Quick answer
A shunt resistor turns current into a small voltage: V = I × R. The design is decided by three numbers, not one: the full-scale drop (which sets resolution), the power dissipated (I²R), and the error budget at the currents you actually measure. At 10 A, a 1 mΩ shunt gives 10 mV and 0.1 W; a 25 µV amplifier offset is already 0.25% of that signal, and at 1 A the same offset is 2.5%. Kelvin (four-wire) sensing and a low-drift amplifier matter as much as the resistor value.
Executive takeaways
- Pick the drop first. 10–50 mV full scale is a practical range; more drop means more power, less drop means more offset error.
- Power rating is not the operating point. A 0.1 W dissipation needs a part rated for at least 0.2–0.25 W, with the derating curve applied.
- Offset dominates at low current. A 25 µV offset is 0.25% at 10 mV but 2.5% at 1 mV — the same circuit is precise at full load and useless at 10% load.
- Kelvin sensing is mandatory above a few amps; sharing force and sense traces adds copper resistance and thermal EMF to the measurement.
1. How shunt sensing works
A shunt is a low-value, low-TCR resistor placed in the current path. The voltage across it is amplified by a current-sense amplifier and digitised by an ADC. Because the resistor is in series with the load, it dissipates power; because the voltage is small, every microvolt of offset and every degree of temperature gradient matters.
2. Choosing the shunt value
| Shunt at 10 A | Full-scale drop | Power | Offset error (25 µV) |
|---|---|---|---|
| 0.5 mΩ | 5 mV | 50 mW | 0.50% |
| 1 mΩ | 10 mV | 100 mW | 0.25% |
| 2 mΩ | 20 mV | 200 mW | 0.13% |
| 5 mΩ | 50 mV | 500 mW | 0.05% |
Resolution is the other half of the choice. With a gain of 100 and a 12-bit, 3.3 V ADC, a 10 mV full-scale signal becomes 1 V, and one LSB is 0.8 mV — about 0.08% of full scale. At 1 mV full scale the same ADC sees 0.1 V and only about 2.4 LSB per millivolt, so the digital resolution adds to the analog error. Size the shunt so the signal at the minimum current you care about is several LSB, not one.
The trade-off is visible in one table: doubling the resistance halves the offset error and doubles the power. For a 10 A rail, 1 mΩ is a common compromise; for a 1 A rail, 10–50 mΩ is more appropriate because the offset error at 1 mΩ would be 2.5%.
3. Power, temperature and TCR
Power is I²R, so it rises with the square of current. A 1 mΩ shunt at 10 A dissipates 0.1 W; at 20 A it dissipates 0.4 W in the same part. Use the derating curve: a 0.25 W part at 70% derating is good for about 0.175 W, which covers 10 A but not 15 A.
Temperature affects the measurement twice. First, the resistor's own TCR changes its value: 50 ppm/°C over a 50 °C rise is 0.25%, which is comparable to the amplifier offset. Second, the temperature gradient across the shunt and its solder joints creates a thermoelectric voltage — a few microvolts per °C for common metal combinations, which at 5 °C is already 15 µV.
Choose a low-TCR part (25–50 ppm/°C), give it copper area, and keep it away from heat sources. A 4-terminal (Kelvin) resistor removes the lead resistance but not the thermal EMF; layout and symmetry handle that.
4. Package, copper and thermal design
A shunt is a power resistor, and its package sets the thermal path. Small chip resistors (0805, 1206) are fine for tens of milliwatts; above a few hundred milliwatts, use a 2512, a 4-terminal part or a metal-strip shunt with a dedicated thermal pad. The copper around the pads is part of the heatsink, and the datasheet derating assumes a specific pad area.
Place the shunt in the current path where the copper is wide, not on a thin trace that adds resistance and heat. Keep the sense traces out of that copper, and keep the shunt away from the amplifier's own heat. A 0.1 W part with a 4-terminal construction is a different thermal design from a 2-terminal chip at the same value.
5. A second operating point: 2 A, 10 mΩ
For a 2 A rail, a 10 mΩ shunt gives 20 mV and 40 mW. The same 25 µV offset is 0.125% of full scale, and the TCR term over 50 °C is 0.25% — the shunt, not the amplifier, is now the dominant error. The lesson is that the shunt value should be chosen with the accuracy target at the minimum current, not with the power budget alone.
6. Kelvin sensing and layout
Kelvin sensing means the force current and the sense voltage use separate connections. The force path carries amps and has milliohms of copper and solder; the sense path carries microamps and must see only the resistor element. On a 4-terminal shunt this is built in; on a 2-terminal part it must be done on the PCB.
