Seebeck Error: When a Sense Wire Adds Its Own Voltage

Every junction between two different metals in a sense-wire path is a small, unwanted thermocouple. If there’s a temperature difference across it, it generates a real voltage that a precision voltmeter cannot tell apart from the signal it’s supposed to be measuring.

1. A Junction You Didn’t Design Is Still a Junction

Anywhere a copper sense wire ends and something else begins, whether that’s a tin-plated ring lug, a nickel-plated connector pin, a brass terminal block, or a gold-flashed test probe, there’s a boundary between two different metals. That boundary doesn’t care that it was never meant to be a sensor. If the two sides of it sit at different temperatures, the junction generates a voltage of its own, stacked directly onto whatever the instrument is trying to read.

This is the same physics that makes a thermocouple work, just showing up somewhere it wasn’t invited. A thermocouple is deliberately two dissimilar wires joined at one end, with the pair generating a voltage proportional to the temperature difference between that junction and the reference end. A sense-wire harness or test fixture is built from several dissimilar-metal junctions that were never intended to measure temperature, but obey exactly the same rule: put a temperature gradient across a copper-to-something junction and it becomes an unintentional thermocouple, quietly adding microvolts to millivolts onto the measurement.

The relationship is close to linear over the small temperature spans found on a test bench: the junction voltage is the Seebeck coefficient of that metal pair multiplied by the temperature difference across it. A pure copper-to-copper joint has almost no Seebeck coefficient at all, which is exactly why it’s the reference point everything else gets compared against.

2. The Seebeck Coefficient Depends Entirely on What’s Touching What

The size of the effect varies by roughly four orders of magnitude depending on the two materials in contact, and on whether the joint is a clean metal-to-metal bond or one with an oxide film sitting in between it. Keithley’s Low Level Measurements Handbook publishes reference values for common junction pairs against copper, gathered from the kind of nanovolt-level instrumentation work where this error source has to be actively designed around:

Junction (vs. copper)	Seebeck coefficient
Cu–Cu (clean, “cold-welded” crimp)	≤0.2 µV/°C
Cu–Ag	~0.3 µV/°C
Cu–Au	~0.3 µV/°C
Cu–Pb/Sn (solder)	1–3 µV/°C
Cu–Kovar (Ni-Fe-Co alloy)	~40–75 µV/°C
Cu–Si	~400 µV/°C
Cu–CuO (oxidised contact)	~1000 µV/°C

Source: Keithley (Tektronix) Low Level Measurements Handbook, Table 3-1.

Two things stand out. First, staying on one metal throughout the sense path (copper wire into a copper lug, copper lug onto a copper or copper-plated terminal) gets you close to zero, because there’s no dissimilar junction to generate a voltage in the first place. Second, the worst case in the table isn’t a dissimilar metal pair at all: it’s an oxidised contact. A film of copper oxide between two nominally identical copper surfaces can behave far worse thermoelectrically than a clean junction between two genuinely different metals, which is one more reason a corroded or poorly-torqued connection is worth chasing down even when its DC resistance still looks acceptable.

3. Where This Turns Up in Battery Test Hardware

Sense-wire harnesses and test fixtures are built from parts chosen for cost, availability, and mechanical fit, not thermoelectric neutrality. A few places it commonly sneaks in:

Ring lugs and crimp terminals: tin-plated or nickel-plated copper lugs crimped onto sense wires and bolted to cell tabs, busbars, or terminal blocks. The lug’s plating, not the wire underneath it, is what actually touches the tab.
Connector pins: test fixtures and cycler channel connectors frequently use nickel, gold-flash, or beryllium-copper spring contacts rather than bare copper, chosen for wear resistance and spring properties rather than thermoelectric performance.
Cell terminals themselves: aluminium tabs bonded or welded to nickel-plated steel or nickel busbars inside the cell, with the sense probe landing on whichever surface is exposed.
Terminal blocks and screw terminals on cycler channels, often brass or plated steel rather than copper.

None of these choices are wrong on their own; plating exists for good reasons like corrosion resistance and contact wear life. The problem only appears when a real temperature gradient sits across one of these junctions: a cycler channel running warm next to a cooler idle neighbour, uneven airflow in a thermal chamber, direct sunlight through a chamber window, or simply a technician’s hand resting near a fixture while probing it.

