Heat Generation in Lithium-Ion Cells and Packs — Where the Energy Goes and Why It Matters

The energy your pack generates that never moves the car

Every time a lithium-ion cell charges or discharges, some fraction of the energy flowing through it becomes heat instead of useful work. In a 77 kWh pack delivering power to the drivetrain, that fraction is small — typically 2–5% during normal driving. But "small fraction of a large number" still means 1.5–4 kW of heat dissipating continuously inside a sealed enclosure. During a 150 kW DC fast charge, the same pack can generate 3–6 kW of heat in a box the size of a large suitcase.

Understanding where that heat comes from — at the electrochemical level, at the cell level, and at the pack level — is what separates a thermal management system that works from one that merely delays failure.

Table of Contents

1. The two sources of heat in a lithium-ion cell

2. Joule heating: resistive losses from the inside out

3. Entropic heat: the reversible component most engineers ignore

4. How cell format affects heat generation and removal

5. Heat generation at the pack level

6. The temperature gradient problem inside a cell

7. Why SOC and C-rate interact to create heat spikes

8. Heat generation during fast charging: the worst-case scenario

9. What happens when heat removal fails: the runaway chain

10. Design implications for thermal management engineers

11. References

1. The Two Sources of Heat in a Lithium-Ion Cell

Heat generation inside a lithium-ion cell has two fundamentally different origins that behave differently with temperature, current, and SOC. Conflating them leads to thermal models that are wrong in exactly the conditions where accuracy matters most.

Source 1 — Joule heating (irreversible): Electrical energy dissipated as heat due to ohmic resistance. This is always positive — it generates heat regardless of whether the cell is charging or discharging. It scales with the square of current. It is the dominant source at high C-rates.

Source 2 — Entropic heat (reversible): Heat exchanged between the cell and its environment due to the entropy change of the electrochemical reaction. Unlike Joule heating, this can be either exothermic (generating heat) or endothermic (absorbing heat), depending on the cell chemistry, the direction of current flow, and the SOC. It is the dominant source at low C-rates and near rest.

The total heat generation rate inside a cell is the sum of both:

Q_total = Q_joule + Q_entropic

>

Q_total = I² × R_internal + I × T × (dU/dT)

>

where I = current (A), R_internal = total internal resistance (Ω), T = absolute temperature (K), and dU/dT = entropic coefficient of the cell voltage (V/K).

Most simplified thermal models only include the I²R term. This is adequate at high currents — but at low discharge rates, the entropic term can rival or exceed Joule heating, and it has the wrong sign for some chemistry-SOC combinations. A model built on I²R alone will predict the wrong temperature in slow-discharge or standby scenarios.

2. Joule Heating: Resistive Losses From the Inside Out

Joule heating is the more intuitive of the two mechanisms. Current flowing through a resistive path generates heat — the higher the current, and the higher the resistance, the more heat. In a lithium-ion cell, the resistive paths are not a single number. They are a stack of components, each contributing differently:

Total DC internal resistance (DCIR) — the number typically cited on a cell datasheet — is the sum of all these contributions measured at a specific temperature, SOC, and pulse duration. A typical 50 Ah prismatic NMC cell might show DCIR of 0.8–1.5 mΩ at 25°C and 50% SOC.

The Joule heating power for that cell at a 1C discharge rate (50 A) is:

Q_joule = I² × R = 50² × 0.001 = 2.5 W

At a 2C rate (100 A):

Q_joule = 100² × 0.001 = 10 W

The quadratic relationship is critical: doubling the current quadruples the heat generation. This is why fast charging generates disproportionately more heat than normal driving — the current, not the energy throughput, is what drives heat generation.

Temperature dependence of DCIR

DCIR is not constant. It changes strongly with temperature, and that dependence is asymmetric:

This has a self-reinforcing implication: a cold cell generates more heat per unit of current than a warm one, and that additional heat warms the cell, which reduces its resistance, which reduces heat generation. Cold cells self-regulate toward moderate temperatures under load — but only if the load is sustained. If a cold cell is subjected to a high-current pulse before it has warmed, the elevated resistance can drive the anode potential below zero vs Li/Li⁺, triggering lithium plating before the thermal self-regulation has time to act.

