Embedded EIS in EV Battery Management: What It Means and What Has Actually Been Proven

Traction-pack use cases, measured performance, production evidence and validation roadmap. Evidence current to 2 September 2026.

Electrochemical impedance spectroscopy (EIS) applies a small current variation and measures the battery voltage response across selected frequencies. In plain language, it gives the battery management system another way to look inside the cell. For an EV owner, the relevant question is whether that extra information improves charging, range, safety, service, or battery life. For an engineer, the harder question is whether the signal remains accurate across pack topology, temperature, aging, power-electronics noise, and real vehicle operation.

This report addresses both questions. It covers BEV, PHEV and HEV traction batteries: measurement opportunities, closed-loop control, safety diagnostics, service and retired-pack routing. Consumer electronics are excluded from the production evidence.

Report position	Finding
Production	One named EV integration: XPENG with Analog Devices; public field-performance data are not disclosed.
Best quantified evidence	Cell and module studies show temperature, SOH, plating and abuse-warning value; transfer to full vehicles and fleets remains the main gap.
Recommended role	A confidence-gated diagnostic layer that augments, but never replaces, conventional voltage, current, temperature and isolation protection.

Evidence note: Supplier statements are used only for product capability and deployment claims. Research performance is taken from peer-reviewed or government sources and is reported with scale, protocol and limitations.

Executive verdict

Verdict: Embedded EIS is technically credible for EVs, but its public evidence is asymmetric: measurement hardware is advancing faster than disclosed vehicle-level diagnostic performance.

The strongest near-term EV case is not a continuous laboratory-style sweep. It is a short, scheduled or event-triggered measurement that changes a bounded vehicle decision: reduce fast-charge current, refine core-temperature estimation, update power limits, confirm an anomaly, set a diagnostic trouble code, or route a retired module.

As of this review, ADI states that XPENG is the first automotive OEM to adopt its EIS architecture for mass-production EV batteries [25]. This is meaningful E4 commercialization evidence. It is not E5 performance evidence: no public source identifies the vehicle population, impedance error distribution, estimator accuracy, false alarms, intervention success, added charging throughput, cycle-life benefit or warranty reduction.

The public research record is strongest at E1-E3. Examples include 0.6 deg C mean absolute error for an instrumented LFP cell under an HEV-like current cycle [4]; 1.974 percentage-point SOH RMS error on 13 retired Nissan Leaf modules [8]; a lithium-plating detection limit below 0.6% of graphite capacity in controlled cells [11]; and 22.5 to 29.2 minutes of warning in specific thermal-abuse tests [16,17]. None of those numbers can be copied directly into a production safety case without topology, noise, prevalence and closed-loop validation.

The practical recommendation is a dual-path BMS: conventional V/I/T/isolation protection remains authoritative; EIS adds a quality-scored diagnostic estimate. Low confidence, excessive state drift, EMI, missing excitation or an out-of-domain spectrum must cause fallback to the conventional estimator and conservative limits.

EV priority map

Rank	Vehicle decision	EIS contribution	Possible action	Public evidence
1	Cold / high-power fast charge	Detect plating risk and core temperature	Warm pack; reduce current; change charge stage	High value; E2-E3 evidence; closed-loop vehicle proof limited
2	Thermal management and SOP	Correct surface-temperature lag; update resistance	Chiller/pump request; charge/discharge derate	Good cell evidence; stringent micro-ohm and EMI requirements
3	SOH, range and warranty	Update capacity/resistance health with uncertainty	Range confidence; power limits; service / warranty triage	Cell and retired-module evidence; fleet labels remain weak
4	Safety anomaly / post-crash	Corroborate soft short, abnormal aging or heating	Isolate, quarantine, service, tow / storage instructions	Promising abuse and lab-pack results; false-alarm evidence absent
5	Manufacturing and retired-pack routing	Compare cells/modules at a standardized state	Hold/rework; reuse class; recycle	Controlled test windows make this the lowest deployment risk

1. Scope, definitions and evidence method

1.1 What this report includes

The core scope is traction batteries in battery-electric, plug-in-hybrid and hybrid vehicles. Evidence qualifies as EV-specific when it uses an automotive cell or module, an EV/HEV duty cycle, vehicle power electronics or CAN data, a named vehicle, or an automotive BMS platform. Large stationary modules are included only when the engineering result transfers to EV packs and are visibly labeled as transfer evidence.

Consumer products are excluded from the production landscape. In particular, Qnovo's LG V30 smartphone claim is not used as EV evidence. Its public description does not establish conventional swept complex EIS, traction-pack behavior, automotive safety integration or field performance.

The review distinguishes three objects that are often conflated: (1) an IC that can synchronize voltage and current or compute a DFT, (2) a system that can generate and measure impedance in a pack, and (3) a validated diagnostic or control function that improves an EV outcome.

