In short: Chopper and auto-zero amplifiers trade bandwidth for near-zero DC offset and almost no 1/f (flicker) noise by modulating the signal to a high frequency where offset and flicker cannot follow, amplifying it there, then demodulating it back — giving sub-µV offsets and offset drift in the tens of nV/°C range.
Every real op-amp carries an input-referred DC offset and a low-frequency "1/f" noise skirt that together swamp tiny signals — a 50 µV bridge output, a thermocouple drifting µV/°C, a current-sense shunt dropping hundreds of µV. A good trimmed bipolar amp still leaves 25–100 µV of offset and a 1/f corner near 1–10 Hz. For precision DC and sub-Hz measurements that is fatal. Chopper and auto-zero techniques attack the problem not by building a quieter transistor, but by moving the measurement away from where offset and flicker live.
Why offset and 1/f noise are the enemy
Flicker noise has a power spectral density that rises as 1/f toward DC. Below the flicker corner it dominates thermal noise, so a straight amplifier integrating a slow signal accumulates unbounded low-frequency error. Offset is simply the DC limit of this problem — a fixed error indistinguishable from the signal itself. Neither is improved by averaging over seconds, because the error is correlated across that whole window.
The key insight: a DC signal and the amplifier's offset/flicker error both sit at DC, so no filter can separate them. Modulation spectrally relocates the wanted signal so it no longer overlaps the error.
Chopping: modulate, amplify, demodulate
A chopper amplifier places a switching network (a "chopper") that multiplies the input by a ±1 square wave at a chopping frequency fchop, typically 10 kHz–1 MHz. The DC input becomes an AC signal centered at fchop. The amplifier gains it up — but the amplifier's own offset and 1/f noise stay at DC. A second chopper, synchronous with the first, demodulates the amplified signal back to DC, while simultaneously up-converting the amplifier's offset and flicker to fchop, where a simple low-pass filter removes them.
Watching the signal and error swap places
The animation below shows the core trade: the blue "signal" packet is relocated to fchop for amplification and returned clean to DC, while the red "offset/flicker" error is simultaneously shoved up to fchop and discarded.
Auto-zeroing: sample the error, then subtract it
Auto-zero (AZ), also called correlated double sampling, works in two phases. In the nulling phase the amplifier's inputs are shorted (or routed to a reference) and its offset-plus-flicker error is measured and stored on a capacitor or in an auxiliary nulling loop. In the amplify phase that stored correction is subtracted from the live signal. Because flicker noise is slowly varying, the sample taken microseconds ago is still highly correlated with the present error, so the subtraction cancels most of it. AZ is a sampled (time-domain) technique; chopping is a continuous (frequency-domain) one.
Chopper-stabilized hybrids and the aliasing catch
Modern precision parts — Analog Devices' ADA4522, Texas Instruments' OPAx188 and the venerable LTC2057 — often combine both: a continuous-time chopper for the main path plus auto-zeroing to suppress chopper ripple and residual charge-injection. The price is subtlety. Auto-zeroing samples, so wideband noise folds (aliases) back to baseband, and chopping injects a ripple tone at fchop plus charge-injection glitches at each switch edge. Designers add ripple-reduction loops, notch filters, or guard banding, and simulate the switched behavior in a transient/PSS (periodic steady-state) analysis in Cadence Spectre or an equivalent — a plain AC analysis will not reveal chopper ripple or clock feedthrough.
| Property | Chopper | Auto-zero |
|---|---|---|
| Domain | Continuous, frequency | Sampled, time |
| Offset | <1 µV typ. | 1–5 µV typ. |
| 1/f noise | Essentially removed | Strongly reduced |
| Main artifact | Ripple at fchop | Aliased wideband noise |
| Bandwidth hit | Signal BW < fchop/2 | Limited by phase rate |
A chopper-stabilized amp can reach offset drift of 10–50 nV/°C — roughly 100× better than a trimmed precision bipolar op-amp — which is why they anchor weigh scales, medical front-ends, and high-side current sensing. These ideas sit next to broader noise behaviour in analog circuits and the instrumentation amplifier design that usually wraps around them.
Where it matters in practice
Consider a 2 mV/V load cell at 5 V excitation feeding a 10 mV full-scale bridge. A 25 µV op-amp offset is 0.25% of full scale — unacceptable for a trade-legal scale. A sub-µV chopper drops that to under 0.01% and, crucially, the drift stays flat across temperature so the system does not need constant recalibration. The same reasoning drives Hall-sensor front-ends, thermopile IR sensors, and precision DACs/ADCs where the reference buffer must not wander. When you design such front-ends you can get hands-on, lab-driven help from analog-focused training like ChipXpert's VLSI and analog programs.
FAQ
Does chopping eliminate thermal (white) noise too?
No. Chopping and auto-zeroing target offset and 1/f (flicker) noise, which live at low frequency. Broadband thermal noise is flat across frequency, so it is unaffected — and auto-zero can slightly raise baseband noise through aliasing. Pick device sizing and bias current for thermal noise separately.
What sets the usable signal bandwidth?
For a chopper the signal bandwidth must stay well below fchop/2 to avoid overlap after demodulation; a 100 kHz chop typically supports tens of kHz of clean signal. Auto-zero bandwidth is bounded by how fast you can cycle the null-and-amplify phases.
Why does chopper ripple appear and how is it removed?
The up-converted offset reappears as a residual tone at fchop plus switch charge-injection glitches. A notch or ripple-reduction loop (a feedback path that senses and cancels the tone) or an added auto-zero stage suppresses it.
How do I verify the design in simulation?
Use transient and periodic steady-state (PSS/pnoise in Cadence Spectre, or the equivalent in Synopsys HSPICE) — the switching makes DC and small-signal AC analyses misleading. PSS captures the folded noise and ripple that define real performance.
Want to design precision front-ends end to end — from noise budgeting to switched-cap layout? Explore the analog track in the CourseTron course catalogue and build real chopper and auto-zero blocks in simulation.
References & Further Reading
- Chopper (electronics) — Wikipedia
- Analog Devices — Auto-Zero and Chopper Amplifiers technical article
- Texas Instruments — Understanding Chopper and Zero-Drift Amplifiers (SBOA269)
- Cadence Spectre Simulation Platform (PSS/pnoise)
- Enz & Temes, "Circuit techniques for reducing the effects of op-amp imperfections," Proc. IEEE (IEEE Xplore)