Current Shunt Measurement and Kelvin Layout

A current shunt is the simplest way to measure current: pass it through a resistor of known value and measure the voltage across it. The accuracy of the measurement, however, depends almost entirely on where the voltage is measured. A few milliohms of track resistance or a small thermal gradient can introduce more error than the tolerance of the resistor itself, which is why the layout around a shunt deserves the same care as the analogue front end.

How a Shunt Measurement Works

A shunt converts current into a voltage by Ohm law, and the amplifier then measures that voltage. A 1 milliohm shunt carrying 20 amperes develops 20 millivolts, which is a small signal compared with the voltages elsewhere on the board. Any additional resistance in series with the shunt adds to the measurement, and any offset in the amplifier is amplified along with the signal, so both the resistor and the measurement circuit have to be treated as a system.

The choice of shunt value is a trade-off between signal amplitude and power dissipation. A larger resistance gives a bigger signal and better noise immunity but dissipates more power and drops more voltage, which matters in a low voltage system. Where a 1 milliohm shunt would drop 100 millivolts at 100 amperes, the same measurement on a 3.3 volt rail consumes three percent of the available headroom.

Two Wire versus Four Wire

A two wire connection measures the voltage at the amplifier rather than at the shunt, so it includes the resistance of the traces and the solder joints between them. On a board where the shunt is a few millimetres from the amplifier, that extra resistance can be a significant fraction of the shunt value, and it changes with temperature and with the solder joint quality.

A four wire, or Kelvin, connection solves the problem by separating the current path from the measurement path. Two terminals carry the load current and two separate terminals carry the sense signal, so no current flows in the sense circuit and the voltage measured is that across the resistive element alone. The separation has to be maintained all the way to the amplifier, which is a layout requirement rather than a component choice.

Kelvin Connection Geometry

The sense traces must connect to the inside of the shunt terminals, inside the current path, so that they sample the element rather than the joint. On a shunt with dedicated sense pins, the connection is defined by the component, and the layout simply has to reach those pins without adding series resistance. On a two terminal shunt, the sense traces must tap the pad at a point where no load current flows, which in practice means inside the current entry area.

The tap points should be symmetric on both sides of the shunt. An asymmetric tap adds a small amount of the current path resistance to one side only, which appears as a gain error rather than a common mode shift and therefore cannot be removed by calibration. Symmetry also keeps the two sense traces the same length, which matters when they run to a differential amplifier with finite common mode rejection.

Sense Trace Routing

Sense traces should be routed as a tight differential pair from the shunt to the amplifier, over a continuous reference plane, and away from switching nodes and inductors. The signal is small, so a trace that runs beside a switching node will pick up capacitive coupling that appears as noise in the measurement. Where the amplifier is far from the shunt, a filter at the amplifier input reduces the bandwidth and rejects the high frequency content that the routing picks up.

The reference plane matters as much as the traces. The two sense traces should refer to a quiet ground, and that ground should be the same reference that the amplifier uses, taken at one point rather than through a shared impedance. Where the shunt is on the high side of the load, the common mode voltage at the amplifier input is close to the supply rail, which requires an amplifier with a suitable input range and a careful layout to avoid leakage into the sense path.

Thermal EMF and Temperature Effects

A temperature difference across a shunt generates a voltage of its own, because two junctions of dissimilar metals at different temperatures behave like a thermocouple. The effect is small, a few microvolts per degree for a copper to alloy junction, but it is comparable to the measurement signal when the shunt is small and the current is low. In practice this means that thermal EMF limits the resolution of the measurement more often than noise does.

Layout measures reduce it. Keeping the two sense connections at the same temperature, avoiding a heat source on one side of the shunt, and using symmetric copper on both sense taps all reduce the gradient. Where the shunt is used for a high precision measurement, a low thermal EMF alloy should be chosen and the calibration should be performed after the assembly has reached thermal equilibrium.

Amplifier Interface and Filtering

The amplifier defines the accuracy of the system as much as the shunt does. Input offset voltage, offset drift, gain error, and common mode rejection all contribute, and their effect should be calculated against the smallest signal that must be resolved. A current sense amplifier with a low offset drift is worth more than a shunt with a tight tolerance, because the offset error appears as a constant current reading that cannot be calibrated out.

Filtering belongs at the amplifier input, close to the device, with the resistor and capacitor values chosen so that the filter does not load the sense traces. A differential capacitor across the inputs and a common mode capacitor from each input to ground form the usual network, and the components should be placed so that the filtered signal is the one the amplifier sees rather than one that has already picked up noise.

Calibration and Verification

Every shunt based measurement should be calibrated against a reference, and the calibration should include the offset at zero current. A two point calibration, at zero and at a known load, removes both the offset and the gain error, provided the relationship is linear. Where the current range is wide, a three point calibration confirms that the linearity is as expected.

Verification under load adds the thermal dimension. Passing a known current through the shunt for long enough for the assembly to reach a steady temperature shows whether the reading drifts, and comparing two identical channels on the same board shows whether the layout is repeatable. The results belong with the ‘+L(‘pcb-pad-design-standards’,’pad design’)+’ and the ‘+L(‘pcb-design-and-fabrication’,’assembly documentation’)+’, because a change in the pad geometry can change the measurement as much as a change in the resistor.

Related reading: our fabrication notes, board quality and design release notes cover the same ground.

FAQ

Why does my current reading include an offset? Usually because the sense traces are tapped outside the shunt element or because the amplifier offset has not been calibrated at zero current. A Kelvin connection and a two point calibration resolve both.

How small a signal can a shunt measurement resolve? It depends on the thermal EMF and the amplifier offset drift rather than on noise alone. Below a few hundred microvolts, thermal effects set the practical limit.

Should the shunt be on the high side or the low side? A low side shunt is simpler because the common mode voltage is near ground. A high side shunt needs an amplifier rated for the supply voltage but does not disturb the load return path.