What this means for thermal management: Cold-soak protection is not just about slowing the charge rate. It is about giving the cell enough time at a low current rate to warm its own electrolyte — reducing DCIR — before the current demands that require a low-resistance path. Pre-conditioning via pack heater is faster and more controllable than self-heating under load.

3. Entropic Heat: The Reversible Component Most Engineers Ignore

The entropic heat term — I × T × (dU/dT) — is less intuitive but equally real. It arises because the electrochemical reactions inside the cell involve changes in the structural order of the electrode materials, and those structural changes have a thermodynamic heat exchange with the surroundings.

The sign and magnitude of dU/dT — the entropic coefficient — varies with both chemistry and SOC:

For an NMC/graphite cell with dU/dT = −0.5 mV/K at 25°C (298 K), the entropic heat rate at 1C (50 A) is:

Q_entropic = I × T × (dU/dT) = 50 × 298 × (−0.0005) = −7.45 W

The negative sign means heat is generated during discharge (the standard convention reverses for discharge current direction). At 7.45 W entropic versus 2.5 W Joule at 1C, the entropic term is actually the dominant heat source at this current.

At 2C (100 A), the same cell generates:

• Joule: 10 W (scales as I²)

• Entropic: 14.9 W (scales as I)

At 3C (150 A):

• Joule: 22.5 W (scales as I²)

• Entropic: 22.4 W (scales as I)

The crossover point — where Joule heating exceeds entropic heating — occurs at approximately 3C for this example cell. Above 3C, Joule heating dominates and scales rapidly. Below 3C, entropic heat is equal to or larger than Joule heat, and any model that ignores it is systematically wrong.

The LFP complication: LFP cells have SOC windows where dU/dT is slightly positive — meaning the cell is endothermic during discharge and actually absorbs heat from its environment. This is why well-instrumented LFP pack temperature measurements sometimes show a slight temperature drop during slow discharge in a warm environment. Thermal models built on NMC assumptions applied to LFP packs can predict the wrong sign of temperature change at low loads.

4. How Cell Format Affects Heat Generation and Removal

Heat generation rate (in watts per unit volume) is primarily a function of current density and internal resistance — both of which scale with cell format. But equally important is how efficiently the generated heat can reach the cooling surface. The ratio of heat-generating volume to heat-removal surface area differs substantially between formats.

Cylindrical cells

A 21700 cell (21 mm diameter, 70 mm tall) has a volume of approximately 24 cm³ and a lateral surface area of approximately 46 cm². In a conventional side-cooled module, only the lateral surface participates in heat removal. The core of a 21700 cell — the jelly roll centre — can be 5–8°C hotter than the surface during 2C discharge, and up to 15–20°C hotter during a 3C pulse.

The 4680 cell (46 mm diameter, 80 mm tall) exacerbates this: volume scales as r², surface area scales as r, so the volume-to-surface ratio is 2.2× worse than the 21700. Tesla's tabless electrode design addresses this specifically — by eliminating the discrete tabs and allowing current to flow across the full electrode width, the effective internal resistance is reduced and heat generation is distributed more uniformly across the jelly roll rather than concentrated near the tabs.

Prismatic cells

Large-format prismatic cells (e.g., 100–300 Ah) have a flat geometry that offers better surface-to-volume ratios than large cylinders. The wide flat faces are natural cooling surfaces. But they introduce a different problem: cell-to-cell variation in swelling. As NMC prismatic cells cycle, the electrode stack expands (1–3% per full cycle over lifetime), and this expansion is constrained by the module housing. The contact pressure between cell and cooling plate changes over the cell lifetime, and variable contact pressure means variable thermal resistance — which means cells at different positions in a module may run at different temperatures even under identical electrical load.

Pouch cells

Pouch cells have the best surface-to-volume ratio of the three formats and the most flexibility in dimensions. The thin polymer laminate casing allows direct contact cooling across the entire large face. The challenge: pouch cells swell significantly during both cycling and calendar aging. A pouch cell may expand by 5–10% in thickness over its service life. Cooling plate designs must accommodate this swelling without losing thermal contact — typically through compliant thermal interface materials (TIMs) with sufficient compression range.

5. Heat Generation at the Pack Level

The cell-level analysis above treats a single cell in isolation. In a pack, additional heat sources compound the picture.

