The Pack That Doesn't Drift Apart
The Pack That Doesn't Drift Apart
The pack is rated 168 watt-hours and delivers 160, because the fullest cell ends every charge and the emptiest ends every discharge, and the 8 watt-hours in between are locked away behind the weakest cell.
The pack is rated 168 watt-hours and delivers 160. The cells are all the same part number, from the same batch, bought on the same day. Nothing is broken. Nothing is out of spec. And 8 watt-hours, 5% of the pack, are locked away behind one cell, because the fullest cell ends every charge and the emptiest ends every discharge, and the capacity in between goes unused.
The pack did not lose those 8 watt-hours. They are still in the cells. They are stranded, sitting in the twelve cells that still have charge in them when the thirteenth hits its lower limit and the BMS stops the discharge for all of them.
And the drift that stranded them grows: the cells leave the factory with tiny differences, and temperature turns small differences into large ones, and a year later the gap is wider and the locked-away share is larger, and the pack that delivered 160 watt-hours now delivers 158, and nobody changed anything, because the pack changed itself.
The simple idea first
Think of a pack as a chain of buckets, tied together so the water has to flow through all of them at once. The buckets are the cells, and they are all supposed to hold the same amount of water, the same charge, in amp-hours. But no two buckets are exactly the same size, and the chain can only fill until the fullest bucket overflows, and can only empty until the emptiest bucket runs dry.
The water left in the other buckets when the emptiest runs dry is the capacity the pack cannot use. And the chain is only as strong as its weakest bucket.
That is the whole pack, in one image. Series means the buckets are in a line: the same current flows through every cell, so the voltage adds up and the capacity is the capacity of the smallest cell. Parallel means the buckets are side by side: the voltage is the same for all of them, and the capacity adds up, and the buckets share their water on their own, because current flows between them until their voltages are equal.
The architecture is the choice of how the buckets are tied together, and the balancing is the work of keeping them level.
The weakest cell limits the string
In a series string, the same current flows through every cell. When charging, the battery management system has to stop as soon as any one cell reaches its upper voltage limit; when discharging, it has to stop as soon as any one cell reaches its lower limit. (Ingro Energy) A 2024 paper in Scientific Reports puts it simply: if the cells are not properly balanced, the weakest cell will always be the one limiting the usable capacity of the pack. (Scientific Reports, via Ingro Energy)
"A pack is only as strong as its weakest cell: if one cell reaches its upper voltage limit before the others during charging, the charger must stop early to protect it, leaving the remaining cells undercharged. The same logic applies during discharge." (EL-Cell) The fundamental constraint, written as an equation, is Qpack = min(Q1, Q2, …, Qn), the pack's capacity is the capacity of its smallest cell. (Electrical Equipment Factory)
And the loss is not small. A pack with a 5% cell voltage spread at end of charge cannot access 100% of its rated capacity. The energy sitting above the weakest cell's cutoff in every other cell is stranded.
(Maxwell Energy) A 100 mV imbalance in a 4S pack can reduce effective capacity by 5–15%. A 300 mV imbalance, and the pack appears nearly full on the low end and nearly empty on the high end. (doflagie) Qnovo's own example makes the same point with two cells: both nominally 7,000 Coulombs (about 1,900 mAh), but in reality one cell is 7,200 Coulombs and the other is 6,800, a 5% difference in capacity, "and can readily happen in a pack without the proper precaution."
(Qnovo) The smaller cell reaches its maximum voltage first, and at that moment one cell is 100% full while the other one is not. (Qnovo) If the charging is not disconnected immediately, one cell will certainly get overcharged and cause a hazard. (Qnovo)
Why cells drift apart
Cells leave the factory with tiny differences in capacity, resistance and self-discharge. (Ingro Energy) A 2025 analysis of battery packs in the journal Energies found that, in series packs, charge inconsistency driven by differences in coulombic efficiency and initial state of charge is the main reason capacity is lost, while differences in resistance mostly limit power. (Energies, via Ingro Energy) Coulombic efficiency is the share of the charge put into a cell that comes back out again, and if one cell keeps 99.5% of every charge and its neighbour keeps 99.8%, the gap between them grows a little on every cycle, and after a hundred cycles the gap is a real number.
Temperature turns small differences into large ones. Sandia's energy storage handbook notes that minimising the temperature spread between cells minimises the variation in how they age. Field data bears this out: in a South Korean grid battery studied over three years, variation between cells within each bank grew every year, concentrated in racks with uneven cooling airflow.
(Sandia, via Ingro Energy) The cells in the middle of the pack run warmer than the cells on the outside, because heat cannot get out of the middle, and the warmer cells age faster, and the faster they age the more they drift, and the more they drift the more they are worked, because the balancing has to work harder on them. The drift is a loop, and the loop runs in one direction.
And the drift has a number on it. A self-discharge difference as slight as 0.5% per month will accumulate into a 6% capacity deficit within a single year if left uncorrected. (Electrical Equipment Factory) Half a percent a month, a difference so small it would not show up on a datasheet, is 6% in a year, and 6% of a 168 Wh pack is 10 Wh, and 10 Wh is the difference between the pack delivering its rated capacity and the pack delivering a tenth less.
Series or parallel: the first decision
The first decision is not the balancing. It is the arrangement, and the arrangement decides what can go wrong.