1.2 Evidence ladder

EV EIS evidence ladder

Figure 1. Evidence ladder used in this report. Commercial availability and diagnostic performance are graded separately.

Grade	Minimum meaning
E1 - cell feasibility	Controlled cell or chemistry; feature correlation or classifier.
E2 - replicated cells	Multiple cells/lots/aging paths; held-out validation; uncertainty or repeatability.
E3 - module, pack or vehicle prototype	Series/parallel topology, common mode, parasitics, embedded excitation, EV duty or abuse test.
E4 - named EV production	Identified OEM and mass-production traction-battery integration. Does not itself prove accuracy or benefit.
E5 - field outcome	Population and operating exposure; sensitivity/specificity; false alarms; intervention and economic outcomes.

Rule: A production IC is not a production EIS use case. A named OEM integration is not proof of field performance. A high-accuracy cell model is not proof of pack generalization.

1.3 Performance is a chain, not one accuracy number

Layer	What must be reported
Metrology	Re/Im or magnitude/phase error versus a reference; repeatability; drift; channel mismatch; de-embedding.
Estimation	MAE/RMSE/bias and uncertainty for core temperature, SOC, SOH, capacity, RUL or SOP.
Detection	Sensitivity, specificity, precision, detection limit, lead time and false alarms per vehicle-hour.
Timing and energy	Acquisition/compute latency, valid spectrum rate, injected Ah/Wh, charge-time impact and thermal load.
Generalization	Held-out lots, chemistry/format, SOC, temperature, aging path, vehicle, charger and EMI environment.
EV outcome	Charge time, accepted power, range accuracy, avoided derates/failures, warranty cost, service time or recovered value.

2. Where EIS fits in an EV

2.1 Measurement windows across the vehicle day

EIS assumes a sufficiently small perturbation, approximate stationarity during the observation window and known timing between current and voltage. A traction battery rarely offers all three continuously. The BMS therefore needs an opportunity scheduler, not just an EIS command.

Window	Battery state	Practical method	Best use	Constraint
Parked / overnight	Rested or slowly drifting	Selected points or fuller sweep	SOH baseline; core T; contact trend	Best repeatability; low-frequency time cost; sleep-current budget
AC charging	Controllable, moderate current	Charger/DC-link sine, multisine or pulse	SOH update; thermal estimate; model ID	Charger loop and harmonics; SOC drift
DC fast charging	High current, rapidly changing SOC/T	Rapid narrowband / operando feature	Plating risk; core T; adaptive charge	High value and highest nonstationarity; measurement must not add material time
Driving / regeneration	Natural broadband current	Passive system identification	ECM update; trend; temperature	Spectral gaps; inverter/motor-control noise; road load
Key-on / service	Controlled diagnostic window	BJB or local excitation	Connection fault; weak module; SOH	Can standardize SOC/T; requires customer-time budget
Crash / thermal event	Time critical, nonstationary	Fast selected-frequency corroboration	Short / runaway confirmation	Must run in parallel with V/T/gas/isolation; never delay contactors

A 2026 analysis of seven series-production vehicles covering 169,760 km did not measure onboard EIS accuracy; it evaluated whether usable idle windows exist. It estimated an average 421 EIS opportunities per 10,000 km at 1 percentage-point SOH resolution, versus 81 opportunities for DVA [10]. This supports scheduling feasibility, not diagnostic validity.

2.2 The end-to-end decision chain

EV EIS measurement-to-action decision chain

Figure 2. An EV EIS function is complete only when a validated measurement changes a bounded vehicle action.

1. Acquire synchronized cell voltage and pack current while recording operating point, contactor state, charger/inverter state and excitation command.

2. Reject distorted or nonstationary records using amplitude limits, coherence/residual tests, drift checks, timing diagnostics and calibrated valid-domain rules.

3. Extract physics-conditioned features or fit a model. Temperature, SOC and aging must be separated; they move overlapping parts of the spectrum.

4. Fuse the estimate with conventional BMS states and attach uncertainty. Compare with a cell/pack baseline, not a universal threshold unless transfer has been proven.

5. Apply a bounded action and record the outcome. The BMS must know what changed because of EIS and whether that change was beneficial.

2.3 Minimum signal contract

Interface	Required content
Inputs	Per-cell voltage samples; pack current/shunt samples; timestamps; SOC/T estimates; excitation command; vehicle mode.
Quality outputs	Valid/invalid; SNR/coherence; harmonic distortion; state drift; saturation; time-sync error; calibration age.
Diagnostic outputs	Estimate/probability, uncertainty, valid domain, baseline deviation, affected cell/module and persistence.
Action interface	Requested charge/power/thermal limit, DTC, contactor or service recommendation, reason and expiry time.
Trace record	Raw or compressed V/I window, spectrum/features, software/calibration versions and resulting vehicle action.

