Electronics Design AURail-to-rail op-amp output still ~200 mV short of VCC under load — is 'rail-to-rail' just marketing?
Asked by stale_biscuit_03 ·
Using an MCP6002 as a unity-gain buffer between a resistive sensor divider and an ESP32 ADC pin, single 3.3V supply. Datasheet says "rail-to-rail input and output" right on the front page, so I assumed I'd get close to a full 0–3.3V swing.
Measuring the actual output with the sensor driven to its extremes: high end tops out around 3.05V, low end bottoms out around 0.25V. That's ~250 mV of headroom lost on each rail, not the few millivolts I expected from "rail-to-rail."
Supply is clean (checked with a scope, no droop). No oscillation, no weirdness on the waveform — it just flattens out short of the rail on both ends. Is "rail-to-rail" just a marketing term that doesn't really mean what it says, or am I missing something about how these output stages actually work?
3 Replies
"Rail-to-rail" is real, but it's a load-dependent spec, not a fixed number — and that's the part almost everyone misses the first time.
A rail-to-rail output stage on a CMOS part like the MCP6002 is a pair of MOSFETs in a common-source configuration, one pulling toward VDD and one toward VSS. Unlike a traditional emitter-follower output stage, which can never fully reach the rail because of a transistor's Vbe drop, a MOSFET output stage genuinely can get very close to the rail — but "close" is set by how much current you're asking it to source or sink, because the output impedance in the saturated (triode) region behaves resistively:
V_drop ≈ I_out × R_on(triode)
Microchip's datasheet for the MCP6002 specifies this as "Output Voltage Swing" (sometimes VOH/VOL), typically measured at multiple load conditions — a lightly loaded condition (often quoted in tens of millivolts from the rail) and a heavier-load condition (often several hundred millivolts, depending on the part). That's the number to check, not the front-page "rail-to-rail" marketing claim, which is normally referring to the best-case, near-zero-load figure.
For your case specifically: an ESP32 ADC input isn't purely capacitive at the instant of sampling — the sample-and-hold capacitor draws a brief but real charging current from your buffer every time a conversion starts, and if there's any other loading downstream (a following filter, a second ADC channel sharing the same buffered rail, or just a lower-value pull resistor than expected), that current adds up. 250 mV of lost headroom on each rail is consistent with a moderate load current against the MCP6002's non-trivial output impedance — check the datasheet's Output Voltage Swing spec at the actual current your circuit draws, not at no load, before concluding anything is wrong with the part.
If you need the full swing at your actual load current, the fix is one of: choose a part with a lower rated output impedance/higher rated Iout for a given swing spec, reduce the load current the buffer has to supply, or — often simplest for a sensor front-end — accept the fixed offset and back it out in firmware, since it's a repeatable, load-dependent error rather than noise. See What Is an Op-Amp? for how the output stage sits in the overall closed-loop model.
Worth adding: that output impedance gets worse with temperature. MOSFET on-resistance has a positive temperature coefficient, so if this board ever sees an elevated ambient (inside an enclosure, near a heat source, direct sun on a field unit), the headroom loss you're measuring at room temperature on the bench will grow. Some manufacturers publish a swing-vs-temperature curve in the datasheet's typical characteristics section — worth pulling up before you calibrate a fixed offset in firmware, because a fixed offset measured once at 22°C won't track a temperature- dependent error correctly across the product's actual operating range.
One thing worth ruling out before you settle on "it's just output swing under load": double-check you're actually looking at rail-to-rail output behaviour and not an input-stage artefact. Rail-to-rail input and rail-to-rail output are two separate specifications on the same part, and it's easy to conflate them.
In a unity-gain buffer, the inverting input sits at the output voltage by definition, so if your sensor divider is genuinely swinging your non-inverting input close to either rail, a part with a weak rail-to-rail input stage can show input-stage crossover distortion or, in a worse case on some older rail-to-rail input designs, phase reversal — the output moving the wrong direction — as the input crosses the transition region between the input stage's two complementary halves. The MCP6002 is specified rail-to-rail on both input and output, so this is unlikely to be your root cause, but it's the first thing I check on any "output won't reach the rail" report before accepting the load-current explanation, because the fix is completely different (choose a part without the input crossover region, or keep the input a safe margin off the rails) if it turns out to be an input problem wearing an output symptom.