Electrochemical Impedance Spectroscopy (EIS) is one of the most widely used non-destructive diagnostic techniques for electrochemical devices. However, analyzing raw EIS data (typically presented as Nyquist plots) remains a significant challenge. Different battery phenomena, such as solid-state diffusion, charge transfer kinetics, and film transport, often occur at similar timescales. This creates overlapping, compressed semi-circles on a Nyquist plot, obscuring individual electrochemical mechanisms.
To overcome this bottleneck, the Distribution of Relaxation Times (DRT) method has emerged as a powerful model-free deconvolution technique. Instead of fitting arbitrary equivalent circuit models (ECMs) with highly correlated parameters, DRT mathematically transforms frequency-domain impedance spectra into a continuous time-constant ($\tau$) domain, resolving overlapping processes into distinct, well-defined peaks.
- Model-Free Deconvolution: DRT mathematically converts frequency-domain EIS spectra into a continuous relaxation time ($\tau$) domain using Tikhonov regularization, separating overlapping physical processes without arbitrary equivalent circuit assumptions.
- High-Frequency & Inductance Insights: Battery EIS does not require open calibration due to negligible stray parallel capacitance compared to low battery impedance, while cable and geometric inductance dominate at frequencies $>10\text{ kHz}$.
- Physicochemical Peak Assignment: Accurately isolates distinct resistance components across the spectrum: bulk ohmic resistance ($\tau < 10^{-5}\text{ s}$), SEI film transport ($10^{-5} - 10^{-3}\text{ s}$), charge-transfer kinetics ($10^{-3} - 10^{-1}\text{ s}$), and solid-state diffusion ($>1\text{ s}$).
1. The Puzzle of Battery EIS: Calibration Discrepancies and Geometrical Inductance
For engineers coming from the field of standard electronic devices or high-frequency circuits, battery electrochemical impedance spectroscopy (EIS) often presents several puzzling discrepancies. Understanding these differences is crucial for correctly interpreting the high-frequency limits of battery impedance.
Why is there no "Open Calibration" in Battery EIS?
In electronic testing using LCR meters or Vector Network Analyzers (VNAs), performing an Open/Short/Load (OSL) calibration (or at least an Open calibration) is a mandatory first step. Open calibration measures and cancels out the stray parallel capacitance ($C_{stray}$) and leakage conductance of the test fixture/leads. Without it, high-impedance measurements are heavily distorted by these parallel parasitics.
However, battery EIS systems (potentiostats/galvanostats) rarely present an "Open Calibration" option in their standard workflows. This is due to two main reasons:
- Low-Impedance Dominance: Lithium-ion batteries are extremely low-impedance devices, typically ranging from $0.1 \text{ m}\Omega$ to $10 \text{ m}\Omega$ (and up to a few ohms for tiny cells). The stray parallel capacitance of typical cables and fixtures is tiny, on the order of picofarads ($C_{stray} \approx 10\text{ pF}$). At $100\text{ kHz}$, this corresponds to an impedance of $Z_{stray} \approx 160\text{ k}\Omega$. Since $Z_{stray}$ is in parallel with the battery, its contribution is mathematically negligible: $$\frac{Z_{cell} \cdot Z_{stray}}{Z_{cell} + Z_{stray}} \approx Z_{cell}$$ Thus, correcting for parallel stray capacitance yields virtually zero change in the measured battery impedance. Instead, the real bottleneck is series lead inductance ($1\text{ }\mu\text{H}$ at $100\text{ kHz}$ equals $0.628\text{ }\Omega$) and contact resistance, making Short calibration and four-terminal Kelvin sensing the only critical compensation steps.
- Active Cell vs. Passive Component Calibration: Standard LCR meter calibration assumes passive, unpowered devices under test. A lithium-ion battery, however, is an active electrochemical energy source with a continuous cell voltage. Short-circuiting the terminals with a live battery in place is extremely dangerous and would draw massive short-circuit currents. Therefore, calibrating the measurement path with the cell connected is not possible. Calibration in electrochemical systems is therefore performed on the test leads and fixtures themselves *before* connecting the cell (i.e., with terminals open or shorted to cancel cable parasitics).