Series (S): The cells are end to end, and the voltage adds up. Thirteen NMC cells in series is 48.1 V nominal (13 × 3.7 V), 54.6 V full (13 × 4.2 V), 39.0 V empty (13 × 3.0 V), the bench pack from Article #6. The capacity is the capacity of the smallest cell, and the current is the same through every cell, so the fullest cell ends the charge and the emptiest ends the discharge. The series string is where balancing lives, because the series string is where the drift locks capacity away.
Parallel (P): The cells are side by side, and the capacity adds up. The voltage is always equal for both cells, and any difference in charge capacity between the cells shows up as a difference in current in the two branches. (Qnovo) In particular, the parallel configuration always guarantees that the cells do not exceed their maximum safe voltage, often 4.35 V. (Qnovo) Cells in parallel self-balance through direct electrical connection, current flows between cells until their voltages equalise. (Cell Saviors) The parallel group does not need a balancer, because the cells balance each other.
But parallel has its own drift, and it is a current drift, not a charge drift. Simulations of an 8-cell parallel-connected module demonstrate that reducing current imbalance leads to more uniform cell aging and improved module and pack-level degradation predictions. (Ayalasomayajula et al., Journal of the Electrochemical Society, 2025, Sandia and Oak Ridge National Laboratory) Busbar interconnect resistance is one of the key factors affecting current imbalance in lithium-ion battery modules and packs, especially those with parallel connections, even small busbar resistances between parallel branches cause current imbalances, leading to inhomogeneity in cell SOCs, temperatures, and impedances within the pack.
(Ayalasomayajula et al., 2025) And terminal location matters: opposite-end terminal connections at intermediate branches minimise the imbalance. (Ayalasomayajula et al., 2025) The parallel group balances itself, but only if the busbar and the terminals let the current share evenly, and the busbar is the part of the pack nobody budgets for.
The bench pack is 13S1P, thirteen cells in series, one in parallel, so it has no parallel group, and the whole architecture question is the series string and its balancing. The gateway pack from Article #14 is 4S2P, four groups in series, two cells in parallel in each group, so it has both: the series string needs balancing, and the parallel pairs need a busbar and terminals that share the current evenly.
Passive or active: the second decision
Balancing is the battery management system's job of bringing cells back into line. Sandia's handbook describes the two approaches. (Ingro Energy) The choice between them is the second decision, and it is a decision about time, heat, and money, in that order.
Passive balancing: The most common approach. Switchable resistors bleed off excess charge from the highest cells, turning it into heat, until the others catch up. (Ingro Energy) Passive balancing dissipates excess energy as heat, while active balancing redistributes energy between cells. (EL-Cell) The BMS activates the bypass resistor on any cell that exceeds a set threshold (usually within 10 to 20 mV of the pack maximum). The current through the resistor is the balance current, typically 50 to 300 mA for BMS units. (Cell Saviors) Passive cell balancing dissipates excess energy as heat through shunt resistors, typically restricted to bypass currents below 300 mA due to enclosure thermal dissipation limits. (Electrical Equipment Factory) The energy dissipated is lost permanently; it does not benefit any other cell. (Maxwell Energy)
Active balancing: Capacitors or small converters move charge from stronger cells to weaker ones, wasting less energy, at higher cost and complexity. (Ingro Energy) Energy transfer circuits move charge from higher-SoC cells to lower-SoC cells, preserving usable energy. (EL-Cell) Active balancing circuits are more complex and more expensive, but they preserve energy by moving it from overcharged cells to undercharged ones. (EL-Cell) Active battery balancer architectures transfer charge between cells using capacitive or inductive topologies, achieving currents up to 5 A at efficiencies exceeding 85%. (Electrical Equipment Factory) The table, from the same guide: passive dissipative, 20 mA to 300 mA, 0% efficiency; active switched capacitor, 500 mA to 2.0 A, 75% to 88%; active inductive or flyback, 1.0 A to 5.0 A, 85% to 94%. (Electrical Equipment Factory) And active balancing moves energy at 80 to 95% efficiency. (Maxwell Energy)
The deciding number is the time. The useful starting point is the amount of charge that must be removed, measured in ampere-hours: balancing time (h) = charge difference (Ah) ÷ average balancing current (A). (EVKX) For two otherwise identical 100 Ah groups, a difference of one percentage point in SOC represents 1 Ah.
Removing that difference at a continuous 0.1 A takes 10 hours. If that branch operates for only half the elapsed time at the same current, its average current is 0.05 A and the correction takes 20 hours. (EVKX) The balance current is the whole decision: a passive 50 mA balancer being asked to close a gap it would need 200 hours to fix is the most common disappointment in the field.
(LJY Energy)
And the window matters as much as the current. Passive balancing only works at the top of charge. The BMS can only bleed off charge from high cells while the pack is near full, it can't push charge into low cells.
This means balancing happens during the final stage of charging, when all cells are approaching full. (Cell Saviors) Passive balancing also has no transfer path to supply a low group from the remaining groups during discharge. (EVKX) During charging, the current entering a bled cell is not the pack current, it is the pack current minus the bleed current: net charging current into the cells = pack charging current − bleed current.
(EVKX) Active balancing, by contrast, can act during charge, discharge, or idle. (Dan-Tech Energy) The passive balancer can only hold the full cell back while the others catch up, and it can only do it at the top of the charge. The active balancer can move charge whenever the pack is awake.
