1. Introduction — Temperature Is the Battery’s Real Limit

Lithium-ion cells can deliver their rated energy only inside a narrow temperature window. Below 0 °C, charging plates lithium and permanently loses capacity; above 45 °C, calendar ageing accelerates by roughly a factor of two for every 10 °C; above 60 °C the cell enters the region where separator shutdown, electrolyte decomposition and thermal runaway become credible. Between those limits sit the practical constraints of any pack design: the cell’s internal resistance grows at low temperature and falls at high temperature, so the heat generated is itself a function of temperature — a positive feedback loop that must be analysed rather than assumed benign. Thermal management is therefore not an accessory to a battery design; it is a first-order constraint on the cell chemistry choice, the pack architecture, the charge/discharge limits and the mechanical design. This guide builds the heat-generation model from the cell’s resistance, constructs the thermal resistance network from cell to ambient, compares the cooling architectures (natural convection, forced air, liquid cold plate, phase-change materials), derives the spacing and airflow requirements, and works a complete example of a 6S4P pack in a sealed enclosure. It complements the battery design and charger design guides on this site.

2. Heat Generation Inside the Cell

The dominant heat source is the ohmic loss in the cell’s internal resistance, plus the entropic heat of the electrochemical reaction:

Q = I²·R_internal + I·T·(dU/dT)

The first term dominates at high current and is always positive; the second is the reversible entropic term, which can be negative during charging and is usually small compared with the first above moderate C-rate. The key subtlety is that R_internal is strongly temperature-dependent through the Arrhenius relation:

R(T) = R_ref · exp( Ea/k · (1/T − 1/T_ref) )

with an activation energy Ea in the 20–60 kJ/mol range, so R can easily double between 25 °C and 0 °C and fall by 30% between 25 °C and 45 °C. The consequence for design: a pack sized for a given heat dissipation at 25 °C will generate substantially less heat once warm (self-limiting to a degree) and substantially more when cold — which is why a battery that performs perfectly in the lab at room temperature can overheat in a hot vehicle and can fail to charge at all in winter. The resistance also rises with age and with the state of charge at the extremes, so the worst-case heat estimate must use the end-of-life resistance (often taken as 1.5–2× the beginning-of-life value) and the lowest permissible operating temperature. Current-squared scaling means a 2C discharge generates four times the heat of 1C: the C-rate chosen for performance has a quadratic thermal price, and the converter/load power budget is where that trade must be evaluated (the electrical side of the same calculation is a natural place for the power calculator).

3. The Thermal Resistance Network

Heat flows from the cell’s jelly roll to the case, through the interface material, into the pack structure and finally to the ambient. Each step is a thermal resistance, and the network is analysed exactly like an electrical circuit:

T_cell = T_ambient + Q · (R_jc + R_interface + R_spreader + R_conv)

Path element	Typical value	Comment
Cell internal (junction to case)	2–10 K/W for a cylindrical 18650; lower for prismatic/pouch	Dominated by the jelly roll’s radial conduction; short in the axial direction
Cell-to-holder interface	1–5 K/W	Air gaps dominate; a thermal pad can cut it by 3–5×
Spreader / pack structure	0.5–3 K/W	Aluminium plate; the spreader’s job is to make the surface isothermal
Convection to ambient (natural)	5–20 K/W	Depends on surface area and orientation; a sealed plastic box is worst
Convection to ambient (forced, 2 m/s)	1–4 K/W	A small fan can be the cheapest fix if the airflow path is clear
Conduction to a chassis	0.2–2 K/W	Best when a metal chassis is available

Two rules follow from the network. First, the largest single resistance dominates the temperature rise — improving anything else is wasted effort, so measure or estimate each element before choosing a mitigation. Second, the internal resistance (R_jc) sets a floor: no cooling system can remove heat from the jelly roll faster than it conducts to the case, so a cell with a 8 K/W internal resistance at 5 W of dissipation will be 40 °C above its case no matter what the pack does. This is why high-C-rate designs use many parallel cells (which divides both the current per cell and the heat) rather than one large cell. The steady-state temperature rise and the required thermal resistance for a given limit are computed directly with the thermal calculator; the same tool verifies the interface-material and heatsink figures before the mechanical design is frozen.