Cell Geometry and the High-Frequency X-Intercept ($R_{ohmic}$) Illusion
In typical equivalent circuit model (ECM) fitting, the intercept where the Nyquist plot crosses the real axis ($Z'' = 0$) at high frequencies is taken as the pure bulk ohmic resistance ($R_{ohmic}$ or $R_{\infty}$), representing the electrolyte and current collector resistances. However, cell geometry can make this X-intercept a deceptive simplification:
- Cylindrical and Wound Cells: In spirally wound cells (e.g., 18650, 4680, or wound prismatic cells), the long metallic current collector foils rolled together act internally like a coil, introducing significant internal geometric inductance ($L_{int}$) and inducing eddy currents.
- Tab Layout and Skin Effect: Studies (such as those by the Technical University of Munich and other research groups) have shown that at high frequencies ($>100\text{ Hz}$ to $10\text{ kHz}$), the current distribution along the jellyroll becomes highly non-uniform due to electromagnetic induction and skin effects. The AC current decays rapidly as it travels along the collector foil away from the tab. Because of this frequency-dependent current distribution, the high-frequency real part of the impedance is distorted and deviates from the true DC bulk ohmic resistance.
Consequently, simply reading the X-intercept as the pure electrolyte resistance, or fitting the high-frequency "inductive hook" with a constant series inductor $L$, introduces errors. When analyzing wound cells, researchers must account for the cell's physical geometry and tab design—such as single-tab vs. multi-tab vs. tabless configurations—to avoid misinterpreting inductive artifacts as bulk electrolyte properties.
2. Principles and Mathematical Formulation of DRT
At the heart of the DRT method lies the assumption that the battery impedance $Z(\omega)$ (excluding inductance and bulk ohmic resistance $R_{\infty}$) can be represented by an infinite series of infinitesimal parallel $RC$ elements connected in series. The overall complex impedance is expressed via the Fredholm integral equation of the first kind:
$$Z(\omega) = R_{\infty} + \int_{0}^{\infty} \frac{g(\tau)}{1 + j\omega\tau} d\tau$$
Where $g(\tau)$ represents the distribution function of relaxation times. In practice, because relaxation times typically span multiple orders of magnitude, we formulate the integral on a logarithmic scale ($\ln\tau$):
$$Z(\omega) = R_{\infty} + \int_{-\infty}^{\infty} \frac{\gamma(\ln\tau)}{1 + j\omega\tau} d(\ln\tau)$$
Calculating the distribution $\gamma(\ln\tau)$ from measured impedance $Z(\omega)$ is an inverse problem. Because numerical inversion is mathematically "ill-posed" (extremely sensitive to measurement noise), standard inversion algorithms easily produce highly oscillatory, non-physical results. To obtain a stable solution, Tikhonov Regularization is applied:
$$\min_{\mathbf{g}} \left\{ \|\mathbf{A}\mathbf{g} - \mathbf{Z}_{meas}\|^2 + \lambda \|\mathbf{L}\mathbf{g}\|^2 \right\}$$
Here, $\mathbf{A}$ is the discretization matrix of the integral kernel, $\mathbf{g}$ is the vector representing the distribution function, $\mathbf{Z}_{meas}$ represents experimental data, $\mathbf{L}$ is a regularization operator (often the identity matrix or a second-order derivative matrix for smoothing), and $\lambda$ is the regularization parameter.
3. EIS Measurement Method and Environment
A high-quality, artifact-free DRT analysis starts with meticulous EIS measurements. Because DRT is highly sensitive to noise and systematic drift, the following measurement protocols must be strictly followed:
- AC Excitation Amplitude: To guarantee that the electrochemical system operates in a linear regime, a small amplitude is critical. For batteries, a voltage perturbation of **5 to 10 mV** (peak-to-peak) is standard. Alternatively, galvanostatic EIS (GEIS) can be used, with the current amplitude carefully chosen to yield a voltage response within 10 mV.
- Frequency Range: A broad frequency sweep is necessary. Typically, the range spans from **10 mHz to 100 kHz** (or up to 1 MHz for inductive studies). Broad sweeps ensure that slow solid-state diffusion (low frequency) and fast contact resistances (high frequency) are both fully captured.
