A battery management board gets specified around its controller — cell count, CAN, balancing. The part that decides whether the pack can be trusted is smaller and harder to change: the chain that turns pack current into a number. A shunt, a difference amplifier, an ADC reference and the copper between them — and state of charge is built on all four.
One of the two gets corrected every time the pack rests. The other does not, and that is the whole problem.
A battery management system is the circuit that keeps a pack inside its safe operating window: it measures cell voltage and pack current, decides whether the pack is inside its limits, and acts by opening contactors or FETs. Cell voltage is the easy half. It is sampled across a resistor divider or a dedicated monitor IC, it is bounded, and it is re-anchored every time the pack sits idle and its open-circuit voltage can be matched against a state-of-charge table.
Current is different. It is measured continuously and then integrated, because the only practical way to know how much charge is left in a working pack is to count what goes in and what comes out. State of charge is that count. So every error in the current measurement is an error that accumulates — and a pack in daily use may never rest long enough to be re-anchored.
The arithmetic is unforgiving. A 10 mA offset in the current reading is 0.01 A, which is 0.24 Ah a day, which is 7.2 Ah a month. On a 100 Ah pack that is 7% of the gauge, produced by an offset too small to notice on a bench meter. A 1 mA offset still costs 0.72 Ah a month. This is why a BMS is a drift problem before it is a resolution problem: a 16-bit converter on top of an amplifier whose offset walks with temperature buys nothing at all.
Any single term can be trimmed at test. The term that cannot is the one printed in copper.
A current-sense chain is not a component choice, it is an addition. The shunt, its temperature coefficient, the amplifier's input offset, the gain and reference behind the converter, and the way the sense traces are taken off the shunt all contribute to the number the controller sees. Quoting any one of them as "the accuracy" is how a board ends up at 5% in the field while the datasheet promised 0.5%.
| Term | Where it comes from | What it costs |
|---|---|---|
| Shunt tolerance | ±0.5% is the common grade; ±0.1% parts exist at a price. | A fixed gain error. It can be calibrated out at end-of-line test — if the test actually measures current. |
| Shunt tempco | Low-TCR alloys still move: 50 ppm/°C is ordinary. | 50 ppm over a 60 °C swing is 0.3%, and it returns with every thermal cycle. Self-heating adds to it: 100 A through 0.1 mΩ is 1 W in the shunt itself. |
| Amplifier offset | Input offset voltage, plus the part of it that drifts with temperature. | Dominant at low current, where the shunt signal is smallest — exactly where a resting pack spends most of its life. |
| Gain and reference | Resistor matching in the amplifier, drift in the ADC reference. | Another few tenths of a percent. Usually the easiest term to trim, and the one most often over-engineered. |
| Layout | Where the sense taps land, and what copper they share with the current path. | Can be larger than the four above combined — and it cannot be calibrated out, because it changes with current and with temperature. |
The terms do not simply add, but they do not cancel either. A worst-case sum is the honest number for a protection threshold; a root-sum-square figure is the honest number for a typical reading. Both have to be quoted over temperature. An accuracy specified at 25 °C is a statement about a bench, not about a pack that spends its life between −10 and 60 °C.
A shunt has two terminals that carry the current and two that measure it. The difference between them is the whole accuracy of the board.
A four-terminal shunt is a resistor with the sense points brought out inside the current terminals. The two heavy terminals carry the pack current; the two small ones tap the voltage across the resistive element only, excluding the joints and the copper leading up to them. Wire the sense pair to the heavy terminals instead — or to a plane that also carries current — and the measurement includes the resistance of a solder joint, which is not a number anyone specified.
The layout rules that follow are simple to state and easy to break:
Tap inside the current terminals — on the pads the manufacturer intended — and route the pair as a differential pair from there to the amplifier. Share no copper with the current path: a sense trace that tees off a heavy pour inherits a fraction of the current's voltage drop, and that fraction changes with load. Keep the pair together and away from switching nodes — the signal is tens of microvolts at low current, and a switching node a few millimetres away is volts. And filter at the pins: an RC pair at the amplifier input, with matched resistors, keeps the filter from becoming a gain error of its own.
There is a second, quieter error in the same place. Two different metals joined at different temperatures make a thermocouple. A shunt alloy against copper-terminated traces does exactly that, and a temperature gradient of a few degrees across the part produces microvolts — the same order as the signal at low current. Low-TCR alloys, a symmetric layout and keeping heat sources (the shunt itself included) away from one side of the pair are what keep that term small.