Which one you need is a research finding, not an opinion. The 2025 Energies analysis found that, for packs built from capacity-screened cells, passive and active balancing are about equally effective, and that active balancing outperforms passive mainly when cell capacities differ significantly, which becomes more likely as a battery ages or when new modules are mixed with old. Passive balancing corrects charge-level differences; it cannot make a smaller cell hold more.
(Energies, via Ingro Energy) Active balancing pays off mainly when cell capacities differ significantly, as packs age or modules are mixed. (Ingro Energy) The bench pack, built from capacity-screened cells and charged fully every cycle, is the case where passive is enough. The pack that ages, or the pack that mixes new modules with old ones, is the case where active earns its cost.
Top or bottom: the third decision
Top balancing and bottom balancing are not two kinds of balancing. They are two choices of where to align the cells, and the choice decides which end of the cycle the pack delivers fully.
Top balancing: Cells are equalised at the top of their charge window, ensuring a uniform full charge. (EL-Cell) Top balancing aligns groups near the upper end of the charge window; bottom balancing aligns them near the lower end. These terms describe the alignment point. Passive and active describe the mechanism used to change charge distribution. (EVKX) One documented implementation is the Orion Jr. 2, a BMS for packs up to 48 V nominal in light vehicles and stationary applications, whose 2018 manual describes adjustable balancing thresholds and recommends top balancing. (EVKX) Top balancing is the most common approach: the charger holds near top-of-charge while the balance dissipates. (doflagie)
Bottom balancing: Cells are equalised at the bottom of their discharge window, ensuring a uniform full discharge. (EL-Cell) Bottom balancing equalises during discharge, so the cells reach 0% together. (doflagie)
The difference is what the pack is for. Top balancing is the choice when the pack is charged fully every cycle and the risk is at the top, the cell that hits its upper voltage limit first, and the overcharge that follows if the BMS does not stop in time. (Qnovo) Bottom balancing is the choice when the pack is discharged deeply every cycle and the risk is at the bottom, the cell that hits its lower limit first, and the discharge that stops with charge still stranded in the other cells.
And there is a hard limit on what either one can do: aligning them at the top of the charge window cannot restore the smaller group's missing capacity. (EVKX) Top balancing makes the cells full together. It does not make the small cell bigger.
If the capacities differ, the pack's capacity is still the smallest cell's, Qpack = min(Q1, Q2, …, Qn), and the only fix for that is a different cell, not a different balance point.
The LFP complication: the flat curve hides the drift
LFP's very flat voltage curve, ideal for delivering steady power, makes balancing harder. Sandia notes that flat-voltage chemistries are bad for cell balancing during prolonged operation at partial state of charge, because cells look alike by voltage even when their charge levels differ. Sloping-voltage chemistries such as NMC can be balanced at almost any state of charge.
(Sandia, via Ingro Energy) LiFePO4 has one of the flattest voltage curves of any lithium chemistry. Between roughly 20% and 80% state of charge, cell voltage barely moves as charge is added or removed. Voltage-based balancing depends on detecting voltage differences between cells, and in the middle of the curve there are no meaningful differences to detect.
(LJY Energy) Meaningful differences only appear in the "knee" above roughly 3.4 V per cell, exactly where balancing is designed to work. (LJY Energy) For lithium iron phosphate chemistry, charge balancing must be executed exclusively in the upper voltage inflection zone above 3.40 V per cell. (Electrical Equipment Factory)
In practice, LFP batteries balance best near the top of charge, where voltages separate. A battery that never reaches full charge, because it is always cycled in a narrow middle window, may drift without the BMS noticing. That is a good reason for an EMS to schedule occasional full charges on LFP systems, for balancing and for recalibrating state of charge.
(Ingro Energy) And the trap runs the other way too: enabling balancing below about 3.35 V per cell can actually leave a pack less balanced, because the balancer acts on voltage readings that do not represent real state-of-charge differences. (LJY Energy, buyer-community consensus, not a manufacturer standard, labelled as such) The bench pack is NMC, with a sloping curve, so the BMS can see the drift at almost any state of charge.
The LFP pack in the same enclosure cannot, and the pack that never reaches full charge is the pack that drifts in silence.
Drift as an early warning: the spread is a signal
Cell spread is more than a capacity problem. A cell that keeps drifting away from its neighbours, needs more balancing than the rest, self-discharges faster or runs warmer is often the first visible sign of a fault developing. Fault-diagnosis research routinely flags cells by comparing them with the pack's typical cell, with one caveat noted in a 2020 review in the journal Algorithms: active balancing can mask the very differences such methods look for, so balancing activity itself is worth tracking. (Ingro Energy)
This is where analytics adds to the BMS. The BMS balances and protects in real time; analytics watches the trend over weeks and months, notices which modules need ever more balancing, and turns that into a maintenance decision before the weak cell starts limiting the rest. (Ingro Energy) The spread is the cheapest diagnostic the pack has: it costs nothing to watch, it is already measured by the BMS, and it moves weeks before the capacity does.
The cell that needs ever more balancing is the cell to replace, and the pack that tracks its balancing time per cycle is the pack that knows which cell that is, before the cell knows it itself.
