Engineering Guides
MEMS Pressure Sensor Datasheets — What Those Specs Actually Mean (And Which Ones Matter)
You've got three datasheets open in three tabs. One says "accuracy ±0.03 hPa." Another says "±1.5 hPa over temperature." A third says "Total Error Band ±0.5% FS." They're all pressure sensors. They all quote accuracy. And the numbers mean completely different things. This article is the decoder ring.
The One Sentence That Frames Everything
Every pressure sensor spec is answering one of two questions:
- "How wrong could this reading be?"
- → Accuracy, TCO, drift, noise, Total Error Band
- This is where the confusion lives.
- "How quickly and reliably can I get the data?"
- → Conversion time, interface speed, power states
- Easier to compare directly.
Il y a five different ways to say "accuracy" and manufacturers cherry-pick whichever one makes their part look best. Here they are, from most misleading to most honest.
Précision relative
aka "Noise Floor" AccuracyWhat it actually means
In ideal conditions — one temperature, one pressure point, right after calibration, no external stress — how precisely can the sensor resolve a pressure change?
Numbers you'll see: ±0,03 hPa ±0.015 hPa ±1 Pa — impressively small.
The catch: This assumes a temperature-controlled lab, at the exact calibration pressure, on a PCB that hasn't warped by a single micron. It tells you how sensitive the sensor is — its resolution ceiling. It does not tell you how accurate your field measurement will be. It's the spec equivalent of a car's horsepower measured on an engine dyno with no exhaust, no accessories, and ideal fuel.
- Measuring changes in pressure, not absolute values
- Altitude tracking on a drone (25 cm height change)
- Leak detection (pressure decay curve)
- Measuring "what is the pressure right now" vs. a calibrated reference
- Any application requiring absolute accuracy
Absolute Accuracy
The Real OneWhat it actually means
Put the sensor at a known pressure and known temperature, take a reading, compare to a calibrated reference standard — how far off is it?
Numbers you'll see: ±1,5 hPa ±5 hPa ±15 hPa — much bigger than relative accuracy.
The catch: This includes manufacturing variation, calibration residuals, and non-linearity — all the things relative accuracy leaves out. But it's still usually quoted at a single temperature point (often 25°C). Change the temperature and this number degrades. More on that next.
Real-world reference: The WF5803F 01BA variant quotes ±1.5 hPa absolute accuracy at 25°C. That's the number to use when asking "will this sensor read the right pressure out of the box?"
- Barometric pressure measurement
- Weather stations
- Raw pressure feeds into altitude/air density calculations
- Relative-change applications
- When you're doing a system-level calibration after assembly — that cancels out much of the absolute error
Temperature Coefficient of Offset (TCO)
The Field KillerWhat it actually means
For every 1 Kelvin the temperature changes, how much does the sensor's zero-point drift? Expressed in pascals of pressure-equivalent drift per degree.
Numbers you'll see: 1.5 Pa/K 3 Pa/K 10 Pa/K
The catch: This is the one that bites engineers who test on the bench at 22°C and wonder why field data looks wrong at −20°C or +60°C. A TCO of 1.5 Pa/K over a 40°C swing shifts your reading by 60 Pa ≈ 0.6 hPa ≈ 5 metres of altitude error. For a drone hovering at a set altitude, that's the difference between "stable" and "slowly drifting into a tree."
Pro tip: Check whether the sensor has on-chip temperature compensation. The WF5803F outputs 24-bit temperature data alongside pressure, with factory calibration coefficients in NVM — the 1.5 Pa/K spec is the compensated figure. The raw, uncompensated element may drift 10× worse.
- Outdoor applications with diurnal temperature cycles
- Engine compartments
- High-altitude UAVs
- Anything not in an air-conditioned room
- Indoor consumer devices
- Lab equipment in temperature-controlled environments
- Short-duration measurements where temperature doesn't change
Long-Term Drift
The Invisible ProblemWhat it actually means
If you leave the sensor powered on for a year — same temperature, same pressure, same supply voltage — how much does the reading shift from its initial calibrated value?
Numbers you'll see: ±0.1% FS/year ±3 hPa/year — or sometimes "not characterized" (translation: we didn't test it), or nothing at all (translation: we really don't want you to know).
The catch: This is the hardest spec to get because it requires a 1,000-hour accelerated life test. Manufacturers that publish it have done the reliability work. Manufacturers that don't — you'll be doing that testing yourself. Drift comes from mechanical relaxation of the silicon diaphragm, moisture ingress, and stress relaxation in the die-attach adhesive. None of it can be compensated in software because it's irreversible.
Red flag: If a datasheet has no long-term drift spec and no reference to accelerated life testing, treat it as a commodity part. Fine for a consumer gadget; not fine for an installed sensor you can't service.
- Installed sensors that can't be recalibrated
- Infrastructure monitoring
- Medical devices held to calibration for the product lifetime
- Consumer devices with 2-year replacement cycles
- Applications with zero-point calibration at startup (vent, read, subtract offset)
Noise Density
24-bit ≠ 24 Usable BitsWhat it actually means
The sensor's intrinsic electrical noise, expressed as pressure-equivalent noise spectral density. Usually in Pa/√Hz or hPa RMS. Sometimes listed; sometimes buried on page 14 of the datasheet; sometimes not there at all.