Bus bar and interconnect resistance

In a 400V, 200A discharge pack, the bus bars and inter-cell connections carry substantial current. A bus bar with 0.5 mΩ resistance at 200 A generates:

Q = I² × R = 200² × 0.0005 = 20 W per bus bar

A pack with 100 series-connected cell groups, each with interconnect resistance of 0.3 mΩ, generates 200² × 0.0003 × 100 = 1,200 W from interconnects alone at 200 A — comparable to the heat from the cells themselves during moderate discharge.

Contactor and fuse losses

Main contactors for a 400V automotive pack typically have contact resistance of 0.5–2 mΩ each. At 200 A, two contactors generate 20–80 W combined. Pyrofuses and pre-charge resistors add further resistive losses during the milliseconds of inrush current at contactor close.

BMS electronics

The BMS draws continuous quiescent current for cell voltage monitoring, balancing, and communication. Passive balancing — dissipating charge as heat through bypass resistors to equalise cell voltages — can add 0.5–5 W per cell group during a balancing event, depending on balancing current and duration.

Total pack heat budget during fast charging

For a representative 77 kWh / 400V pack (192 series cells, 2 parallel) charging at 150 kW:

This 3.7–4.6 kW must be removed continuously during the charge session. A liquid cooling plate with 5 L/min coolant flow and 15°C inlet temperature can manage approximately 4–5 kW with a 10°C coolant temperature rise — which is why that flow rate is a common design target for fast-charge-capable packs.

6. The Temperature Gradient Problem Inside a Cell

Even if a pack's average temperature is perfectly controlled, individual cells develop internal temperature gradients during high-rate operation. These gradients matter because the electrochemical reactions inside the cell are not uniform — and non-uniform temperatures produce non-uniform reaction rates, which produce non-uniform current density, which produces non-uniform degradation.

Radial gradient in cylindrical cells: Heat is generated throughout the jelly roll volume but can only escape radially through the surface. The temperature difference between core and surface is approximately:

ΔT_radial ≈ Q_gen × r² / (4 × k_eff × L)

>

where Q_gen = volumetric heat generation rate (W/m³), r = cell radius (m), k_eff = effective radial thermal conductivity (W/m·K), L = cell length (m).

For a 21700 cell, k_eff in the radial direction is approximately 0.2–0.5 W/m·K — much lower than the axial direction (~20–30 W/m·K) because heat must cross the alternating electrode-separator-electrode stack layers rather than flowing along them. At 2C discharge with 10 W total heat generation, the core-to-surface temperature gradient in a 21700 is approximately 5–10°C.

The consequence of internal gradients: The hotter core experiences faster SEI growth, earlier lithium plating risk, and faster capacity fade than the cooler surface. The cell ages from the inside out. A cell that measures 30°C at the surface — within all operating limits — may be running its core at 38–45°C during aggressive fast charging, where SEI growth rate (following Arrhenius kinetics) is 2–3× faster than at 30°C.

Surface-mounted temperature sensors — which is what every automotive BMS uses — cannot see this gradient. They systematically underestimate the thermal stress the cell core is experiencing.

7. Why SOC and C-Rate Interact to Create Heat Spikes

Heat generation is not uniform across a charge or discharge curve. DCIR varies with SOC because the electrochemical activity — and therefore the resistance — of both electrodes changes as lithium intercalation state changes.

For NMC/graphite cells, DCIR is typically lowest between 30–70% SOC and rises at both extremes:

• Above 80% SOC: The cathode is approaching full delithiation (NMC) and fewer available sites means higher ionic resistance. The anode is nearly fully lithiated (LiC₆), and the staging phase transitions of graphite increase anode impedance. Heat generation spikes in the last 20% of charge.

• Below 20% SOC: The cathode is approaching full lithiation and the anode is nearly empty — both electrodes are in their high-resistance regions. Heat generation spikes in the last 20% of discharge.

• Near 0% SOC: The anode potential approaches 0 V vs Li/Li⁺ during discharge. Continued high-rate discharge in this window risks copper dissolution from the anode current collector — an irreversible degradation mechanism that creates internal short circuit risk on subsequent charging.