The bench pack, and the 8 watt-hours
The bench pack from Article #6: 13S1P NMC, 3.5 Ah per cell, 48.1 V nominal (13 × 3.7 V), 54.6 V full (13 × 4.2 V), 39.0 V empty (13 × 3.0 V), 168 Wh. The demo is the article's own worked example, labelled, built on the cluster's figures and the verified sources above.
Step 1, the drift. The cells leave the factory with small differences. Qnovo's example: two cells nominally 1,900 mAh, in reality 7,200 and 6,800 Coulombs, a 5% difference in capacity, "and can readily happen in a pack without the proper precaution." Apply the 5% to the bench pack: twelve cells at 3.5 Ah, and one cell at 3.5 × 0.95 = 3.325 Ah, the article's own worked example, labelled. Qpack = min(Q1, …, Q13) = 3.325 Ah. The usable energy is 48.1 V × 3.325 Ah = 159.9 Wh, and the locked-away share is 168.3 − 159.9 = 8.4 Wh, 5.0% of the pack. The fullest cell ends the charge at 4.2 V while the weakest is at 95%, and the emptiest ends the discharge at 3.0 V while the strongest still holds 5%. The 8.4 Wh are still in the cells. They are stranded.
Step 2, the drift grows. A self-discharge difference of 0.5% per month accumulates into a 6% capacity deficit within a single year if left uncorrected. (Electrical Equipment Factory) After a year, the weakest cell is at 3.5 × 0.94 = 3.29 Ah, and the usable energy is 48.1 × 3.29 = 158.2 Wh, and the locked-away share is 10.1 Wh, 6.0%. And the temperature makes it worse: the bench pack sits in the sealed box from Article #7, where 5 W of self-heating is a ~25 °C rise, and the hub's inside reaches 85 °C at 60 °C ambient (Articles #50 and #52), and Sandia's handbook notes that minimising the temperature spread between cells minimises the variation in how they age, and the South Korean grid battery's variation grew every year, concentrated in racks with uneven cooling airflow. (Sandia, via Ingro Energy) The drift is a loop, and the loop runs in one direction.
Step 3, the balancing time. The gap to close is the 5% spread: 3.5 Ah × 5% = 0.175 Ah, 175 mAh. And the window: passive balancing only works at the top of charge, so the pack has to sit near full for the whole of that time. The bench pack charges at 0.5C, 1.75 A, so the charge itself takes 3.5 ÷ 1.75 = 2.0 hours, and the balancing needs 1.75 hours at the top, on top of the charge. The net charging current into a bled cell is the pack current minus the bleed current (EVKX): 1.75 − 0.10 = 1.65 A. The EVKX cross-check, on a 100 Ah pack: 1% of SOC is 1 Ah, and removing it at a continuous 0.1 A takes 10 hours, and at half duty, 20 hours. (EVKX)
Step 4, the heat. The passive balancer burns the excess as heat. The energy bled per cycle is 0.175 Ah × 4.1 V ≈ 0.72 Wh, and over a year of daily cycles, 0.72 × 365 = 262 Wh, burned. The heat while it bleeds is small but real: three cells bleeding at 100 mA and 4.2 V is 3 × 4.2 × 0.1 = 1.26 W, and all thirteen at once would be 5.46 W, in the sealed box from Article #7, where 5 W is a 25 °C rise, and the bleed heat is heat the pack has to shed. Dan-Tech's range for the same number is 0.2–0.7 W per cell, 1–2 W total across three cells. (Dan-Tech Energy) The active balancer, at 85% efficiency, recovers 0.85 × 0.72 = 0.61 Wh per cycle instead of burning it, but the recovered energy is not the argument. The time is.
Step 5, the active advantage. At 2 A, the middle of the active inductive range of 1.0 to 5.0 A (Electrical Equipment Factory), the same 0.175 Ah gap closes in 0.175 ÷ 2.0 = 0.088 hours, 5.3 minutes. Against the passive 1.75 hours at 100 mA, that is 20× faster. At 5 A, it is 2.1 minutes. And the active balancer can act during charge, discharge, or idle (Dan-Tech Energy), it can supply the low cell during discharge, which the passive balancer cannot do, because passive balancing has no transfer path to supply a low group from the remaining groups during discharge. (EVKX) The active balancer's advantage is not the 0.61 Wh it recovers. It is the 5 minutes against the 1.75 hours, and the discharge it can cover, and the idle it can use.
Step 6, the decision. For the bench pack, the recommendation is passive top balancing, and the reason is the research, not the price. The pack is built from capacity-screened cells, it is charged fully every cycle, and the 2025 Energies analysis found that, for packs built from capacity-screened cells, passive and active balancing are about equally effective. (Energies, via Ingro Energy) The passive balancer closes the 0.175 Ah gap in 1.75 hours at the top of a 2-hour charge, and the heat is 1.26 W across three cells, and the pack delivers its 159.9 Wh. The active balancer earns its cost when the pack ages, or when new modules are mixed with old ones, or when the pack is LFP and rarely reaches full charge, and then the 5 minutes, and the discharge, and the idle, are worth the converter. And the balance point is the top: the pack is charged fully every cycle, the risk is at the top (OVP 4.20–4.25 V per cell, Article #24), and the Orion Jr. 2's manual recommends top balancing. (EVKX)
Step 7, the early warning, and the test it has to pass. Track the balancing time per cycle. If the 1.75 hours becomes 3.5 hours, the drift doubled, and the cell that needs ever more balancing is the cell to replace, before it starts limiting the rest. (Ingro Energy) And the architecture has to survive the test it will be asked to pass: UN38.3, from Article #42, a transport test, not a product-safety approval, and T.5, the external short circuit, at 54.6 V across 0.1 Ω, is 546 A and 29,812 W. The fuse, the busbar, and the BMS's short-circuit protection are the architecture that survives it, and the lab fees are $4,500–$8,500 and 3–6 weeks, so the architecture that fails T.5 is the architecture that costs a quarter.
