.

Two boards were built and characterised this way, assembled about a week apart, with a separate diagnosis session afterward. Building a second unit and comparing it against the first is exactly the stability argument this page is making: a design that only exists once cannot show whether its behaviour is repeatable or coincidental.

No trim potentiometer

The usual way to land on exactly 10 V is a trimpot. This design avoids one. A mechanical wiper is another component that drifts with temperature, vibration and age, sitting in the middle of a circuit built for stability.

Instead the gain is set by a fixed network of bulk metal foil resistors, with additional branches placed in parallel to correct the ratio. The resistor technology is the point: foil parts reach single-digit and sub-ppm/°C temperature coefficients, far better than a trimmer wiper. To choose values I wrote a small solver that searches combinations of the foil values available and only proposes networks that can actually be populated on the board.

The trade-off is real and worth stating: changing the reference device means recomputing and repopulating resistors rather than turning a screw. That is exactly what the LTZ1000-to-ADR1000 change required. For a board whose whole purpose is holding still, that is the right side of the trade.

What the numbers mean

The solver reports a resistor-fit residual far below a ppm. That figure describes how closely the resistor arithmetic lands on the target. It is not the accuracy of the finished board, and presenting it as such would be misleading.

The real error sources are much larger. The reference has its own noise. The amplifier contributes offset. Real resistors sit somewhere inside their tolerance bands rather than exactly on nominal. Thermal EMFs at dissimilar-metal junctions generate microvolts on their own, and the meter carries its own calibration uncertainty: the 3458A's 24-hour 10 V DC specification is 0.5 ppm of reading plus 0.05 ppm of range, and factory traceability to NIST adds roughly 2 ppm of reading on top of that — call it 2.5 ppm from the meter alone, larger than the 0.7 ppm deviation being reported above.

Any claim of sub-ppm accuracy would require a full traceable uncertainty budget. What is shown here is a development measurement, not a specification.

Grounding

At this level a ground plane is not automatically a single node. The heater draws a relatively large and varying current; if that current shares copper with the reference return, the resulting drop lands straight on the measurement.

The board keeps the reference, heater, power and output returns separate and joins them at one deliberate point, rather than letting them reconnect somewhere in the supply.