4. Worked Example: How Small a Gradient Actually Matters

Take a sense-wire ring lug bolted to a cell tab, with a modest 3°C gradient across that junction, easily produced by a warm neighbouring channel, uneven chamber airflow, or a fixture sitting partly in direct light. Compare two plating choices for that lug, using the coefficients from the table above:

Tin-plated lug (Cu–Sn junction, 1–3 µV/°C): 3°C × 1–3 µV/°C = 3–9 µV of offset.
Nickel-plated lug (closer to the Cu–Kovar end of the table, since nickel alloys sit in a similar range, ~40–75 µV/°C): 3°C × 40–75 µV/°C = 120–225 µV of offset.

Whether that offset matters depends entirely on what’s being measured against it:

DCIR / HPPC pulse resistance. A 5A pulse into a cell with 20 mΩ of internal resistance produces a 100 mV signal (V = IR). The tin-lug offset is 0.003–0.009% of that signal, which is irrelevant. The nickel-lug offset is 0.12–0.23%, small in absolute terms, but worth noting if the test is tracking resistance growth of only a few percent per year across an aging study, where that offset starts to compete with the trend it’s trying to detect.

OCV–SOC curve characterisation. LFP cells are well documented as having a wide, flat plateau in their OCV–SOC curve, commonly cited as moving only tens of millivolts across 60–70 percentage points of SOC in the flattest section, which puts the local slope well under 1 mV per percent SOC in places. Against a slope that shallow, the nickel-lug offset of 120–225 µV alone can correspond to something in the region of half a percent or more of SOC, baked directly into the OCV–SOC lookup table before any BMS estimation algorithm runs on top of it. That’s not a rounding error; on a flat curve it’s a meaningful fraction of the very characteristic the test exists to capture.

The point isn’t that either plating choice is “wrong”; it’s that a connector or lug spec chosen purely for cost or corrosion resistance can quietly set the noise floor for a precision voltage measurement, long before anyone looks at the instrument’s own datasheet accuracy.

This post has been built based on the support of and sponsorship from: illumion, Custom Power, Quarto Technical Services, Eatron Technologies and About:Energy

5. Reducing It at the Hardware Level

Because thermoelectric offsets are approximately constant over a short measurement window, instruments correct for them mathematically (current reversal, offset-compensated ohms, delta mode, covered in more depth in the Four-Wire Kelvin Resistance Measurement piece). But that compensation only cancels a constant offset; it does nothing for an offset that’s drifting as a chamber ramps or a channel warms up mid-test, so it’s still worth minimising the problem at the hardware level rather than relying on the instrument to clean it up:

Keep sense-path junctions single-metal wherever possible: copper wire into copper (or copper-alloy) lugs, onto copper or copper-plated terminals, rather than introducing tin, nickel, or brass into the loop unnecessarily.
Where a dissimilar-metal junction can’t be avoided, keep it a clean, tight, well-torqued mechanical joint: a loose or lightly oxidised contact behaves worse than the base metal pairing would suggest, per the Cu–CuO row above.
Match junctions in pairs. If the current-carrying loop has to include a dissimilar-metal junction, put an equivalent junction elsewhere in the loop, oriented so the two thermoelectric voltages oppose rather than add.
Keep matched junction pairs physically close together and thermally coupled, so they see the same temperature rather than drifting apart under a gradient.
Route sense harnesses away from local heat sources (cycler power stages, adjacent warm channels, chamber heater elements, direct light through a chamber window) and away from directional airflow that puts one side of a connector at a different temperature to the other.
Let fixtures and instrumentation reach thermal equilibrium before starting a precision measurement; a fixture still settling from being handled or freshly connected is actively generating drifting, not constant, thermoelectric offsets.

6. Recommendations Checklist

Treat every dissimilar-metal junction in a sense-wire path as a potential thermocouple, not just an electrical connection.
Specify sense-path connectors and lugs with thermoelectric behaviour in mind, not just contact resistance and corrosion resistance, especially for microvolt-to-low-millivolt signals like OCV characterisation or micro-ohm resistance testing.
Where practical, keep the entire sense path a single metal.
For junctions that can’t be avoided, prioritise a clean, matched, thermally-coupled pair over trying to eliminate every junction outright.
Rely on instrument-side compensation (current reversal, OCO, delta mode) for the residual constant offset, not as a substitute for controlling gradients at the hardware level.