The decision test
Five questions. If the answer to any one of them is no, the architecture is a guess, and the decision is to measure first, and to build the architecture the measurement supports.
- What is the spread, in amp-hours, not in millivolts? The voltage spread is the symptom. The charge difference is the number, and the balancing time is the charge difference divided by the balancing current. If the spread is stated in millivolts, convert it: a 100 mV imbalance in a 4S pack is 5–15% of capacity, and the bench pack's 5% is 0.175 Ah. (doflagie)
- How long does the balancing take, and does the pack sit at the top long enough for it? At 100 mA, the bench pack's gap takes 1.75 hours at the top of charge. If the pack charges in 2 hours and is unplugged at full, the balancer never gets its window, and the drift grows 0.5% a month. (EVKX; Cell Saviors)
- Is the balancing passive or active, and does the pack need the active one? Passive is enough for capacity-screened cells charged fully every cycle. Active earns its cost when the cells age, when modules are mixed, or when the pack rarely reaches full charge. (Energies, via Ingro Energy)
- Top or bottom, and does the chemistry let the BMS see the drift? NMC's sloping curve can be balanced at almost any state of charge. LFP's flat curve hides the drift between 20% and 80% SOC, and the balancing has to happen above 3.40 V per cell, at the top, with occasional full charges. (Sandia; LJY Energy; Electrical Equipment Factory)
- Is the spread tracked, and is the trend watched? The cell that keeps drifting, needs ever more balancing, self-discharges faster, or runs warmer is the first visible sign of a fault. The BMS balances in real time; the analytics watches the trend over weeks and months. If the balancing time per cycle is not tracked, the pack has no early warning. (Ingro Energy)
Objections
"The cells are matched, we don't need balancing." The cells leave the factory with tiny differences in capacity, resistance and self-discharge. (Ingro Energy) Qnovo's own example is two cells from the same batch, nominally identical, in reality 5% apart in capacity, "and can readily happen in a pack without the proper precaution." (Qnovo) The matched cells are the ones that drift: the coulombic efficiency differs, the self-discharge differs, the temperature differs across the pack, and the gap grows a little on every cycle. The balancing is not for the cells as bought. It is for the cells as they age.
"Active balancing is better, just use active." Not always. Research shows both are about equally effective for packs built from well-matched cells. Active balancing is more useful when cell capacities differ significantly, for example in older batteries or when new modules are added beside old ones. (Ingro Energy) The active balancer costs more, is more complex, and its advantage is time and the discharge it can cover, not the energy it recovers, which on the bench pack is 0.61 Wh a cycle. If the pack is capacity-screened and charged fully every cycle, the passive balancer closes the gap in 1.75 hours at the top of a 2-hour charge, and the active balancer's 5 minutes buy nothing the pack needs.
"Passive balancing wastes energy, the pack should not burn charge as heat." The passive balancer burns 0.72 Wh a cycle on the bench pack, and 262 Wh over a year of daily cycles, and the pack it saves is 168 Wh a cycle, 61,320 Wh a year. The burned share is 0.43% of the energy the pack delivers. The waste is not the energy. The waste is the capacity the pack would lose without the balancer: 8.4 Wh a cycle, locked away behind the weakest cell, every cycle, for the life of the pack. The heat is the price of the capacity, and the price is less than half a percent.
"We'll just top up the weak cell by hand." The weak cell is weak because it drifts, and it drifts because its coulombic efficiency, its self-discharge, and its temperature differ from its neighbours, and the hand top-up fixes the charge level once, and the drift starts again on the next cycle. The balancing is not a one-time correction. It is a process that runs on every charge, at the top, for the life of the pack, and the BMS is the part of the pack that runs it, at 50 to 300 mA, on a threshold of 10 to 20 mV from the pack maximum. (Cell Saviors)
"LFP doesn't need balancing, the voltage is flat, so the cells are equal." The voltage is flat, so the cells look equal. Sandia notes that flat-voltage chemistries are bad for cell balancing during prolonged operation at partial state of charge, because cells look alike by voltage even when their charge levels differ. (Sandia, via Ingro Energy) The LFP pack that never reaches full charge is the pack that drifts without the BMS noticing, and the fix is the occasional full charge, for balancing and for recalibrating state of charge, not the absence of balancing.
