Thermistor and ADC Resolution Calculator
NTC resistance, divider voltage and ADC counts across a range — plus the temperature resolution you really get, which is nowhere near uniform.
Response across the range
| Temp | Resistance | Voltage | ADC count | °C per count |
|---|
"10-bit" tells you nothing on its own
A thermistor is exponential and an ADC is linear, so temperature resolution is nowhere near uniform across a range. A 10-bit reading of a 10 kΩ NTC through a 10 kΩ resistor is good to about a tenth of a degree near room temperature and roughly half a degree at 100 °C — the same converter, the same circuit, a factor of six or more apart. Quoting the bit depth without saying at which temperature is meaningless.
- The series resistor picks where the sensor is sharp. Setting it to the thermistor's own resistance at the middle of your range puts the divider near half the supply there, which is where volts-per-degree is steepest. Sharpening the middle blunts the ends — that is a trade, not a free improvement.
- The sensor heats itself. Current through the divider dissipates power in the thermistor, and a thermistor is by definition a device whose reading changes when it warms. At 3.3 V through 10 kΩ that is about a tenth of a degree. A 1 kΩ thermistor on 5 V is over three degrees — larger than every other error combined. It is an offset, not noise, so it does not average away.
- Use the counts you have. If your range only spans a third of the converter's scale, two thirds of your resolution is being thrown away. The table above shows exactly how much of the range you are using.
- The beta equation is a two-point fit. It is exact at 25 °C and at whichever temperature beta was measured against, and drifts between and beyond them — within a percent or two over 0 to 100 °C for a typical part, and noticeably worse outside. Steinhart-Hart adds a cubic term and a third calibration point and is the accurate version.
- Your reference voltage matters as much as your bits. A divider ratio is independent of supply only if the ADC uses the same supply as its reference. If it does not, supply noise appears directly as temperature noise.
How to use
- Pick a thermistor and enter the series resistor, reference voltage and ADC bits.
- Set the temperature range you actually care about.
- Read the table: resistance, voltage, count and degrees per count at each point.
- Compare the best and worst resolution, and check the self-heating figure.
Frequently asked questions
Why is my temperature resolution worse at one end of the range?
Because a thermistor is exponential and an ADC is linear. Equal steps in ADC counts are not equal steps in temperature. A 10-bit reading of a 10 kilohm NTC through a 10 kilohm resistor is good to about a tenth of a degree near room temperature and roughly half a degree at 100 degrees — the same converter and the same circuit, a factor of six apart. Quoting a bit depth without saying at which temperature does not mean anything.
What series resistor should I use?
The thermistor's own resistance at the middle of the range you care about. That puts the divider near half the supply there, which is where the volts-per-degree slope is steepest and the reading is sharpest. It is a genuine trade rather than a free improvement: sharpening the middle of the range blunts both ends, so pick the middle of the range you actually need rather than the middle of what the sensor can do.
Does the thermistor heat itself up?
Yes, and it is a real error rather than a theoretical one. Current through the divider dissipates power in the thermistor, and a thermistor is by definition a device whose reading changes when it warms. At 3.3 volts through 10 kilohms the error is about a tenth of a degree. A 1 kilohm thermistor on 5 volts dissipates over six milliwatts and reads more than three degrees high — larger than every other error combined. It is a consistent offset rather than noise, so averaging does not remove it.
How do I reduce self-heating?
Raise the series resistor, lower the reference voltage, or power the divider from a spare output pin and only turn it on while you are sampling. The last one costs a settling delay of a few milliseconds and essentially eliminates the problem, which is why battery-powered loggers almost always do it. A part with a higher dissipation constant — a larger body, or one in contact with liquid rather than air — also helps.
How accurate is the beta equation?
It is a two-point fit, so it is exact at 25 degrees and at whichever temperature beta was measured against, and it drifts between and beyond those points. Against a real B3950 datasheet it lands within a percent or two across 0 to 100 degrees and noticeably worse outside that range. Steinhart-Hart adds a cubic term and a third calibration point and is the version to use when the accuracy actually matters.
Should the thermistor be the top or bottom leg?
Either works and the arithmetic mirrors. With the thermistor to ground, a rising temperature lowers its resistance and so lowers the reading. With it to the supply, a rising temperature raises the reading. Pick whichever gives you the more convenient direction, and note that a shorted probe reads a rail either way — which is worth detecting, since a broken sensor otherwise reports a plausible extreme temperature.
Do more ADC bits fix this?
They help proportionally — going from 10 bits to 12 gives four times as many counts and four times finer resolution everywhere. What they do not fix is the non-uniformity, since the ratio between the sharp end and the blunt end stays exactly the same. If you need even resolution across a wide range, a linearising network or a digital sensor with the curve built in is a better answer than more bits.
🔒 This tool runs entirely in your browser. Nothing you enter is uploaded, logged, or stored.