- State of Charge (SOC) Stabilization: Batteries must reach electrochemical equilibrium at each SOC point before initiating EIS. A rest (relaxation) time of **2 to 4 hours** (or until the open-circuit voltage drift is less than 1 mV/hour) is essential. Performing EIS on a drifting cell violates the stationarity requirement, generating severe artifacts in low-frequency DRT data.
- Temperature Control: Reaction kinetics and diffusivity scale exponentially with temperature (Arrhenius behavior). Measurements must be performed inside a high-precision thermal chamber (e.g., stable to within $\pm 0.1^\circ\text{C}$).
- Four-Terminal (Kelvin) Connection: Eliminating test lead and contact resistance is vital, especially for low-impedance high-energy cells where contact resistance can overshadow internal reaction parameters.
4. Peak Attributes and Broadening in DRT Spectra
Once the regularization equation is solved, the resulting DRT spectrum (plotting $\gamma(\ln\tau)$ against $\tau$ or frequency $f$) reveals several distinct peaks. Understanding what these peaks represent is key to interpreting the cell state:
Peak Position ($\tau$) and Area ($R$)
- Horizontal Axis ($\tau$): The peak location defines the characteristic timescale of the physical process. Alternatively, it can be viewed as characteristic frequency $f_c = 1/(2\pi\tau)$.
- Peak Area: The integral (area) under a peak directly corresponds to the polarization resistance ($R_p$) associated with that specific process. As the cell degrades, the growth of specific peak areas reveals which component is deteriorating.
Why do DRT peaks exhibit broadening?
In an ideal case with a single, perfectly uniform $RC$ element, the DRT would show an infinitely sharp Dirac delta function. In reality, DRT peaks are always broadened due to two main factors:
- Physical Heterogeneity: Electrodes are composed of billions of active particles with varying size distributions. The porous network introduces non-uniform tortuosity, and local reaction rates vary spatially. This physical dispersion creates a continuous distribution of time constants, naturally broadening the peak.
- Mathematical Regularization (Smoothing): The Tikhonov regularization parameter $\lambda$ directly controls the smoothness of the solved curve. If $\lambda$ is too high (over-smoothing), numerical noise is suppressed but peaks broaden and merge, lowering resolution. If $\lambda$ is too low (under-smoothing), peaks become very sharp but the spectrum is corrupted by oscillatory noise artifacts. Selecting the optimal $\lambda$ (e.g., using the L-Curve method) is essential.
5. Frequency Components and Physical Phenomenon Assignment
While the exact timescales vary depending on chemistry (e.g., LFP, NMC, Graphite, Silicon), temperature, and design, the literature generally maps the resolved time constants to the following physical mechanisms:
| Relaxation Time ($\tau$) | Frequency Range ($f$) | Assigned Physical Process | Key Characteristics & Dependencies |
|---|---|---|---|
| $\tau < 10^{-5}\text{ s}$ | $f > 15.9\text{ kHz}$ | Ohmic Resistance & Inductance | Bulk electrolyte resistance, current collector foil resistance, tab connections, and high-frequency inductive wiring effects. Typically SOC-independent. |
| $\tau = 10^{-5}\text{ to } 10^{-3}\text{ s}$ | $159\text{ Hz } \le f \le 15.9\text{ kHz}$ | SEI / CEI Layer Passivation | Lithium-ion transport through the Solid Electrolyte Interphase (SEI) on the anode or Cathode Electrolyte Interphase (CEI). Shows minimal SOC dependence but grows significantly during high-temperature storage or cycling. |
| $\tau = 10^{-3}\text{ to } 10^{-1}\text{ s}$ | $1.59\text{ Hz } \le f < 159\text{ Hz}$ | Charge Transfer Kinetics | Charge transfer reactions at the anode and cathode interfaces. Strongly dependent on temperature (Arrhenius relation) and displays non-linear variations with SOC. |
| $\tau > 10^{-1}\text{ s}$ | $f < 1.59\text{ Hz}$ | Solid-state & Liquid-phase Diffusion | Lithium diffusion within the active material bulk (solid-state) and transport in the electrolyte pores (liquid-phase). Highly sensitive to particle size and localized concentration. Note: Warburg-type diffusion impedance does not conform to the parallel RC element assumption underlying DRT, so its spectral representation appears as a characteristic broad distribution rather than a simple peak. |
6. Key Pitfalls and Precautions in DRT Interpretation
While DRT is exceptionally powerful, naive interpretation can easily lead to erroneous scientific conclusions. Researchers must keep the following constraints in mind:
- The Two-Electrode Superimposition Limit: In a standard commercial battery cell
(2-terminal configuration), the measured impedance is the sum of both the anode and cathode.