On the schematic, draw the shunt as a four-terminal symbol so the two sense wires are visually separate from the force path. That small convention prevents the most common layout error: a sense trace that quietly rejoins the power copper a few millimetres away.
- Route the sense traces from the inside edges of the pads, not from the ends of the current path.
- Keep the two sense traces symmetric and close together, so they pick up the same thermal gradient.
- Do not run sense traces under the shunt or through the power copper.
- Use a differential pair into the amplifier, with the filter close to the amplifier pins.
7. Error budget: where the accuracy goes
A precision current-sense amplifier such as the INA240 quotes ±25 µV offset, 0.20% gain error, 2.5 ppm/°C gain drift, 250 nV/°C offset drift and 132 dB DC CMRR. Those numbers are only useful when they are converted into errors at the shunt's full-scale voltage.
| Error source | Value at 10 A, 1 mΩ | Share |
|---|---|---|
| Amplifier offset (25 µV) | 0.25% | Largest single term |
| Gain error (0.20%) | 0.20% | Fixed |
| Offset drift (250 nV/°C × 50 °C) | 0.125% | Temperature |
| Gain drift (2.5 ppm/°C × 50 °C) | 0.013% | Small |
| Shunt TCR (50 ppm/°C × 50 °C) | 0.25% | Temperature |
| Thermal EMF (3 µV/°C × 5 °C) | 0.15% | Layout |
The table lists absolute values; how you combine them depends on whether the errors are independent. RSS is the usual assumption for random terms, while offset and gain error are often treated as worst-case for a conservative budget. Either way, the ranking is the same: offset and shunt TCR dominate at 10 mV full scale.
Root-sum-square of those terms is about 0.45% at 10 A. At 1 A the same chain is roughly 4.5%, because the offset terms do not scale with current. If the application needs 1% at 1 A, the shunt must be larger, the amplifier offset lower, or the measurement must be calibrated at the operating point.
8. High-side or low-side?
High-side sensing puts the shunt between the supply and the load. It keeps the load ground intact and detects a short to ground, but the amplifier must handle the full common-mode voltage — the INA240's −4 V to +80 V range is designed for this. Low-side sensing puts the shunt between the load and ground, so the amplifier sees a small common-mode voltage and can use a cheaper part; the cost is a lifted ground and the loss of ground-fault detection.
In motor drives, high-side or in-line sensing is preferred because the phase current must be measured regardless of the PWM state. In battery packs and low-voltage rails, low-side is often enough.
9. Amplifier selection
Four specifications matter more than the rest:
- Offset and offset drift — set the low-current accuracy.
- Gain error and drift — set the full-scale accuracy.
- Common-mode range and CMRR — set whether the part survives the rail and the PWM edge.
- Bandwidth and PWM rejection — set whether the current waveform is usable in a switching converter.
For PWM applications, choose an amplifier with enhanced PWM rejection rather than simply high bandwidth; the switching edge couples through the shunt and can saturate an ordinary amplifier. For battery or DC loads, a lower-bandwidth part with good DC accuracy is usually better.
10. Bidirectional and switching currents
Battery charge and discharge, motor regeneration and H-bridge outputs all reverse current. A unidirectional amplifier clips at zero; use a bidirectional part with a reference at mid-supply, or an amplifier whose output is centred on VREF/2. The shunt itself is symmetric, so the sensing chain, not the resistor, sets the direction capability.
In a switching converter the shunt sees the PWM ripple on top of the average current. The amplifier's bandwidth must be high enough to reproduce the ripple if you need it, and low enough not to amplify the switching edge if you do not. Many designs add an RC filter at the amplifier input; the filter's resistor must be small enough not to load the shunt and matched on both legs to preserve CMRR.
11. Filtering, bandwidth and noise
The signal at full scale is 10–50 mV, so noise matters. A differential RC filter with a cutoff around 1/10 of the switching frequency removes the edge without distorting the average. Keep the two filter resistors equal and place the capacitor close to the amplifier pins; mismatch converts common-mode noise into differential error.
Check the amplifier's input bias current through the filter resistors: 1 µA through 1 kΩ is 1 mV, which is 10% of a 10 mV signal. For high-impedance filters, choose an amplifier with low bias current, or reduce the resistor values.
12. Calibration and verification
Even a 0.45% design benefits from a two-point calibration: zero current and a known full-scale current, both at the operating temperature. Store the gain and offset correction in firmware. A single-point calibration at room temperature leaves the TCR and drift terms uncorrected.
