Investigating Thermoelectric Zero Offset Around a Current Shunt

A current measurement can drift even when the amplifier is stable and no load current flows. Dissimilar metal junctions at different temperatures generate thermoelectric voltages. At a low-value shunt, a few microvolts can represent a significant apparent current, especially after one terminal has been heated by a neighboring power component.

Separate zero voltage from zero current

The Keithley Low Level Measurements Handbook explains thermoelectric EMFs in low-voltage measurement paths. Kelvin connections remove much of the voltage error caused by load current in shared copper, but they do not remove every voltage source along the sense circuit. Connector contacts, soldered junctions, and instrument leads also belong to the measured loop.

Suppose a 1-milliohm shunt develops a 3-microvolt zero-current residual. The equivalent indicated current is 3 mA. This arithmetic does not identify the physical source; it establishes whether the observed voltage is large enough to matter. Compare the residual with the amplifier’s input offset and its specified drift rather than assuming the entire error belongs to the shunt.

Create a controlled thermal experiment

First establish a true zero-current state while keeping the measurement electronics powered. Observe the reading after a heavy-load interval and during cooling. Next, change which nearby circuit dissipates power without changing the measured shunt current. A slow residual that follows the spatial heating pattern suggests a thermal mechanism, whereas a fast synchronous spike suggests electrical coupling.

• Keep sense leads and connectors thermally sheltered from moving air.
• Record both shunt terminal temperatures where the instrumentation permits.
• Avoid touching one terminal immediately before a microvolt reading.
• Repeat with current direction reversed when the circuit supports it.
• Distinguish amplifier warmup from shunt and connector thermal gradients.

Reduce the gradient, then verify

More symmetric copper, distance from concentrated heat sources, and matched sense-path construction can improve repeatability. A zero calibration taken before a load pulse may not remain valid immediately afterward. A correction model based on one ambient sensor may miss a local differential temperature, so validate it with realistic operating sequences.

Use Kelvin pickup placement to remove shared-current errors first, and common-mode transient testing to identify fast artifacts separately. Preserve the full time trace after load removal; a single settled zero reading cannot describe a board that must measure small currents immediately after delivering a large one.