4. Cooling Architectures

Architecture	Capability	Cost / complexity	Typical application
Natural convection	<1–2 W per cell	Lowest	Small consumer packs, low C-rate
Forced air	2–10 W per cell	Low; needs air paths and a filter	Power tools, e-bikes, stationary racks with fans
Liquid cold plate	>10 W per cell	High: pump, coolant, seals, service	EV packs, high-power grid storage
Phase-change material (PCM)	Buffers peaks; no steady-state removal	Medium; adds mass	Pulsed loads, drones, peak-shaving
Heat pipe / vapour chamber	Transports heat to a remote sink	Medium-high	Space-constrained or sealed enclosures

Natural convection with a well-designed conduction path is the most cost-effective option for packs below roughly 1 W per cell; above that, forced air or a cold plate is required, and the choice usually follows the enclosure’s sealing requirement (a sealed IP-rated box cannot use forced air without a heat exchanger). For pulse loads, a PCM can absorb a burst of heat that would otherwise trip the temperature limit, but it does not remove the average heat — the steady-state path must still exist. The distribution of the cooling matters as much as its capacity: cells near the airflow inlet run cooler than those at the outlet, and the temperature difference across a pack directly causes capacity mismatch, unbalanced state of charge and accelerated ageing of the hottest cell. Design targets of ≤5 °C cell-to-cell difference and ≤10 °C cell-to-ambient rise are common for a well-engineered pack.

5. Mechanical Design: Spacing, Interfaces, Direction

The mechanical details decide whether the calculated thermal resistance is achieved in reality. Rules that matter most:

• Spacing: cylindrical cells dissipate heat mainly through their radial surface, so adjacent cells should not touch. A gap of 1–3 mm between cells, with the airflow (or the conduction path) crossing that gap, typically halves the thermal resistance compared with cells stacked in contact. A hexagonal close packing in a plastic holder with air gaps is the standard compromise between density and cooling.
• Anisotropy: a cylindrical cell’s axial conductivity (through the metal can ends) is far higher than the radial path; prismatic/pouch cells conduct best through their flat faces. Orient the cooling path to exploit the low-resistance direction.
• Interfaces: air gaps between the cell and the holder are 3–5× worse than a filled gap; thermal pads, gap fillers or potting material convert the air gap into a conductive path. Never rely on the cell’s own wrapper as a thermal interface.
• Spreaders: an aluminium plate or a heat spreader makes the cooling surface isothermal so that the cell-to-cell difference stays small; without it, the cells nearest the cooling inlet are cold and the remote ones are hot.
• Direction of airflow: for a rack, pass the air across the cell axes rather than along the pack’s length, so every cell sees a similar temperature; if forced air is used, the outlet air must not be recirculated into the inlet.
• Sealing and condensation: a sealed enclosure must move heat through its walls (conduction to the chassis) because convection inside a sealed box is poor; and any liquid-cooled design must be evaluated for condensation on the cold plate in humid conditions.
• Temperature sensing placement: measure the cell surface at the hottest location (near the positive terminal for a cylindrical cell during discharge, where the current density is highest) and use more than one sensor for packs larger than a few cells; the BMS’s protections are only as good as its thermistor placement.

6. Limits, Protections and Fast-Charge Interaction

The BMS’s thermal protections must reflect the cell’s real limits, not the marketing specification: charge inhibit below 0 °C (some chemistries allow a limited low-temperature charge with reduced current), charge current derating above ~40 °C, discharge derating as the cell approaches its upper temperature limit, and an absolute cut-off above the cell’s specified maximum. Fast charging interacts directly with thermal management: a charge protocol that pushes current until the cell hits its voltage or temperature limit generates the most heat exactly when the pack is least able to dissipate it (at a low state of charge the current is high and, in a hot environment, the ambient is high too). The practical approach is a charge profile whose current is limited by the cell’s temperature as well as by its voltage — a temperature-derated current limit implemented in the charger’s control loop rather than a fixed current. In cold conditions, a low-rate self-heating phase (charging with a small current that heats the cell through its own resistance, or using a heater element) is often necessary before the main charge can start; the energy cost of heating the pack is part of the system’s energy budget, and it should be accounted for when comparing chemistries or pack architectures. Finally, at the system level, the thermal design must survive a single-point fault: a failed fan, a blocked airflow path or a stuck coolant valve must trip a temperature-based power limit rather than allow the pack to run to destruction.

7. Worked Example — 6S4P Pack in a Sealed Enclosure

Target: a 6S4P pack of 18650 cells (2.5 Ah, 20 mΩ each), continuous 30 A discharge (7.5 A per cell, 3C), sealed aluminium enclosure 200 × 120 × 80 mm, ambient 35 °C, the cell’s specified maximum surface temperature 60 °C.

