Electrochemical impedance spectroscopy is a standard laboratory technique for characterizing battery cells. It measures the cell's complex impedance across a range of frequencies, typically from 10 kHz down to a few millihertz, by applying a small AC perturbation and measuring the response. The resulting Nyquist plot or Bode plot reveals the contributions of different physical processes: ohmic resistance, charge transfer kinetics at the electrode-electrolyte interfaces, solid-state diffusion in electrode particles, and double-layer capacitance effects.
Most BMS firmware treats EIS data, if it uses it at all, as a tool for internal resistance estimation. Take the value at a specific frequency (often 1 kHz as a proxy for ohmic resistance), track how it changes over time, infer state of health. That is a useful but narrow application. The full impedance spectrum carries significantly more information, including thermal state information that can be used to build a cell-specific thermal fingerprint: a characteristic signature of how that cell's impedance varies with temperature.
How temperature shows up in the impedance spectrum
Temperature affects every major physical process that contributes to a cell's impedance spectrum, and it does so with characteristic signatures at different frequency ranges.
At high frequencies (above 1 kHz), the primary observable is the pure ohmic resistance of the electrolyte and contact resistances. Electrolyte ionic conductivity has a strong temperature dependence: it increases with temperature following an Arrhenius-type relationship. In a 1M LiPF6 electrolyte, conductivity roughly doubles from 0 to 40 degrees Celsius. This means the high-frequency intercept of the Nyquist plot (the real-axis x-intercept) shifts measurably with temperature, typically by 1 to 2 milliohms per degree Celsius for an 18650 cell. This shift is a well-established temperature proxy and is the basis for so-called high-frequency resistance thermometry in battery characterization.
At intermediate frequencies (10 Hz to 1 kHz), the semicircle that appears in the Nyquist plot corresponds to charge transfer resistance at the electrode interfaces combined with double-layer capacitance effects. The charge transfer resistance is strongly temperature-dependent, decreasing as temperature increases because the Butler-Volmer exchange current density at the electrode surface follows Arrhenius kinetics. The radius of this semicircle therefore encodes temperature information. In a fresh cell, the charge transfer resistance at 10 degrees Celsius is typically two to three times its value at 40 degrees Celsius.
At low frequencies (below 1 Hz), the Warburg element dominates, representing solid-state lithium diffusion in the electrode particles. Diffusion coefficients in electrode materials are strongly temperature-dependent: lithium diffusivity in graphite increases by roughly an order of magnitude from 0 to 60 degrees Celsius. The slope of the low-frequency Warburg tail in the Nyquist plot shifts with temperature.
Extracting the thermal fingerprint
A cell thermal fingerprint is a parameterized model that maps the relationship between measured impedance features and actual cell temperature. The practical process for building one involves three steps: characterization measurements, equivalent circuit fitting, and fingerprint parameterization.
The characterization measurements are taken at a set of known temperatures, typically covering the range from 10 to 55 degrees Celsius in steps of 5 to 10 degrees. For each temperature setpoint, you measure a full EIS spectrum after the cell has equilibrated thermally. It is important to control SoC during this measurement because many impedance features have SoC dependence as well as temperature dependence, and the two effects need to be disentangled.
Equivalent circuit fitting involves applying a model circuit (typically an Randles circuit or a modified version with additional elements for the anode and cathode interfaces separately) and extracting the element values at each temperature. This gives you a set of (temperature, element value) pairs for each circuit element. The high-frequency resistance, the charge transfer resistance, and the Warburg coefficient are the most temperature-sensitive and the most useful for thermal state estimation.
Fingerprint parameterization fits an Arrhenius or empirical functional form to each (temperature, element value) series. This gives you an equation that predicts element values from temperature, or equivalently, that infers temperature from measured element values. The high-frequency resistance and charge transfer resistance together are typically sufficient to constrain temperature to within 2 to 4 degrees Celsius under typical operating conditions.
Using the fingerprint in thermal modeling
The thermal fingerprint is useful for initializing and calibrating a thermal model of the cell rather than as a standalone temperature sensor. Here is the specific way we use it: when a cell begins a charge session, we have a thermistor reading of surface temperature and an EIS-derived estimate of the average internal impedance state. The EIS measurement gives us an independent temperature estimate that we can compare to the surface thermistor reading.
If the EIS-derived temperature and the surface thermistor agree within measurement uncertainty, we can initialize the thermal model with high confidence that the surface temperature and internal temperature are equilibrated (the cell has been at rest long enough to reach thermal equilibrium with its environment). The model is then reliable from the start of the charge session.
If the EIS-derived temperature is significantly higher than the surface thermistor, it indicates that the cell's interior is still warm from a prior discharge event, even though the surface has begun to cool. In this case, initializing the thermal model with the surface temperature would underestimate the starting core temperature, and the model would be optimistic about the available thermal headroom early in the charge session. Using the EIS-derived internal temperature estimate for model initialization corrects this.
Practical limitations and what they mean for implementation
EIS as a production tool for BMS thermal state estimation has limitations that are worth being clear about.
Full broadband EIS (10 kHz to millihertz) requires a relatively long measurement time, typically 5 to 10 minutes for a complete spectrum down to the low-frequency region. This is not practical as a continuous real-time measurement during charging. What is practical is a narrowband high-frequency measurement (10 Hz to 10 kHz) that takes less than 30 seconds and captures the ohmic resistance and charge transfer semicircle without the low-frequency Warburg region. The thermal information in the high-frequency region is sufficient for initialization purposes.
Some embedded BMS hardware supports single-frequency impedance measurement but not broadband EIS. A single-frequency measurement at 1 kHz gives the ohmic resistance, which is the least ambiguous temperature proxy (though it also encodes contact resistance and SEI thickness, so it needs a reference baseline). Single-frequency at 10 to 100 Hz gives a point on the charge transfer semicircle that encodes temperature along with SoC and cell age. Disentangling these effects requires the fingerprint model that maps the measurement space to the temperature estimate while accounting for the known SoC from the BMS state estimation.
The fingerprint is also cell-specific. A fingerprint built from one cell lot may not transfer accurately to a different lot or a different cell chemistry, even within the same nominal product line. For production use, the fingerprint should be built from cells sampled from the same lot as the production batch, or the model should include uncertainty bounds that account for expected lot-to-lot variation in cell impedance parameters. This is extra characterization work, but it is one-time work per cell type rather than ongoing.
The payoff is a cell thermal characterization capability that goes beyond what a thermistor alone can provide: an evidence-based understanding of the cell's internal thermal state derived from its own electrochemical signature, which is the physical basis for everything we do in building thermally-informed charge protocols.