High-side or low-side is not a preference. It decides which amplifier you can use, which ground the sense pair references, and how much distance the barrier needs.
A low-side shunt sits in the pack negative, so the amplifier sees a common-mode voltage near ground and an ordinary difference amplifier will do. The cost is that the pack negative is no longer the same node as the system ground — everything downstream has to accept that the "ground" it is measuring against moves with load current.
A high-side shunt sits in the positive path and keeps one continuous ground, which most systems prefer. Now the amplifier has to survive the full pack voltage at its inputs, or the signal has to cross an isolation barrier — an isolated amplifier or a digital isolator with the converter on the pack side. That barrier is a physical object on the board: a gap in the copper, a distance across the laminate, and a creepage requirement that grows with the pack voltage and with how dirty the environment is allowed to get.
The distances on either side of that gap are the same ones a power board lives or dies by, and they are set in the artwork: creepage and clearance are dimensions of the drawing, not properties of the assembly. A slot milled under the isolator, the CTI of the laminate and the pollution degree the product is designed for all enter the calculation before anyone quotes. A split ground plane under the barrier is not a layout mistake — it is the barrier doing its job.
On a battery board, assembly is part of the measurement — in a way it is not on a signal board.
A shunt is a resistor, and the path through it runs through two solder joints. A void under a terminal does not just look bad under X-ray; it removes conducting area from the path, raises the local resistance and changes the value the board was calibrated against. That is why X-ray is the inspection that matters on this class of joint, and why a first article is worth looking at before a panel is built.
The current path itself is usually heavy copper — 2 oz and up, sometimes a busbar formed in the laminate. Heavy copper brings two assembly consequences that a signal board never sees: the joint has enough thermal mass to fight the iron or the reflow profile, and the sense pair has to be kept out of the pour that makes the joint easy. A shunt that is pressed or screwed into the board instead of soldered moves the problem to the mechanical assembly, and the torque spec becomes part of the measurement.
Two smaller things belong on the same list. Flux residue and moisture matter more here than on a digital board, because the sense nodes are high-impedance by design; cleaning and coating are process decisions, not cosmetics. And if the pair is coated, the coating has to be specified while the board is being designed, for the same reason a coating cannot repair creepage after the fact.
Our own fabrication runs 0.5 oz to 12 oz copper and assembly to IPC-A-610 Class III, with 3D X-ray on the inspection list — the parameters relevant to a shunt joint are in the fabrication table, and the inspection sequence is in how a board is tested.
It can, and sometimes it should. A Hall or fluxgate sensor measures the field the current produces, so it is isolated by construction and adds no resistance to the path. The trade is offset and drift: at low current the field is small, and the sensor's offset is what the measurement rests on. That is the same failure mode as a shunt chain, with less control over it. Where accuracy over temperature matters, a shunt with an isolated amplifier is still the usual choice; where the current is very large and isolation is mandatory, the contactless sensor earns its place.
Two different numbers hide inside that question. A protection threshold needs repeatability — the board must trip at the same current every time, and a fixed gain error can be calibrated out. State of charge needs long-term offset stability, because the error is integrated and never resets. A chain that is 1% accurate but drifts 10 mA with temperature will produce a worse gauge than one that is 2% accurate and stable. Ask for the error budget over temperature, not a single figure at 25 °C.
Low-side is cheaper and easier: the amplifier sees a common-mode voltage near ground, so an ordinary difference amplifier works. The cost is that the pack negative is no longer system ground, and everything downstream has to tolerate that. High-side keeps one continuous ground but puts the pack voltage at the amplifier inputs, so it needs a high-voltage part or an isolation barrier. That barrier is a layout decision with its own creepage distance, and it cannot be added later without changing the board.
The other half of a battery board: how the current path is sized, where the heat goes, and how the spacing distances are set.
X-ray is what sees the void under a shunt terminal; functional test is what proves the sense chain reads the current it should.
What the monitor IC, the isolation barrier and the fine-pitch parts around them demand from placement.
The shunt part number, the pack voltage and where the isolation barrier sits tell us more than a BOM does. If the sense pair or the spacing is still open, we will say so at the quote stage — while the artwork can still change, rather than after the boards come off the line.
