Current Sense Shunt Design and Error Calculator
Enter your current range, shunt and amplifier. The calculator returns dissipation and the error at both ends of the range — separating the gain errors that stay constant from the offset errors that grow without limit as current falls.
| Error term | At full scale | At minimum current | Behaviour |
|---|---|---|---|
| Shunt tolerance | gain error — constant % of reading | ||
| Shunt tempco | gain error — constant % of reading | ||
| Amplifier gain error | gain error — constant % of reading | ||
| Amplifier offset | offset — dominates at low current | ||
| ADC quantisation | offset-like — also worse at low current |
Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.
Why current sensing is accurate at the top and terrible at the bottom
Current-sense errors split into two families that behave completely differently, and conflating them is why a design that measures 0.5% at full scale turns out to be 30% wrong at low current.
Gain errors — shunt tolerance, shunt tempco, amplifier gain error — are a constant percentage of the reading. They look the same at 10 A and 10 mA.
Offset errors — amplifier offset, ADC quantisation — are a fixed number of microvolts. As a percentage of reading they grow without limit as the current falls. An amplifier offset of 50 µV across a 5 mΩ shunt is equivalent to 10 mA of current, permanently, whether you are measuring 10 A or 20 mA.
That is why the table gives both columns. The full-scale figure is the one datasheets quote and the low-current figure is the one that decides whether your sleep-mode measurement means anything.
The shunt value is the whole design
Vshunt = I · R ·
P = I²R ·
Ioffset-equivalent = VOS / REverything pulls against everything else. A larger shunt produces more signal, so the amplifier’s offset represents less current and low-current accuracy improves proportionally. But dissipation rises with the square of current, and the burden voltage — the volts you have taken away from the load — rises too, which matters on a low-voltage rail or a battery near end of discharge.
A smaller shunt fixes power and burden and makes the offset problem worse in exact proportion. There is no setting that wins; there is only the trade, which is why the tool states it explicitly at your operating point rather than leaving you to discover it.
Self-heating closes a loop
A shunt dissipating a watt gets hot, and a hot shunt changes value by its tempco, which changes the reading. This is genuinely a feedback path, and it is why shunt tempco is specified in ppm/°C and why the good ones are metal foil or manganin rather than thick film. A 50 ppm part over a 60 °C swing contributes 0.3%; a 500 ppm thick-film part contributes 3% and has just become your dominant error.
Kelvin connection matters for the same reason. Solder joints and trace copper have resistance of their own, and copper’s tempco is about 3900 ppm/°C — nearly a hundred times worse than a decent shunt. If the sense connections pick up any of the current path, that copper is in series with your precision resistor and it dominates the drift. Four-terminal shunts exist entirely to prevent this, and a two-terminal shunt with carelessly placed sense traces throws away most of what you paid for.
What this does not cover
- Common-mode rejection at the amplifier’s input, which for high-side sensing on a high rail can be the dominant error and is not modelled here.
- Bandwidth and step response. Shunt inductance matters for fast current steps, and a current-sense amplifier’s settling time may exceed the interval you sample in.
- Amplifier offset drift over temperature, which behaves like offset but grows — a chopper amplifier is the usual answer when low-current accuracy matters.
- Thermal EMF at dissimilar-metal junctions, tens of microvolts per degree of gradient, which is the same order as a good amplifier’s offset and is why layout symmetry around the shunt matters for precision work.
More free tools
Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.