Series connection

What it is (precise)

Connect the positive terminal of one cell to the negative terminal of the next and two invariants follow from circuit law. Voltages add: n cells in series present n times the cell voltage at the string terminals, at every state of charge and under every load.

Ampere-hours do not: there is a single conduction path, so the current — and therefore the charge moved per hour — is identical for every cell in the chain, and the string's capacity in Ah remains that of one cell. Datasheets encode the arrangement in xPyS notation, parallel groups times cells in series: 1P104S means one parallel group of 104 series cells — pure series, no parallel paths inside the module.

The arithmetic climbs the hierarchy unchanged. A 314 Ah LFP cell sits at ~3.2 V nominal and stores about 1 kWh. Put 104 of them in series and the module is ~330 V nominal — the 1P104S building block usual in current 5 MWh-class containers, where the earlier 280 Ah generation used 1P48S–52S modules at 150–170 V.

Put four 104S modules in series and the rack string is ~416 cells at roughly 1,330 V nominal — and still a 314 Ah battery. Only energy accumulates: 416 × ~3.2 V × 314 Ah ≈ 418 kWh per rack, because energy is the product of the voltage that series stacking multiplied and the ampere-hours it left alone.

The reason to stack is current. A 314 Ah string discharging at 0.5C carries about 157 A whether it is one cell or 416 — series stacking raises power by raising voltage at constant current, so conductor cross-sections, contactor ratings and I²R losses are set by the current of a single cell, not by the megawatts of the rack. That is the same argument that pushed DC architectures from 1000 to 1500 VDC. The mirror arrangement is the parallel connection, which adds ampere-hours at constant voltage; real containers use both, but the series count is what sets the system's voltage class.

Why it matters in a real grid-scale project

The single conduction path has a governing consequence: the string inherits the limits of its weakest element. Every cell is forced to pass the same current, so the first cell to reach a protective voltage limit ends the charge or discharge for all of them — a series string delivers the minimum cell's capacity, not the average, and one lagging cell caps the power of its whole rack.

The mechanics of that rule, the stranded energy it costs and the balancing that manages it belong to the weakest-cell limitation; what matters here is that the exposure is created the moment elements are wired in series, and it grows with every element added to the chain.

The second consequence is insulation. Stacking multiplies not just the terminal voltage but the potential to ground along the chain: cell number one and cell number 416 sit far apart electrically, so every component the string touches — module housings, busbar carriers, sense harnesses, connectors — must be insulated for the rack's up-to-1500 VDC system voltage, not for the ~3.2 V of the cell it serves.

Creepage and clearance distances inside modules are sized to that system voltage, and the low-current voltage-sense wiring needs the same insulation class as the power path because it taps the same potentials.

The split between module size and module count is partly a handling decision for the same reason — a ~330 V module is the unit an installer lifts, while the lethal full-string voltage only exists once the modules are linked in the rack. Emerging 2,000 V architectures push the same trade further: more cells in series, less current per megawatt, and a higher insulation class for everything in the enclosure.

Typical values and standards

Anchor numbers for stationary LFP: ~3.2 V nominal per cell with an operating window of roughly 2.5–3.65 V; 104 cells in series give the ~330 V module of current 5 MWh-class designs, and ~416 cells in series is the practical ceiling on a 1500 VDC platform. That ceiling is set by the top of the cell window, not the nominal: the string's maximum voltage is the series count times the per-cell charge limit — which is why deployed 416S platforms hold the operating charge ceiling below the cell's 3.65 V datasheet maximum — so sizing against count × 3.2 V overstates the room available.

Series stacking also multiplies the swing — the ~1.15 V a single LFP cell traverses between empty and full becomes on the order of 480 V of string-voltage swing across 416 cells, which is why the PCS must accept a wide VDC window rather than a single DC voltage.

No standard prescribes a series count. The number falls out of two documents: the cell datasheet's voltage window and the DC system voltage the equipment is designed and certified to — 1500 VDC for today's mainstream platforms, with 2,000 V architectures emerging.

The series count then propagates into the insulation coordination of everything downstream, and into the BMS specification through a simple counting rule: N cells in series need N+1 voltage-sense taps, because both ends of every cell must be measured. A 104S module therefore carries 105 sense connections, and the BMS channel count of a container is largely a restatement of its series architecture.

How it shows up in specs, studies and contracts

On a datasheet the series architecture is the xPyS code and the three voltages printed beside it — nominal, minimum, maximum — each of which should equal the series count times the corresponding cell figure. The energy line is series count × parallel count × cell energy, and it holds only at nominal voltage and reference conditions; the watt-hours actually delivered follow the voltage trajectory, which series stacking amplifies along with everything else.

In sizing studies the check is envelope compatibility: the string's full voltage range — cold, hot, empty, full, aged — must sit inside the PCS's DC input window, and because every per-cell effect arrives at the terminals multiplied by the series count, a 0.1 V per-cell sag under load is roughly 42 V at the rack terminals.

In supply contracts the series count is a compatibility constraint. A replacement module must match the series architecture and cell revision of the string it joins, because a module with a different count or window shifts the whole string's limits; that is one reason vendors bin and document replacement modules rather than treating them as interchangeable.

Commissioning tests read the series arithmetic backward — a string voltage that does not equal count times a plausible cell voltage points at a miswired or dead element — and the witnessed capacity test then measures what the assembled series chain, weakest element included, actually delivers between its voltage endpoints.

Common pitfalls

The persistent confusion is what series stacking multiplies. It multiplies voltage and therefore energy; it does not multiply ampere-hours, and it does not multiply current capability — a 416-cell string sources the current of one cell, and its C-rate limits are the cell's C-rate limits.

The arithmetic error that follows is sizing the string against nominal voltage: the binding numbers are count × cell maximum against the 1500 VDC system limit on one side and count × cell minimum against the PCS window floor on the other, and both must survive the temperature and aging corners, not just the datasheet reference row.

The structural trap is forgetting that a series chain has no redundancy. Every weld, busbar joint and connector is in the only current path, so a single high-resistance joint heats at full string current — at the tens of microohms of a typical laser weld, roughly a watt per weld at 157 A, and hundreds of welds per module — while a single open joint takes the entire string offline.

Parallel paths offer graceful degradation; series paths offer none, which is why joint resistance measurements and torque checks appear in factory and commissioning records, and why the move to larger cells is partly a move to fewer series joints per megawatt-hour.