Sources
- Ingro Energy, "Cell Imbalance and Balancing: Why the Weakest Cell Wins" (opened in full)
What we used
"Because every cell in a series string must stay inside its limits, the cell that fills first stops the charge and the cell that empties first stops the discharge, so a small drift can lock away a surprising amount of capacity." "In a series string, the fullest cell ends every charge and the emptiest ends every discharge." "Cells drift because of small manufacturing differences, amplified by temperature differences across the pack." "Passive balancing bleeds off excess charge as heat; active balancing moves charge between cells." "Active balancing pays off mainly when cell capacities differ significantly, as packs age or modules are mixed." "LFP's flat voltage curve makes balancing harder at partial charge, so occasional full charges help." "A cell that keeps drifting or needs ever more balancing is an early warning worth acting on." "In a series string, the same current flows through every cell. When charging, the battery management system has to stop as soon as any one cell reaches its upper voltage limit; when discharging, it has to stop as soon as any one cell reaches its lower limit. A 2024 paper in Scientific Reports puts it simply: if the cells are not properly balanced, the weakest cell will always be the one limiting the usable capacity of the pack." "The drawing below shows the effect. When cells sit at different charge levels, the fullest one ends every charge and the emptiest ends every discharge, and the capacity in between goes unused." "Cells leave the factory with tiny differences in capacity, resistance and self-discharge. A 2025 analysis of battery packs in the journal Energies found that, in series packs, charge inconsistency driven by differences in coulombic efficiency and initial state of charge is the main reason capacity is lost, while differences in resistance mostly limit power." "Temperature turns small differences into large ones. Sandia's energy storage handbook notes that minimising the temperature spread between cells minimises the variation in how they age. Field data bears this out: in a South Korean grid battery studied over three years, variation between cells within each bank grew every year, concentrated in racks with uneven cooling airflow." "Balancing is the battery management system's job of bringing cells back into line. Sandia's handbook describes the two approaches. Passive balancing: The most common approach. Switchable resistors bleed off excess charge from the highest cells, turning it into heat, until the others catch up. Active balancing: Capacitors or small converters move charge from stronger cells to weaker ones, wasting less energy, at higher cost and complexity." "The 2025 Energies analysis found that, for packs built from capacity-screened cells, passive and active balancing are about equally effective, and that active balancing outperforms passive mainly when cell capacities differ significantly, which becomes more likely as a battery ages or when new modules are mixed with old. Passive balancing corrects charge-level differences; it cannot make a smaller cell hold more." "LFP's very flat voltage curve, ideal for delivering steady power, makes balancing harder. Sandia notes that flat-voltage chemistries are bad for cell balancing during prolonged operation at partial state of charge, because cells look alike by voltage even when their charge levels differ. Sloping-voltage chemistries such as NMC can be balanced at almost any state of charge." "In practice, LFP batteries balance best near the top of charge, where voltages separate. A battery that never reaches full charge, because it is always cycled in a narrow middle window, may drift without the BMS noticing. That is a good reason for an EMS to schedule occasional full charges on LFP systems, for balancing and for recalibrating state of charge." "Cell spread is more than a capacity problem. A cell that keeps drifting away from its neighbours, needs more balancing than the rest, self-discharges faster or runs warmer is often the first visible sign of a fault developing. Fault-diagnosis research routinely flags cells by comparing them with the pack's typical cell, with one caveat noted in a 2020 review in the journal Algorithms: active balancing can mask the very differences such methods look for, so balancing activity itself is worth tracking." "The BMS balances and protects in real time; analytics watches the trend over weeks and months, notices which modules need ever more balancing, and turns that into a maintenance decision before the weak cell starts limiting the rest." Sources cited on the page: Khan et al., Scientific Reports, 2024 (a novel active cell balancing topology); Duan et al., Energies, 2025 (decoupling analysis of parameter inconsistencies); Sandia, DOE Energy Storage Handbook, Chapter 3, 2021; Mun et al., Energies, 2025 (DCIR and SOH in field-deployed ESS, thermal non-uniformity); Tran & Fowler, Algorithms, 2020 (fault diagnostic algorithms review); Communications Engineering, 2024 (degradation in parallel-connected packs under thermal gradients)
- EL-Cell, "How do you balance cells in a lithium-ion battery pack?"
What we used
"Cell balancing in a lithium-ion battery pack is the process of equalizing the state of charge (SoC) across all individual cells connected in series or parallel. Because no two cells are perfectly identical, differences in capacity, internal resistance, and self-discharge rate cause cells to reach full charge or full discharge at different times, reducing the usable capacity of the entire pack." "A pack is only as strong as its weakest cell: if one cell reaches its upper voltage limit before the others during charging, the charger must stop early to protect it, leaving the remaining cells undercharged. The same logic applies during discharge." "The two principal methods for cell balancing in a lithium-ion battery pack are passive balancing and active balancing. Passive balancing dissipates excess energy as heat, while active balancing redistributes energy between cells." "Passive balancing: Resistors bleed off charge from higher-SoC cells until they match the lowest cell in the pack." "Active balancing: Energy transfer circuits move charge from higher-SoC cells to lower-SoC cells, preserving usable energy." "Top balancing: Cells are equalized at the top of their charge window, ensuring a uniform full charge." "Bottom balancing: Cells are equalized at the bottom of their discharge window, ensuring a uniform full discharge." "The key distinction between passive and active cell balancing is energy efficiency. Passive balancing wastes excess charge as heat through a bleed resistor, making it simple and low-cost but thermally inefficient. Active balancing transfers charge between cells using inductors, capacitors, or DC-DC converters, recovering energy that would otherwise be lost."