• Heat generation: per parallel group, 7.5 A through 4 cells of 20 mΩ → each cell carries 1.875 A, so Q_cell = 1.875² × 0.020 = 70 mW at beginning of life. That seems tiny — but use the end-of-life resistance (1.5×) and the worst-case temperature (R doubles at 0 °C, falls ~30% at 45 °C): at 45 °C, R ≈ 14 mΩ → Q_cell ≈ 49 mW; at 0 °C, R ≈ 40 mΩ → Q_cell ≈ 140 mW. The pack total (24 cells) is about 1.7 W at 25 °C and 3.4 W at 0 °C. The low per-cell figure is the payoff of the 4P configuration: parallel cells divide both the current and the I²R heat.
• Thermal network: cell internal R_jc ≈ 6 K/W, holder interface 2 K/W, spreader-to-case 1 K/W, case-to-ambient (natural convection, ~0.05 m² effective, sealed aluminium) ≈ 4 K/W. Total ≈ 13 K/W per cell’s local path, but with a spreader the pack shares the case area: use the pack-level network, R_total ≈ 0.8 K/W pack-level with the 24 cells’ heat spread over the case.
• Temperature rise: ΔT = Q × R = 1.7 W × 0.8 K/W ≈ 1.4 K at 25 °C — the sealed case is not the limit for this continuous current. At 0 °C with 3.4 W: ΔT ≈ 2.7 K, still small. The design’s limit is therefore the cell’s internal gradient, not the case: verify with the thermal calculator that the jelly roll’s hot spot stays below the 60 °C surface spec plus the internal gradient (a few K at these currents).
• Worst case: the real design case is a fast charge at 20 A in a 35 °C ambient. The pack resistance is 6 series groups of four 20 mΩ cells in parallel: R_pack = 6 × (20 mΩ / 4) = 30 mΩ. At 20 A, Q_total = I²R = 400 × 0.030 = 12 W. ΔT = 12 × 0.8 = 9.6 °C → the case reaches ~45 °C and the cells ~48 °C, acceptable but close; derate the charge current above 40 °C ambient.
• Mitigations, in the order of the network’s largest term: (1) fill the cell-to-holder gaps with a thermal pad to cut the interface resistance; (2) add an aluminium spreader plate to make the case isothermal and reduce the cell-to-cell difference to a few K; (3) if the enclosure stays sealed and the charge current must rise, conduct to the external chassis through a thermal bridge or derate the current with temperature. The order matters: improve the dominant resistance first, and verify the result with the thermal calculator after each change.
• Sensing and protection: thermistors on two locations inside the pack (the centre of the array and near an end cell), charge inhibit below 0 °C, charge derating above 40 °C, discharge derating above 55 °C, and a hard cutoff at 60 °C with hysteresis.
• Verification: instrument a prototype pack with 6–8 thermocouples, run the worst-case charge and discharge, and record the cell-to-cell spread; the measured resistances then replace the estimates in the network model.

8. Thermal Test and Validation

A thermal design is not finished until it has been measured, and the measurements that matter are the ones taken at the worst case: the highest ambient the product will see, the fastest charge the charger will permit, and the heaviest load the system will draw — simultaneously. A short Python model of the thermal network, calibrated against a single measurement, is enough to predict the rest of the operating envelope and to decide where the limits must be set:

"""First-order (two-node: cell -> surface) thermal model of a battery pack."""

# Measured thermal masses and resistances, calibrated from one test run
C_cell   = 185.0    # J/K   cell mass (4P x 3.4 Ah x 3.7 V ~ 250 g, cp ~ 0.85 J/gK)
C_surf   = 90.0     # J/K   case/plate mass in good contact with the cells
R_int    = 1.1      # K/W   cell -> surface  (holder + plate + interface)
R_amb    = 2.4      # K/W   surface -> ambient (natural convection, sealed box)

def simulate(I_load, I_charge, minutes, dt=1.0, T_amb=25.0):
    T_c = T_s = T_amb
    n = int(minutes * 60 / dt)
    for k in range(n):
        # Ohmic heat: cells only, split evenly across the parallel group
        P_cell = I_load**2 * 0.012 + I_charge**2 * 0.008      # W, calibrated
        T_c += (P_cell - (T_c - T_s) / R_int) * dt / C_cell
        T_s += ((T_c - T_s) / R_int - (T_s - T_amb) / R_amb) * dt / C_surf
        if T_c > 60.0:
            return k * dt, T_c, T_s, "CELL LIMIT REACHED"
    return minutes * 60, T_c, T_s, "ok"

t, Tc, Ts, msg = simulate(I_load=8.0, I_charge=3.0, minutes=45)
print(f"t={t:5.0f}s  T_cell={Tc:5.1f}C  T_surf={Ts:5.1f}C  {msg}")