3. EV use cases: how impedance becomes action

Design Principle: Start with the vehicle decision and its acceptable error. Then choose the smallest frequency set and shortest window that improves that decision over the V/I/T baseline.

3.1 Plating-aware fast charging

Why it matters. Lithium plating becomes more likely at low temperature, high charge rate and high SOC. A voltage-only limit is deliberately conservative because the negative-electrode potential is not measured. Operando impedance can add a feature correlated with plating onset [11-14].

How to use it. During a controllable charging stage, obtain a short baseline-conditioned impedance feature - for example the reported abnormal decrease around 1 Hz in multi-stage constant-current charging [13]. Correct for SOC, temperature, age, prior current and relaxation. Fuse the feature with a thermal/electrochemical model and produce a calibrated plating probability or margin.

Vehicle action. If risk rises with adequate confidence, request battery preheating, reduce current, change the next charge stage or hold SOC. If measurement quality is poor, do not increase current; retain the existing safe charge map. Log accepted energy, charge time and later capacity loss so benefit can be quantified.

Evidence limit. Controlled studies have sensitive ground truth and module prototypes, but no public fleet study reports how often EIS permits higher charge power, how many plating events it prevents, or whether it improves cycle life relative to an optimized V/I/T/model baseline.

3.2 Core temperature and thermal-management control

Why it matters. Surface sensors lag the cell core during fast charge, high-power driving and preconditioning. Impedance phase in roughly the 100 Hz to 1 kHz region is strongly temperature-sensitive, but large automotive cells demand micro-ohm-level precision [5].

How to use it. Schedule several high/mid-frequency points during parked or stable-current windows, or estimate them from a validated passive excitation. Use surface temperature as an independent anchor and estimate core/max temperature plus uncertainty. Recalibrate with age and cell replacement because connector and electrochemical resistance drift.

Vehicle action. Command pump/chiller/preheater, limit regenerative charge, reduce traction or charging power, or update the permissible power envelope. Cross-check against thermistors; a single inconsistent EIS estimate must not suppress an overtemperature response.

EV-specific constraint. A 2026 traction study measured Citroen C-Zero pack-current noise and found significant content overlapping the optimal temperature band, with disturbances potentially affecting EIS up to 3 kHz [5]. It estimated that <1 K uncertainty for high-energy cells above 23 deg C requires <4 micro-ohm uncertainty in both Re and Im. This turns EMI and calibration into first-order product requirements.

3.3 SOH, SOP, range and warranty

Why it matters. Capacity determines usable energy; resistance and thermal state constrain available power. EIS can separate high-frequency connection/ohmic changes from charge-transfer and diffusion changes more richly than a single DC resistance value, but temperature and SOC confound the same features.

How to use it. Measure at repeatable parked or charging states, compare with an individual pack baseline, and update a joint capacity-resistance health state. Use hierarchical models that retain cell/module variation and report uncertainty. DVA/ICA during suitable AC charging can provide occasional degradation-mode confirmation; EIS can provide more frequent updates [10].

Vehicle action. Refine range and power confidence bands; reduce charge/discharge limits for a high-resistance cell group; trigger service; support residual-value or warranty triage. Do not approve a warranty claim or a safety derate from one spectrum without persistence and corroboration.

Evidence limit. Cell studies report 0.75-1.5 percentage-point RMS SOH error [7]; 13 retired Nissan Leaf modules yielded 1.974 percentage-point RMS and 4.935 percentage-point peak error on 128 samples [8]. These are useful benchmarks, not proof for in-vehicle packs across seasons and vehicle variants.

3.4 Safety anomaly, soft short and post-crash quarantine

Why it matters. Incipient internal shorts, abnormal aging and thermal abuse may alter impedance before voltage or surface temperature crosses a conventional threshold. The safety value is early corroboration and fault localization, not replacement of primary protection.

How to use it. Trigger rapid selected-frequency or pulse-derived impedance after an abnormal voltage residual, temperature gradient, gas/pressure event, crash signal or isolation change. Compare each cell/module to its own history and peers. Require persistence and sensor fusion before a service decision; act immediately on conventional hard limits.

Vehicle action. Reduce power, inhibit charging, open contactors, issue a DTC, request controlled discharge, prescribe towing/storage separation, or keep the pack quarantined for repeat diagnostics. In an active thermal event, EIS processing must never delay the primary disconnect or occupant warning.

Evidence limit. NHTSA/Sandia obtained 22.5 minutes of rapid-EIS warning in an 11.6 Ah NMC single-cell overtemperature test and 29.2 minutes in a 1s4p, 46 Ah configuration [16,17]. The same source shows modality ranking changes with abuse mode; gas/H2 can outperform EIS under overcharge. Lab faults and controlled heating do not establish field sensitivity, false alarms or spontaneous-fault coverage.