- EVKX, "EV battery cell balancing: passive and active methods"
What we used
"Cell balancing adjusts the charge held by series-connected cell groups so that differences between them do not unnecessarily restrict the battery's usable energy. Passive systems remove charge from selected groups; active systems transfer energy between parts of the pack." "Balancing time (h) = charge difference (Ah) ÷ average balancing current (A)." "For two otherwise identical 100 Ah groups, a difference of one percentage point in SOC represents 1 Ah. Removing that difference at a continuous 0.1 A takes 10 hours. If that branch operates for only half the elapsed time at the same current, its average current is 0.05 A and the correction takes 20 hours." "Net charging current into the cells = pack charging current − bleed current." "Passive balancing also has no transfer path to supply a low group from the remaining groups during discharge." "NXP's MC33775A automotive cell controller supports passive balancing up to 300 mA per channel." "Top balancing aligns groups near the upper end of the charge window; bottom balancing aligns them near the lower end. These terms describe the alignment point. Passive and active describe the mechanism used to change charge distribution." "One documented implementation is the Orion Jr. 2, a BMS for packs up to 48 V nominal in light vehicles and stationary applications. Its 2018 manual describes adjustable balancing thresholds and recommends top balancing." "Aligning them at the top of the charge window cannot restore the smaller group's missing capacity."
- Cell Saviors, "How to Balance Lithium Batteries: Series, Parallel, Active vs Passive"
What we used
"The BMS activates the bypass resistor on any cell that exceeds a set threshold (usually within 10 to 20 mV of the pack maximum). The current through the resistor is the balance current, typically 50 to 300 mA for BMS units." "Passive balancing only works at the top of charge. The BMS can only bleed off charge from high cells while the pack is near full, it can't push charge into low cells. This means balancing happens during the final stage of charging, when all cells are approaching full." "Cells in parallel self-balance through direct electrical connection, current flows between cells until their voltages equalize."
- Qnovo, "What is cell balancing?"
What we used
"If the cells are electrical tied in series, then the pack is called s-configuration. If they are tied in parallel, then they are in a p-configuration. The former serves to raise the maximum voltage of the pack in multiples of 4.35V, whereas the latter serves to increase the maximum current through the pack without increasing the voltage." "In a parallel configuration, the voltage is always equal for both cells. Any difference in charge capacity between the cells will manifest itself as a difference in current in the two branches. In particular, this parallel configuration always guarantees that the cells do not exceed their maximum safe voltage, often 4.35V." "If the two cells are truly identical, then they will reach their maximum capacity and their maximum voltage at the same moment. But if there is a difference in capacity between them, then the cell with a smaller capacity will reach 4.35V before the other cell does. At this point time, one cell is 100% full while the other one is not. If the charging is not disconnected immediately, one cell will certainly get overcharged and cause a hazard." "They are both nominally 7,000 Coulombs (or about 1,900 mAh) but in reality, one cell is 7,200 Coulombs and the other one is 6,800 Coulombs. This is about 5% difference in capacity, and can readily happen in a pack without the proper precaution."
- Maxwell Energy, "Cell Balancing in BMS: How It Works and Why Li-ion Needs It"
What we used
"A lithium-ion battery pack is only as strong as its weakest cell." "A pack with 5% cell voltage spread at end of charge cannot access 100% of its rated capacity. The energy sitting above the weakest cell's cutoff in every other cell is stranded." "Active balancing moves it to weaker cells at 80 to 95% efficiency." "Cell balancing interrupts this cycle. By keeping all cells at equivalent SOC throughout charge and discharge, balancing distributes aging stress evenly across the full pack, extending the cycle life of the weakest cells and slowing the overall pack degradation rate."
- LJY Energy, "LiFePO4 Cell Balancing: Passive vs Active, 7 Specs to Check"
What we used
"LiFePO4 has one of the flattest voltage curves of any lithium chemistry. Between roughly 20% and 80% state of charge, cell voltage barely moves as charge is added or removed. Voltage-based balancing depends on detecting voltage differences between cells, and in the middle of the curve there are no meaningful differences to detect." "Meaningful differences only appear in the 'knee' above roughly 3.4V per cell, exactly where balancing is designed to work." "Enabling balancing below about 3.35V per cell can actually leave a pack *less* balanced, because the balancer acts on voltage readings that don't represent real state-of-charge differences. The conventional wisdom in experienced communities is to let balancing start only near the top, typically in the 3.40–3.55V per cell range reported across BMS parameter sheets." (labelled by the guide as buyer-community consensus, not a manufacturer standard) "Passive balance currents commonly fall in the 35–100mA range." (labelled by the guide as industry experience) "The buying test is a single equation, mismatch amp-hours ÷ balance current = hours needed, and most disappointments come from a passive 50mA balancer being asked to close a gap it would need 200 hours to fix."
