Every precision measurement is ultimately a ratio against a voltage reference. When a design measures beautifully on the bench and drifts in the field, the reference is the usual suspect — its errors multiply through the entire signal chain. Here is how the specifications interact and how to choose an IC that keeps its promises.
Initial Accuracy Is the Least of Your Problems
A 0.05 % initial accuracy sounds excellent — until you realize it can be calibrated away in one software step. What cannot be calibrated away:
- Temperature coefficient: a 10 ppm/°C reference wanders 250 ppm across an industrial −40 to +125 °C range — five times its initial error. Compare tempco numbers at your temperature span, and read the curves, not just the box.
- Long-term drift: typically quoted in ppm/√kHr — a reference that drifts 20 ppm in its first 1,000 hours keeps drifting. Burn-in helps; designs that must hold accuracy for years buy the low-drift references and still budget for periodic calibration.
- Reflow shift: soldering permanently moves the output (30–100 ppm is common). Calibrate after assembly, not before.
Topologies: Three Families, Three Jobs
| Family | Typical Tempco | Noise | Best For |
|---|---|---|---|
| Bandgap (series) | 3–30 ppm/°C | Moderate | General precision, mixed-signal, ADC/DAC references |
| Buried zener | 1–5 ppm/°C | Low, but needs supply headroom | 6.5-digit DMMs, metrology, high-end instrumentation |
| Shunt (2-terminal) | 10–100 ppm/°C | Depends on bias | Cost-sensitive bias points, negative rails, simple regulation |
Shunt references need a bias resistor and draw current regardless of load — flexible, but wasteful in battery designs. Series references regulate their own current and often include shutdown, which matters for power budgets.
Noise: The Spec That Sneaks Into Every Measurement
- 0.1 Hz–10 Hz (1/f) noise dominates DC measurements. A reference with 3 µVpp low-frequency noise sets a resolution floor no averaging can remove. Precision designs pair low-noise references with filtering.
- Broadband noise matters for AC-sampled systems; check the full specification, not the headline.
- Filtering works: an R-C between reference and ADC REF pin trades a little transient response for substantially lower noise — but verify the converter's reference input current does not droop the node between conversions.
The Interactions That Bite
- Load regulation: SAR ADCs draw charge spikes from the reference; the reference's output impedance and any series filter resistance interact. A reference rated for "±10 ppm load regulation" at steady DC may still sag during a conversion — check the converter's reference-input specifications, and buffer if needed.
- Line regulation matters where the supply is unregulated; spec it at your real supply variation, not nominal.
- Turn-on settling: references can take milliseconds to seconds to reach final accuracy; designs that duty-cycle the reference must wait or measure the settle time.
Selection Workflow
- Write the accuracy budget in ppm: initial (calibrated away), tempco over range, drift over lifetime, noise floor.
- Pick topology by precision class and power budget.
- Verify noise at your bandwidth and load-transient behavior with your converter.
- Plan calibration after reflow, and re-check after thermal cycling.
JTDZ Tech supplies series and shunt references from TI, ADI, and Renesas with full tempco and noise data — plus the precision amplifiers and converters they feed. Tell us your accuracy budget and we will quote the reference that closes it.