The value of such a model is not precision — its parameters are only approximate — but direction: it tells you immediately whether the cell-to-ambient path or the internal generation dominates, and therefore whether the fix is a bigger heat spreader (R_amb) or a lower-resistance interface (R_int). Run it once with the constants above, then adjust R_amb or R_int by 30% and observe the change in the peak temperature; the component that matters will make itself obvious.

The test set itself should include: a thermocouple or an infrared spot on the cell body (not on the terminal or the PCB), a second sensor on the enclosure surface, ambient logging beside the product, and a current log synchronised with the temperature log. For pack-level tests, an insulated enclosure held at the maximum rated ambient is the honest worst case — testing at 25 °C open-bench and subtracting a margin is how packs pass in the lab and fail in the field. The specific measurements worth recording are:

• Steady-state rise per watt at a fixed current — the empirical R (in K/W) that can be compared between mechanical variants.
• The thermal time constant (the time to reach 63% of the final rise), which determines how long a load can be sustained before the limit is reached.
• The temperature difference between cells in the pack — imbalance above 3–5 °C accelerates the ageing of the hottest cell and, in a series string, changes its effective capacity, so the pack’s usable capacity is set by that cell.
• The charge/discharge limit entry points — the temperature at which the BMS begins derating, which should be well before the cell’s absolute limit and validated as a smooth, hysteresis-free transition.

One last point connects thermal validation back to the BMS: the temperature sensors’ placement must be chosen during this test, not before it. A pack with a single sensor placed on the coolest cell will not protect the hottest one; the standard practice is one sensor per series group (or per two groups) placed on the cell body, plus validation that the sensor-to-cell offset measured at full load is recorded in the BMS configuration rather than assumed to be zero.

9. Common Mistakes

• Using the room-temperature internal resistance: the cold and end-of-life resistance is 2–3× higher; the worst-case heat estimate must use those values.
• Ignoring the internal gradient: the case temperature is not the cell temperature; a 5–10 K internal gradient is common at high C-rate, and the BMS is measuring the surface.
• Stacking cells in direct contact: the thermal resistance rises sharply and the centre cells overheat first.
• Sealed plastic enclosure with natural convection: the worst possible thermal configuration; conduct to a metal chassis or accept a much lower current.
• Cooling the wrong direction: a cylindrical cell’s axial path is far better than the radial one; orient and design the spreader accordingly.
• No single-fault protection: a blocked airflow path or a failed fan must trip a temperature-based power limit, not merely raise the temperature.
• Charging below 0 °C at high current: permanent lithium plating; inhibit or limit the charge current when cold.

10. FAQ

Q: What cell-to-cell temperature difference is acceptable? A: Below about 5 °C for good life; every degree of imbalance accelerates the ageing of the hottest cell and worsens the capacity mismatch.

Q: Is a Peltier cooler a good idea for a battery? A: Rarely — it removes heat at the cost of significant extra electrical power and rejects it into the same enclosure; the energy cost usually exceeds the benefit unless the ambient is unusually hot.

Q: How do I choose between air and liquid cooling? A: From the required heat rejection per cell and the sealing requirement: below a few watts per cell with an open enclosure, air is almost always cheaper; a sealed high-power pack needs a cold plate or conduction to a chassis.

Q: Does higher temperature really double the ageing rate? A: For calendar ageing, an Arrhenius-type rule of roughly 2× per 10 °C above room temperature is a common approximation; the exact factor depends on chemistry and state of charge.

Q: Should the pack be heated in cold weather? A: Yes, if the product must charge below 0 °C: a low-rate self-heating phase or a heater element is required, and its energy cost must be included in the system budget.

11. Conclusion

Battery thermal management is a resistance-network problem with a temperature-dependent source. Compute the ohmic heat with the worst-case resistance and the worst-case C-rate, build the cell-to-ambient network and improve its dominant element, design the mechanical layout so that the calculated resistances are achieved (gaps, spreaders, orientation, interfaces), limit charge and discharge by temperature as well as by voltage, and place the sensors where the cells are hottest. Temperature is the battery’s real performance limit; treating it as a design input rather than an afterthought is what makes a pack last.