- Electrical Equipment Factory, "Lithium Ion Cell Balancing: BMS Engineering Design Guide"
What we used
"The fundamental constraint governing any series string is expressed as: Qpack = min(Q1, Q2, ..., Qn)." "A self-discharge difference as slight as 0.5% per month will accumulate into a 6% capacity deficit within a single year if left uncorrected." "Passive cell balancing dissipates excess energy as heat through shunt resistors, typically restricted to bypass currents below 300 mA due to enclosure thermal dissipation limits." "Active battery balancer architectures transfer charge between cells using capacitive or inductive topologies, achieving currents up to 5 A at efficiencies exceeding 85%." "For lithium iron phosphate (LFP) chemistry, charge balancing must be executed exclusively in the upper voltage inflection zone above 3.40 V per cell." Balancing current table: passive dissipative 20 mA to 300 mA, 0% efficiency; active switched capacitor 500 mA to 2.0 A, 75% to 88%; active inductive / flyback 1.0 A to 5.0 A, 85% to 94%. Worked calculation: an industrial rack utilising 280 Ah LFP prismatic cells, V_cell = 3.45 V at top of charge, R_shunt = 39 Ω → I_bal = 3.45 / 39 = 0.0885 A (88.5 mA)
- Dan-Tech Energy, "Active vs Passive Cell Balancing"
What we used
"Passive balancing bleeds the higher cells in a series string down to match the lowest one, using a resistor and heat, and it can only act while the pack is charging. Active balancing moves charge between cells instead of burning it off, using one of three circuit topologies, and it can act during charge, discharge, or idle." "A passive bleed circuit typically draws somewhere in the 50-200mA range per cell being bled, which is the industry-typical range for resistor-based balancing designs. At a nominal cell voltage around 3.6-3.7V, that works out to roughly 0.2-0.7W dissipated as heat per cell being actively bled, or on the order of 1-2W total across those three cells for the duration of the balancing window."
- doflagie, "Sch Cell Balancing"
What we used
"The LOWEST voltage cell determines discharge cutoff. The HIGHEST voltage cell determines charge cutoff. A 100 mV imbalance in a 4S pack can reduce effective capacity by 5–15%. A 300 mV imbalance → pack appears nearly full on low end, nearly empty on high end." "TOP-BALANCE: equalize during charging (cells reach 100% together). Most common approach. Charger holds near top-of-charge while balance dissipates." "BOTTOM-BALANCE: equalize during discharge (cells reach 0% together)." "I_bal = V_cell / R_BAL. For R_BAL = 10Ω, V_cell = 3.6V (LiFePO4 full): I_bal = 360 mA. For R_BAL = 47Ω: I_bal = 76 mA." "TIME TO BALANCE: For 50 mV imbalance at C = 50 Ah: ΔQ = ΔV × C / V_cell = 0.05 × 50 / 3.6 = 0.69 Ah; t = ΔQ / I_bal = 0.69 / 0.360 = 1.9 hours (at 360 mA)." "Passive balancing is adequate when: Pack < 20 Ah (fast to balance even at low current); Low cycle rate (weekend activations); Cells are well-matched from factory."
- Ayalasomayajula et al., "Physics-Based Analysis of Cell Imbalances and Aging in Lithium-Ion Battery Modules and Packs,"
What we used
Journal of the Electrochemical Society, 2025 (Sandia and Oak Ridge National Laboratory), "Simulations of an 8-cell parallel-connected module demonstrate that reducing current imbalance leads to more uniform cell aging and improved module/pack-level degradation predictions." "Busbar interconnect resistance is one of the key factors affecting current imbalance in lithium-ion battery modules and packs, especially those with parallel connections." "Terminal location significantly affects imbalance, with opposite-end terminal connections at intermediate branches minimizing the imbalance." "The pack circuit construction influences the accuracy of physics-based analysis at the pack scale."
- ZSD Power, "LiFePO4 battery pack cell balancing: passive vs active methods compared"
What we used
Bench test of 48V/100Ah LFP packs (16S) under 0.5C partial-state-of-charge cycling: "cell voltage spreads of 60–120mV were common after 200 cycles under 0.5C partial-state-of-charge (PSOC) conditions… At that spread level, usable capacity had dropped to roughly 88% of rated. By contrast, the same cell chemistry managed by an active balancer held spread below 20mV through 200 cycles." At 500 cycles: passive 80–150mV, active 10–30mV. (Vendor bench test, labelled as such; the article uses the 200-cycle figures only as a cross-check.)
- Demo arithmetic verified in Python before writing: 13 × 3.7 = 48.1 V nominal; 13 × 4.2 = 54.6 V full; 13 × 3.0
What we used
= 39.0 V empty; 48.1 × 3.5 = 168.3 Wh (cluster: 168 Wh); weakest cell 3.5 × 0.95 = 3.325 Ah; usable 48.1 × 3.325 = 159.9 Wh; locked away 8.4 Wh = 5.0%; after a year 3.5 × 0.94 = 3.29 Ah, usable 158.2 Wh, loss 10.1 Wh = 6.0%; gap 3.5 × 5% = 0.175 Ah = 175 mAh; balancing time at 50 mA = 3.5 h, at 100 mA = 1.75 h, at 300 mA = 35 min, at 2 A = 5.3 min, at 5 A = 2.1 min; EVKX cross-check 100 Ah × 1% = 1 Ah, 10 h at 0.1 A, 20 h at 0.05 A; energy bled per cycle 0.175 × 4.1 = 0.72 Wh, over 365 cycles 262 Wh; active recovers at 85% = 0.61 Wh per cycle; heat 3 cells × 4.2 V × 0.1 A = 1.26 W, all 13 at once 5.46 W; per-cell bleed power 0.42 W (Dan-Tech 0.2–0.7 W); net charge into a bled cell 1.75 − 0.10 = 1.65 A; charge time 3.5 / 1.75 = 2.0 h; active 2 A vs passive 100 mA = 20× faster; doflagie cross-check ΔQ = 0.05 × 50 / 3.6 = 0.69 Ah, t = 1.9 h at 360 mA; 100 mV imbalance = 5–15% of 168 Wh = 8.4 to 25.2 Wh locked away