Can an Op Amp Really Achieve 1 ppm Accuracy? Error Sources and Design Techniques

Can an Op Amp Really Achieve 1 ppm Accuracy? Error Sources and Design Techniques

As converters and measurement systems move toward ppm-level resolution, selecting an op amp by offset voltage, noise, or bandwidth alone is no longer sufficient. A 1 ppm target forces the designer to budget every important error source at the same time: broadband and 1/f noise, input bias current, CMRR, PSRR, open-loop gain, input-stage nonlinearity, output-stage distortion, supply impedance, and layout.

The source uses approximately -120 dBc harmonic distortion as a useful reference for 1 ppm nonlinearity. Very few commercial amplifiers achieve that level under every operating condition, so the problem becomes one of choosing a suitable device and then optimizing the application around its weaknesses.

1. Why a 'Precision' Op Amp May Still Miss 1 ppm

Different amplifier families fail in different ways. Video and line-driver amplifiers are fast but often have poor dc offset and 1/f noise. Classical precision bipolar amplifiers can have excellent dc specifications yet insufficient high-frequency distortion. Chopper amplifiers excel at offset and drift but may be limited by input current or ac linearity. MOS-input amplifiers minimize bias current but can be weaker in low-frequency noise and linearity.

A single ppm-class specification therefore does not make the complete amplifier ppm-class in a real circuit.

2. Input Noise: Broadband Noise Can Be Filtered, 1/f Noise Is Harder

With 6 nV/√Hz input-voltage-noise density and 100 kHz system bandwidth, the input-referred rms noise is about 1.9 µV. Reducing the bandwidth to 1 kHz lowers it to roughly 0.19 µV rms, or about 1 µV p-p.

Low-frequency 1/f noise is more troublesome. It is commonly specified as peak-to-peak noise from 0.1 Hz to 10 Hz and is not easily removed by simple averaging or ordinary filtering. Many op amps exhibit roughly 1 to 6 µV p-p in this region, which is already significant for ppm-level dc measurements.

Example LT1468 input voltage and current noise

Current noise also matters because INOISE multiplied by the source impedance becomes additional input-referred voltage noise. The current-noise terms at the two inputs are generally uncorrelated and do not cancel simply because resistor values are matched.

3. CMRR and Bias Current Create Hidden Nonlinearity

CMRR is not only a static dB number. Offset voltage can curve as common-mode voltage changes, and the nonlinear part appears directly as distortion. In the source example, the LT6018 corresponds to about 133 dB CMRR in its linear region, with only about 0.5 µV deviation from an ideal line across roughly 30 V of common-mode range.

LT6018 input offset versus common-mode voltage

Input bias current also changes with common-mode voltage. Once multiplied by circuit resistance, that variation becomes an additional offset and nonlinearity term. Ppm-level designs therefore require close attention to both bias-current magnitude and its variation with input level.

LT1468 input bias current versus common-mode voltage

4. Input-Stage gm Nonlinearity Limits High-Frequency Linearity

The input stage can be viewed as a differential transconductance element. Ideally gm would be constant, but in real devices it changes with differential input voltage. As signal frequency rises, loop gain falls and the amplifier needs a larger differential input voltage to produce the output, pushing the input stage further into its nonlinear region.

Current and transconductance behavior of several differential input stages

The source notes that ppm-level gain accuracy requires open-loop gain on the order of one million or more. With a 15 MHz GBW amplifier driving a 5 V p-p buffer output, a standard bipolar input stage may reach -120 dBc only to roughly 548 Hz. Alternative input structures can extend that estimate into the several-hundred-hertz to roughly 1.5 kHz range, but the overall message is unchanged: ppm linearity generally belongs to low-frequency, high-loop-gain operation.

5. PSRR and the Supply Network Can Become Distortion Sources

Load current modulates the supply rails through wiring resistance, bypass-capacitor ESR/ESL, and source impedance. Finite PSRR then couples that supply disturbance back into the signal path.

A useful example from the source is a 50 mV p-p supply disturbance with a desired residual below 5 µV p-p at the output. That requires about 80 dB of PSRR at the signal frequency. At ppm levels, the supply and current-return geometry are part of the signal chain.

6. Output-Stage Errors: Clipping, Compression, Crossover, and Asymmetry

Representative output-stage transfer curves under different loads

Output stages exhibit several nonlinear mechanisms: rail clipping, gain compression as load current rises, crossover distortion near zero, and unequal positive/negative gain under load. All become worse as the load gets heavier.

The source offers a practical rule of thumb: for good linearity, keep the required linear output current well below the short-circuit current—roughly 35% can be used as a reference limit. Very low-power op amps are especially likely to be output-stage limited.

7. Circuit Techniques That Help an Op Amp Approach ppm Performance

· Choose input-voltage noise comparable to or below the resistor network's thermal noise, and avoid unnecessarily high circuit impedance.

· Where common-mode variation is a major concern, an inverting topology can hold the input common-mode level nearly fixed and reduce CMRR-related nonlinearity.

· With bipolar-input amplifiers, a matching resistance at the opposite input can compensate bias-current-induced offset, although the added resistor contributes noise.

· Avoid forcing rail-to-rail input amplifiers through their internal input-stage crossover region when low distortion is critical.

· Keep supply-current return paths away from high-impedance inputs and feedback nodes; above roughly 100 kHz, magnetic coupling can matter.

· For heavy loads, composite or boosted amplifier structures can reduce the distortion produced by the main output stage.

8. Composite and Boosted Amplifiers

Composite-amplifier structure compared with a single amplifier

In the source LTspice example, a standalone LTC6240 produces about -78 dBc distortion at 10 kHz and 4 V p-p. Adding a broadband output amplifier inside a composite loop improves the simulated result to about -135 dBc. The source also cautions that many macro-models do not model distortion accurately, so simulation must be validated against data-sheet behavior and hardware measurements.

Boosted amplifier in which an auxiliary stage carries most of the load current

A boosted architecture lets the helper amplifier supply most of the load current while the main precision amplifier remains lightly loaded. The example improves distortion from about -78 dBc to roughly -106 dBc—nearly 30 dB—while sacrificing less bandwidth than the full composite approach.

9. Device Selection Strategy

The source identifies AD8597, ADA4807, ADA4898, LT1468, LT1678, and LT6018 as some of the easier candidates for near-ppm linearity. Other parts—including AD797, ADA4075, ADA4610, ADA4805, ADA4899, and LTC6228—can also offer strong distortion performance but may require more careful treatment of input current, impedance, or common-mode behavior.

Conclusion

One-ppm accuracy is not a label attached to a single op-amp model. It is the outcome of an error budget that includes noise, bias current, common-mode behavior, loop gain, PSRR, input-stage gm linearity, output-stage loading, and PCB current return paths. Reaching ppm-class performance requires defining bandwidth, signal amplitude, load, and temperature first, then choosing the amplifier and circuit topology around the dominant errors—and finally validating the result in hardware.