Because anode and cathode charge-transfer processes often occur at similar timescales, their DRT
peaks frequently overlap into a single, merged peak. Without reference electrode (3-electrode)
testing or physical simulation templates, assigning a merged peak to one specific electrode is
highly speculative.
How researchers address this separation:- Three-Electrode Cells: Introducing a reference electrode (e.g., lithium wire) allows the measurement of WE vs. RE and CE vs. RE separately, providing independent DRT spectra for individual electrodes.
- Symmetrical Cell Method: Harvesting electrodes from identical cycled/fresh cells to build symmetrical cells (e.g., cathode-cathode or anode-anode) to isolate single-electrode impedance.
- Temperature and SOC Sensitivity Analysis: Evaluating how peaks shift over a matrix of temperatures and SOCs. Since cathode and anode charge-transfer processes have distinct activation energies and lithium-ion concentration dependencies, their peak areas and time constants follow different trajectories.
- Model-Based Decoupling: Combining experimental DRT curves with physics-based models (such as P2D/DFN models) to compare experimentally observed peaks with physically simulated peaks.
- Regularization Artifacts (Fake Peaks): A small change in the regularization parameter $\lambda$ can make a peak split into two, or make two peaks merge into one. If the raw data contains high-frequency noise, a low $\lambda$ will interpret this noise as physical processes, producing spurious peaks at low time constants.
- Violations of Stationarity: If the cell is discharging, charging, or simply resting but still chemically relaxing (drifting voltage) during the EIS sweep, the low-frequency data becomes distorted. DRT inversion will translate this time-dependent drift into false peak features at large time constants ($\tau > 10\text{ s}$).
- High-Frequency Inductive Distortion: High-frequency inductance from long cables can bleed into the real part of the impedance spectrum if not mathematically corrected. This creates false negative or positive oscillations in the DRT at very short time constants.
- Kramers-Kronig (KK) Consistency Check: Before running DRT, the raw impedance data must be verified using a KK test (e.g., using Lin-KK tools) to confirm the data complies with linearity, causality, and stability. Inconsistent data will inevitably result in mathematically valid but physically nonsensical DRT spectra.
7. Excel DRT: Democratizing DRT Analysis
Traditionally, performing DRT required installing complex, specialized programming environments in Python or MATLAB. To make this methodology accessible to material scientists and bench engineers, we developed the **Excel DRT Tool**.
By implementing Tikhonov regularization and L-curve parameter selection algorithms directly inside an optimized Excel VBA framework, the tool allows users to import raw EIS text files and compute high-resolution DRT curves immediately within a familiar spreadsheet environment. This lowers the barrier to entry, enabling rapid diagnostic workflows right next to the battery cycler.
Explore the Excel DRT Analysis Tool
Process your EIS impedance data and extract distribution of relaxation times in a few clicks with Excel VBA.
Go to Excel DRT Tool →8. References
- Schichlein, H., Müller, A. C., Voigts, M., Krügel, A., & Ivers-Tiffée, E. (2002). Deconvolution of electrochemical impedance spectra for the identification of electrode processes. Journal of Applied Electrochemistry, 32, 875–882. DOI: 10.1023/A:1020599525160
- Wan, T. H., Saccoccio, M., Chen, C., & Ciucci, F. (2015). Influence of the discretization methods on the distribution of relaxation times deconvolution: implementing radial basis functions with DRTtools. Electrochimica Acta, 184, 483–499. DOI: 10.1016/j.electacta.2015.09.097
- Illig, J., Ender, M., Weber, A., & Ivers-Tiffée, E. (2015). Modeling graphite anodes with serial and transmission line models. Journal of Power Sources, 282, 335–347. DOI: 10.1016/j.jpowsour.2015.02.038