A calibrated shunt and a calibrated meter are two different things; verify both, or the reference becomes the largest error in the chain.
Verify at 10%, 50% and 100% load against a calibrated meter, and repeat at the temperature extremes. The 10% point is where offset dominates, and it is the point most often skipped.
13. Who should not use a shunt
- Very high currents (hundreds of amps). The power loss becomes unacceptable; use a Hall sensor or a current transformer.
- Isolated or high-voltage rails. A shunt needs an isolated amplifier or digital isolator; a Hall sensor may be simpler.
- Situations with no room for a drop. A 50 mV drop is 1% of a 5 V rail; on a 1.2 V core rail it is 4%.
- Measurements that must survive a short circuit. The shunt and amplifier need protection; a shunt alone is not a fault-tolerant sensor.
14. Five mistakes
1. Sizing the shunt for full load only. Accuracy at 10% load is usually worse than at 100%, often by a factor of ten.
2. Ignoring TCR. A 100 ppm/°C part over 60 °C drifts 0.6% — more than the amplifier's gain error.
3. Sharing force and sense traces. The copper and solder add resistance that changes with temperature.
4. Placing the shunt next to a hot component. The thermal gradient creates EMF and changes TCR.
5. Forgetting the common-mode range. A 5 V amplifier on a 48 V rail will not survive the first switching event.
15. Checklist
- Full-scale current and the minimum current that must be accurate.
- Shunt value, drop and power at both points.
- Power rating with derating at the maximum ambient.
- TCR and thermal EMF of the chosen part.
- Amplifier offset, gain error and drift at the operating temperature.
- Common-mode range and CMRR for the rail and the switching node.
- Bandwidth and PWM rejection if the load switches.
- Kelvin layout and filter placement.
- Calibration plan: one point or two, and at what temperature.
- Verification: compare against a calibrated meter at 10%, 50% and 100% load.
FAQ
Can I calibrate out the shunt TCR?
Only at the calibration temperature. A two-point calibration removes offset and gain, but the drift between calibration points remains; a temperature sensor and a correction table are the robust answer for wide-temperature products.
Should the shunt be before or after the filter?
The shunt stays in the current path; the RC filter goes between the shunt's sense terminals and the amplifier inputs. Do not put the filter in the force path.
How do I measure 1 A accurately with a 1 mΩ shunt?
A 1 mV signal is dominated by offset. Use a larger shunt (10–50 mΩ), a zero-drift amplifier, or calibrate at 1 A; otherwise the error will be several percent.
What shunt value should I start with?
Choose a full-scale drop of 10–50 mV. For 10 A that means 1–5 mΩ; for 1 A, 10–50 mΩ. Then check the power and the offset error at your minimum current.
Why is my current reading wrong at low currents?
Amplifier offset. A 25 µV offset is 0.25% at 10 mV but 2.5% at 1 mV; the error is fixed in volts, so it grows as a percentage as current falls.
Do I need a four-terminal shunt?
Above a few amps, yes, or a careful Kelvin layout on a two-terminal part. The sense traces must not carry the load current.
Can I use a shunt on a 48 V rail?
Yes, with an amplifier rated for that common-mode voltage. The INA240 family, for example, covers −4 V to 80 V; a 5 V-only amplifier will not.
How do I reduce thermal EMF?
Keep the two sense junctions at the same temperature, use symmetric copper, and avoid placing the shunt near heat sources. A chopper-stabilised amplifier also removes low-frequency offset drift.
Glossary
- Shunt
- A low-value resistor used to convert current into a measurable voltage.
- Kelvin sensing
- Four-wire measurement that separates force and sense connections.
- TCR
- Temperature coefficient of resistance, in ppm/°C.
- Thermal EMF
- Voltage generated by a temperature gradient across dissimilar metals.
- CMRR
- Common-mode rejection ratio; how well the amplifier ignores the rail voltage.
Sources
- Texas Instruments INA240 datasheet — ±25 µV offset, 0.20% gain error, 132 dB CMRR, −4 V to 80 V common mode.
- Texas Instruments, Low-Drift In-Line Motor Current Measurements with PWM Rejection (SBOA160) — in-line sensing and PWM rejection.
- Shunt resistor datasheet for TCR, power derating and thermal EMF of the specific part.
Selecting shunts and current-sense amplifiers?
Browse our resistor line and amplifier line, or send the current range, rail voltage and accuracy target for a part recommendation.