Improving battery health insights with optimized EIS measurements

Whether hunting for the perfect electrolyte or diagnosing a failing cell, electrochemical impedance spectroscopy (EIS) is the most powerful—and often most frustrating—diagnostic tool available to battery researchers. This blog post discusses the application of EIS to batteries, the common errors that can affect the validity of the data, and the best practices to adopt to optimize measurements for cleaner, more reliable data.

This article covers the following topics:

• EIS as a diagnostic tool
• EIS: Pitfalls, issues, and errors
• Experimental setup: Temperature fluctuations, contact resistance, and inductive loops
• Modelling errors: Overfitting, K–K test

EIS as a diagnostic tool

It is useful to review the primary insights gained from EIS studies in battery research, which generally fall into a few key categories [1].

EIS: Pitfalls, issues, and errors

In EIS, errors can be introduced in several ways and at different points in the workflow. Errors can arise even before the measurement begins, for example, if the wrong parameters are selected, if the wrong conditions are not accounted for in the experimental setup, or after the fact in the data analysis if physical realities are not considered. Luckily, there are best practices that can be adopted to avoid these errors becoming so large that confidence is lost in the measured data.

The Big Three: Linearity, stability, and causality  

For EIS data measured on batteries (or any other system) to be valid, the system must meet three measurement criteria. If these are violated, confidence in the measured data may be questioned.

Linearity

The excitation signal (current or voltage) is a compromise. It must be small enough (typically <10 mA) so the response is linear, but also large enough to achieve a decent signal-to-noise ratio. If the signal is too large, nonlinear harmonics are measured rather than true impedance. In the context of batteries, the best practice for galvanostatic EIS is to usually aim for an amplitude which corresponds to a current of C/10. 

Stability

This comes from the requirement that the system should not change during the measurement. For batteries, this means that the state of charge should not change during the measurement. Therefore, do not make an EIS measurement that could take up to one hour on a battery that is actively discharging or charging, as doing so leads to artifacts that often appear as a distorted tail at low frequencies. Most EIS measurements on batteries are therefore done following a rest (>30 minutes) after the battery has been charged or discharged, and usually also with a 0 V / 0 A DC bias. There are exceptions, like for operando EIS, but this is an advanced technique which requires some extra data analysis and will be covered in a separate blog post.

Causality

The final criteria is that the response must be caused solely by the excitation signal – not by external noise or internal chemical fluctuations caused by, e.g., the battery cooling down after a charging step is completed. Therefore, the final criteria relates to the previous one (stability), in that sufficient time should be taken before the measurement commences to allow the battery to reach a steady state.

Experimental setup: Temperature fluctuations, contact resistance, and inductive loops

Up to this point, the focus has been on the universal criteria required to obtain high-quality, valid EIS data. Here we address the errors that arise from the experimental setup that are particularly noticeable when specifically dealing with batteries.

Contact resistance

It is essential to consider the impact of contact resistance between the battery terminals and the measurement leads. Poor or variable contact can introduce additional resistance that distorts the true impedance profile of the battery. It is best practice to use a specially designed battery holder for highly accurate EIS measurements. This ensures the contact is reproducible and can be accounted for by calibration, if necessary. This effect becomes especially significant when using a two‑wire measurement setup, where the same pair of leads is used to both supply current and measure voltage. In this configuration, the resistance of the leads, connectors, and contact interfaces is added directly to the measured impedance, potentially obscuring the real internal resistance of the battery.

To avoid this error, a four‑wire (Kelvin) connection should be used whenever accurate impedance or low-resistance measurements are required. In a Kelvin setup, one pair of leads carries the current while a separate pair senses the voltage directly at the battery terminals. Because the sensing leads draw negligible current, their resistance contribution is effectively eliminated, ensuring that the measured impedance reflects the battery itself rather than the measurement hardware. Another advantage of using specially designed holders is that they often come with four terminals for exactly this purpose. If using a holder is not possible, for example with a prismatic battery, then four-contact measurement should still be made, according to Figure 4. 

Inductive loops

At high frequencies, impedance spectra often exhibit a «curl» that drifts into the positive imaginary quadrant of the Nyquist plot. This behavior is almost never related to electrochemical phenomena and is a classic sign of parasitic inductance in the measurement setup.