The practical implication for fast charging: Most DC fast chargers automatically taper current above 80% SOC — not just to prevent overcharge, but because the cell is entering its high-resistance window and heat generation per unit of energy added increases sharply. A charge curve that appears to slow at 80% is not the charger being conservative — it is managing the transition into the heat spike region. BMS implementations that allow constant-current charging through this window to satisfy a driver's "charge to 100%" preference are trading pack longevity for convenience.

8. Heat Generation During Fast Charging: The Worst-Case Scenario

DC fast charging at 150–350 kW is the highest sustained thermal stress a pack experiences in normal operation. Three factors combine to make it more demanding than high-power discharge:

Factor 1 — Current direction reversal changes entropic heat sign. During charging, the entropic term for NMC/graphite changes sign: what was exothermic during discharge becomes endothermic during charge (heat is absorbed from the environment). This sounds helpful — and at low charge rates, it genuinely reduces net heat generation. At high charge rates, Joule heating (scaling as I²) overwhelms the entropic benefit and total heat generation exceeds the discharge case at the same current magnitude.

Factor 2 — Charging always ends in the high-resistance SOC window. Unlike discharge, which typically stops before reaching 0% SOC, charging always targets the high-SOC region where cathode and anode resistance are both elevated. The Joule heating rate during the last 20% of charge is substantially higher per ampere than during the middle of the SOC window.

Factor 3 — Pack is typically cold at the start of a DC fast charge. A driver arriving at a public charger after a highway cruise has a pack that may be warm, but a driver using a charger after overnight parking has a cold pack with elevated DCIR. The first minutes of fast charging into a cold pack generate disproportionate heat and risk lithium plating at the anode, which is why thermal pre-conditioning (triggered by navigation routing to a charger) is not just a comfort feature — it is a degradation mitigation strategy.

The result is a heat generation curve during fast charging that looks roughly like:

• Minutes 0–5 from cold: peak heat generation (cold DCIR × high current)

• Minutes 5–20: moderate heat as pack warms and DCIR falls

• Minutes 20–35: rising heat again as SOC approaches 80%

• Minutes 35–45: rapid heat spike as CC-CV transition occurs and current tapers

A thermal management system sized for steady-state conditions can be overwhelmed by the cold-start spike. Pack thermal models that assume a pre-warmed cell temperature at the start of a charge event will underestimate peak coolant temperatures by 5–15°C.

9. What Happens When Heat Removal Fails: The Runaway Chain

When the thermal management system cannot remove heat as fast as the cells generate it, temperature rises. The rise is initially slow — manageable. But lithium-ion cells have a property that makes sustained temperature rise self-accelerating above a threshold: the exothermic reactions that drive thermal runaway have activation energies that are exceeded at temperatures the pack can reach.

The runaway sequence in an NMC pack proceeds in stages:

Stage 1 — SEI decomposition (80–120°C): The solid electrolyte interphase on the anode decomposes exothermically, releasing heat and producing gases (CO₂, hydrocarbons). Heat generation rate increases. Internal pressure begins to rise.

Stage 2 — Separator melting / shutdown (120–150°C): Most polyethylene separators melt at ~130°C, collapsing their pores and shutting down ionic current flow. In a well-functioning cell, this is a safety mechanism. In a cell that is already hot from external heat input (not from internal current), the shutdown occurs without the current that would have triggered it electrochemically — the separator melts without stopping an internal short that is driven by direct electrode contact from the mechanical collapse.

Stage 3 — Electrolyte decomposition and electrolyte-anode reaction (150–200°C): The organic electrolyte reacts exothermically with the lithiated graphite anode. For a fully charged cell, this reaction releases approximately 300–800 J/g of electrode material. The heat generation rate is now self-sustaining — the cell temperature is rising faster than any external cooling can remove.

Stage 4 — Cathode decomposition (170°C for NMC 811, 220°C for NMC 622, 270°C for LFP): As described in the NMC stoichiometry article, the charged cathode releases oxygen above its onset temperature. The released oxygen reacts violently with the electrolyte. For NMC 811, this stage can release 500–1,500 J/g of cathode material. Peak cell temperatures can reach 400–800°C.

Stage 5 — Propagation to adjacent cells: A single cell in runaway can transfer enough heat to neighbouring cells through direct conduction, radiation, and hot gas flow to push them past their Stage 1 onset. If adjacent cells also enter runaway, the propagation is exponential. Pack design for thermal runaway containment (not prevention) is the engineering discipline that accepts Stage 4 as inevitable in at least one cell and focuses on ensuring Stage 5 does not consume the vehicle.