The catch: A 24-bit ADC does not give you 24 bits of pressure resolution. The bottom several bits are noise. Example: if pressure noise is 0.5 Pa RMS at 100 Hz output rate and full-scale is 700,000 Pa (7 Bar), effective resolution is log₂(700,000 / 0.5) ≈ 20.4 bits. The bottom 3–4 bits are random. Averaging reduces noise by √2 per doubling of samples — but you pay in response time.
Practical takeaway: Don't get hypnotized by "24-bit ADC" on page one. Look for the noise spec. No noise spec? Budget time on the bench to characterize it yourself — because you will care eventually.
- High-speed pressure measurement (no time to average)
- Low-pressure applications where full scale is small
- Slow, averaged measurements (weather stations at 1 sample/min)
- High-pressure applications where noise is a tiny fraction of full scale
Total Error Band (TEB)
The Honest OneWhat it actually means
Take all error sources — absolute accuracy, TCO over the full operating temperature range, non-linearity, hysteresis, long-term drift, and power supply rejection — combine them RSS or worst-case. The result is the maximum error the sensor will produce in the real world.
Numbers you'll see: ±0,5 % FS ±1% FS ±2.5% FS — always the biggest accuracy number in the datasheet.
The catch: TEB is the honest spec — and the one manufacturers least want to highlight. A sensor with ±0.03 hPa relative accuracy and ±5 hPa absolute accuracy over temperature might have a TEB of ±10 hPa or more once you factor in drift, supply voltage variation, and mounting stress.
TEB is the number you should use for worst-case tolerance analysis. Not relative accuracy. Not absolute accuracy at 25°C. TEB.
- Worst-case tolerance analysis
- Safety-critical systems
- Any system that must guarantee ±X hPa under all operating conditions
- Systems with a calibration step (offset/gain errors get cancelled)
- Applications needing precision (repeatability) not accuracy (closeness to true value)
- Measuring changes over minutes rather than absolute values over years
The Three Specs You Actually Need to Read
If you only have 30 seconds per datasheet, look at these in order:
Absolute Accuracy at 25°C
The sensor's fundamental quality — how well it was calibrated at the factory. For a 7 Bar sensor: under ±0.2% FS (±14 hPa) is solid. Under ±0.1% FS (±7 hPa) is excellent. If this number is bad, nothing else matters.
TCO or "Accuracy Over Temperature"
The difference between "works on my bench" and "works in the field." If the datasheet only quotes accuracy at 25°C, they're hiding the temperature performance. The WF5803F's compensated TCO of 1.5 Pa/K means a 60°C swing adds less than 1 hPa of offset error.
Does the Datasheet Show a Noise Plot?
Not the value — the mere existence of FFTs, Allan deviation curves, or noise density vs. frequency plots tells you the manufacturer characterized the dynamic behavior. Written for engineers, not procurement managers.
When comparing sensors with inconsistent specs: default to the one that publishes the most of these, not the one with the best single number. Transparency signals confidence in the part.
The Small Print That Isn't Small
Three parameters that don't make the headline spec table but will absolutely ruin your design if ignored:
MEMS sensors are mechanical structures. Reflow too hot, too long, or too many times and the diaphragm stress-relaxes — permanently shifting the offset. Most specify JEDEC J-STD-020, peak 260°C, one reflow cycle. Exceed it and your accuracy spec is void. Hot-air rework after that? Your call.
The LGA package is soldered directly to your PCB. If the board flexes — from mounting screws, thermal expansion, or someone leaning on the enclosure — that stress couples into the sensor die through the solder joints. Sometimes called "board-bending sensitivity." Not always in the datasheet.
LGA packages absorb moisture. Expose them too long before reflow and trapped moisture vaporizes during soldering, delaminating the package ("popcorning"). Most MEMS pressure sensors are MSL-3: 168-hour floor life after opening the sealed bag. Follow the bake-out instructions or budget for scrap.
Spec Cheat Sheet
| Spec | What It Tells You | Read It When |
|---|---|---|
| Précision relative | Sensor resolution in ideal conditions | Measuring changes, not absolute values |
| Absolute Accuracy | Out-of-box accuracy at room temp | Measuring true pressure; no system-level cal planned |
| Coût total de possession | How much temperature affects zero-point | Outdoor use; wide temp range; unheated enclosures |
| Long-Term Drift | Will it still be accurate next year? | Installed sensors; no recalibration budget |
| Noise Density | How many ADC bits are actually real | High-speed sampling; low-pressure ranges |
| Total Error Band | The worst-case real-world error | Tolerance analysis; safety-critical systems |
TL;DR
- Précision relative — looks impressive, only relevant for change-detection applications.
- Précision absolue — the real out-of-box number. Always at 25°C; always degrades with temperature.
- Coût total de possession — the field killer. Any outdoor or wide-temperature application lives or dies by this spec.
- Long-term drift — if it's missing from the datasheet, ask why. Then test it yourself.
- Noise density — "24-bit ADC" means nothing without this. Find the plot or measure it.
- Total Error Band — the honest number. Use it for worst-case analysis. Not relative accuracy.
- The rule: choose the sensor that publishes the most specs, not the best single number.
Still comparing datasheets? Tell us your application — range, temperature, environment — and we'll tell you which specs actually matter for your design, usually within an hour.
Get in Touch →