Stray inductance is commonly introduced by long cable runs, untwisted or poorly routed leads, or measurement geometries that create unintended loops. Even a few microhenries (μH) of inductance can dominate the impedance response at higher frequencies. To minimize these artifacts, the wiring should be kept as short as practicable; if possible, the current- (WE+CE) and voltage-sensing (RE+S) cables can also be twisted tightly together into pairs and arranged to avoid forming large loops. Using proper cable management and maintaining a compact, low-inductance measurement geometry helps ensure that the high-frequency region reflects the true behavior of the cell rather than limitations of the test setup. Although, because of the internal geometry of the battery, it is not usually possible to remove this effect entirely.

Modelling errors: Overfitting, K–K test

The preceding sections covered the workflow up until the measurement is made. This final section presents a few things which can be done to avoid post-measurement mistakes during the analysis of the collected data.

Overfitting

When fitting impedance data, it is tempting to keep adding additional resistor–capacitor (RC) or constant phase elements to the equivalent circuit model (ECM) until the residuals shrink and the fit appears nearly perfect. However, this approach can be misleading. Every element in an ECM must correspond to a real, physically meaningful process in the electrochemical system. Adding arbitrary components simply to improve statistical fit can produce models that look impressive but lack any connection to actual charge‑transfer pathways, diffusion behavior, or interfacial processes.

An over‑parameterized circuit often masks underlying issues such as noisy data, inductive artifacts, or measurement errors. It can also lead to incorrect assumptions by assigning an RC circuit to something that does not exist. A good ECM should be as simple as possible and grounded in physical principles, reflecting what is known about the cell chemistry, geometry, and operating conditions.

Ignoring the Kramers–Kronig (K–K) transform

The Kramers–Kronig (K–K) test is a crucial—yet frequently overlooked—tool for validating impedance data. Mathematically, the K–K relations are integral equations that connect the real and imaginary parts of a complex function (the impedance in this case). If the data was measured according to the three criteria above (i.e., linearity, stability, and causality) then the real part of the impedance is predictable from the imaginary part and vice-versa. The K–K test works by calculating what the theoretical imaginary part of the impedance should be, based on the measured real data. If they do not match, then the data is not valid.   

Most modern EIS software, e.g., NOVA, has a built-in K–K tool. Behind the scenes, these work by generating a pseudo-circuit made of a series of RC circuits (usually as many as the number of frequencies measured), since this circuit is inherently K–K-compliant. The software will then compare the experimentally measured data points to the K–K-compliant model and generate a residual error plot. 

The residual plot should be randomly distributed around zero, across the entire measured frequency range—typically within at 1% to 2% range. If distinct patterns emerge, like waves or spikes (>5%), then the data is considered to have failed the K–K test. 

If a dataset fails the K–K test, it is a strong indication that the system was nonstationary (changing during the measurement), nonlinear (excited too strongly), or corrupted by instrumentation errors, drift, or poor contact. Care should therefore be taken with any conclusions made from such data. 

Conclusion

Electrochemical impedance spectroscopy remains one of the most important diagnostic tools in battery research, but its value depends entirely on the quality and interpretability of the data collected. The best practice is to always ensure that the fundamental requirements of linearity, stability, and causality are met, while also correcting and controlling the other common errors in experimental setup—temperature drift, poor contact, or inductive wiring.

In short: cleaner measurements lead to clearer conclusions.

References

[1] Meddings, N.; Heinrich, M.; Overney, F.; et al. Application of Electrochemical Impedance Spectroscopy to Commercial Li-Ion Cells: A Review. Journal of Power Sources 2020, I, 228742. DOI:10.1016/j.jpowsour.2020.228742

[2] Choi, W.; Shin, H.-C.; Kim, J. M.; et al. Modeling and Applications of Electrochemical Impedance Spectroscopy (EIS) for Lithium-Ion Batteries. Journal of Electrochemical Science and Technology 2020, 11 (1), 1–13. DOI:10.33961/jecst.2019.00528

Additional resources 

Blog post: High-frequency EIS: a powerful tool for the future of mobility

Blog post: Solid-state batteries: A promising technology thriving under pressure