The design hierarchy that follows from this: Prevention (cooling system keeps cells below 45°C continuously) → Protection (BMS opens contactors, disconnects current, triggers pre-cooling when any cell reaches 55°C) → Containment (structural barriers between cells rated to absorb and redirect the thermal and gas energy of a single-cell runaway without propagating to the next cell) → Venting (directed gas ejection path that moves combustion products away from vehicle occupants). These four layers are sequential. A system relying only on the first is not adequately designed.

10. Design Implications for Thermal Management Engineers

The physics in the preceding sections translate directly into design requirements that are often underspecified in initial pack architecture documents:

Specify heat generation at cold soak, not at nominal temperature. A pack sized for 4 kW heat rejection at 25°C cell temperature needs to handle 6–8 kW in the first minutes of fast charging from −10°C. Cold-start heat generation peaks are the sizing case for the coolant pump, not steady-state.

Model entropic heat or accept systematic error below 1C. At C/5 discharge — the rate relevant for range prediction and low-power accessory loads — the entropic term is the dominant heat source. Thermal models that ignore it will predict lower cell temperatures than actual, giving false confidence in thermal margins.

Account for contact resistance degradation over life. A new pack with 0.1 mΩ cell-to-busbar contact resistance may show 0.5 mΩ after 5 years of vibration, thermal cycling, and corrosion. The 5× increase in contact resistance adds meaningfully to pack-level heat generation and changes the spatial distribution — a hot spot at a high-resistance interconnect may trigger a thermal event that the cell temperature sensors do not predict.

Design cooling for end-of-life resistance, not beginning-of-life. DCIR increases approximately 20–40% over the service life of an automotive NMC cell as the SEI thickens and electrode particle cracking increases contact resistance. A thermal system designed for beginning-of-life DCIR will be undersized at end of life, when the cells are simultaneously generating more heat and have less thermal stability margin due to capacity loss.

Separate the question of average pack temperature from cell-to-cell temperature variance. A pack with an average cell temperature of 30°C but a 15°C spread between the hottest and coolest cell is thermally worse than a pack at an average of 35°C with a 3°C spread. The hot cells age faster, lose capacity, see higher local DCIR (increasing their heat generation relative to cooler cells), and are the first to reach runaway conditions. Temperature uniformity is a more important design target than average temperature.

References

1. Bernardi, D., Pawlikowski, E. and Newman, J. — "A general energy balance for battery systems," Journal of the Electrochemical Society, 132(1), 1985. https://doi.org/10.1149/1.2113792

2. Rao, L. and Newman, J. — "Heat-generation rate and general energy balance for insertion battery systems," Journal of the Electrochemical Society, 144(8), 1997. https://doi.org/10.1149/1.1837832

3. Thomas, K.E. and Newman, J. — "Thermal modeling of porous insertion electrodes," Journal of the Electrochemical Society, 150(2), 2003. https://doi.org/10.1149/1.1531194

4. Pesaran, A.A. — "Battery thermal models for hybrid vehicle simulations," Journal of Power Sources, 110(2), 2002. https://doi.org/10.1016/S0378-7753(02)00200-8

5. Forgez, C. et al. — "Thermal modeling of a cylindrical LiFePO4/graphite lithium-ion battery," Journal of Power Sources, 195(9), 2010. https://doi.org/10.1016/j.jpowsour.2009.10.105

6. Feng, X. et al. — "Thermal runaway mechanism of lithium-ion battery for electric vehicles: a review," Energy Storage Materials, 10, 2018. https://doi.org/10.1016/j.ensm.2017.05.013

7. Kim, G-H., Pesaran, A. and Spotnitz, R. — "A three-dimensional thermal abuse model for lithium-ion cells," Journal of Power Sources, 170(2), 2007. https://doi.org/10.1016/j.jpowsour.2007.04.018

8. Waldmann, T. et al. — "Temperature dependent ageing mechanisms in lithium-ion batteries — a post-mortem study," Journal of Power Sources, 262, 2014. https://doi.org/10.1016/j.jpowsour.2014.03.112