3.5 Imbalance, interconnect and model maintenance

Per-cell synchronized responses can reveal a module that deviates from peers, while high-frequency features can expose busbar/contact changes. The BMS can use persistent deviations to rebalance, revise thermal or resistance parameters, restrict power or schedule service. Pack-only EIS may hide the weak element: series voltage sums and parallel current sharing dilute local defects [19-21].

For online model maintenance, identify a small ECM or physics-model parameter set from scheduled or passive data. The action is not a direct safety command; it is an updated SOC/SOP/thermal observer that must be regression-tested before activation.

3.6 Manufacturing, service and retired-EV routing

Factory and service conditions can standardize SOC, temperature, rest, fixture and excitation, so they avoid much of the moving-vehicle problem. EIS can compare incoming cells, verify wetting/formation, identify connection or module outliers, and shorten capacity screening. For retired EV batteries it can estimate capacity/resistance health and match modules before second-life assembly.

The action must be explicit: pass, hold, rework, reject, reuse class, restricted-power use or recycle. A useful model reports false accept/reject at production prevalence, gauge repeatability and reproducibility, cycle time, downstream life correlation and routing economics. SOH alone is insufficient; safety history, isolation, swelling/leakage and fault memory remain mandatory.

3.7 Use-case completeness checklist

Use case	Measurement	Output	EV action	Critical proof
SOC correction	Parked or rich passive drive segment	SOC residual + uncertainty	Correct coulomb count / range band	Benchmark against OCV, model and DVA
SOH / RUL	Repeatable SOC/T, periodic	Capacity/resistance health	Limits, warranty/service, value	Seasonal and fleet-label gap
SOP	Dynamic or scheduled ECM update	Resistance + thermal state	Traction/regen/charge limit	Must prove transient prediction
Core T	100 Hz-1 kHz selected points	Tcore/max + confidence	Cooling, heating, derate	Micro-ohm precision and inverter noise
Plating	Operando fast charge	Onset/risk probability	Warm, reduce/change charge	Need pack false-alarm and life benefit
Short / TR	Event-triggered rapid diagnostic	Anomaly type/severity	Isolate, warn, quarantine	Never sole safety channel
Imbalance/contact	Parked/service or coherent global excitation	Peer deviation / resistance	Balance, service, replace	Topology masks localization
Factory/service	Standardized station	Pass/hold/rework score	Disposition	Gauge R&R and prevalence
Second life	Removed module/pack	SOH, power, safety class	Reuse class or recycle	Chemistry/history uncertainty

4. What performance has actually been shown

The following table reports the result, the physical scale and the transfer gap together. Values are not directly comparable because frequency band, cell chemistry, ground truth, rest, temperature and target differ.

Target	Scale / protocol	Reported result	What it proves	What it does not prove
Core temperature	2.3 Ah LFP 26650; +/-20 A pulses; 3,500 s HEV cycle	0.6 deg C MAE; uniform-temperature assumption 2.6 deg C	Strong cell observer result	One instrumented cell; not pack EMI
Traction T measurement requirement	Meta-analysis: 68 publications / 83 cells + Citroen C-Zero current noise	100 Hz-1 kHz preferred; <4 micro-ohm Re/Im uncertainty for <1 K on high-energy cells above 23 deg C	Translates cell sensitivity into BMS specification	Not a vehicle estimator accuracy result
SOH	Commercial cells; multiple institutions/methods	0.75-1.5 percentage-point average RMS error	Replicated cell benchmark	Fixed test conditions; cell variation dominates
Retired EV SOH	13 end-of-first-life 2011 Nissan Leaf modules; 128 samples	1.974 percentage-point RMS; 4.935 peak error	Real retired automotive modules	Small dataset; offline; model split risk
EIS opportunity	7 series-production EVs; 169,760 km	421 idle EIS opportunities / 10,000 km vs 81 DVA	Measurement scheduling is plausible	No EIS was validated onboard
Automotive module metrology	7 kWh, 11S36P, 396 x 21700 cells; 260 cycles	Median EIS-fit RMS error 56 micro-ohm; full EIS + pulse spot about 40 min	Pack-scale calibration and aging trends	Bench tester; too slow for routine onboard use
Lithium plating	Three-electrode cells with titration; two-electrode feasibility	Detected plated Li below 0.6% of graphite capacity	Sensitive mechanism-grounded limit	Cell lab result; no vehicle false alarms
Module plating prototype	Series module topology + equivalent sampling	0.1-1,160 Hz measurement; plating confirmed by capacity/post-mortem	Embedded module feasibility	No sensitivity/specificity or closed-loop benefit
Soft short classifier	7 cells; external 200-10 ohm shunts; about 840 spectra	97.5% average accuracy; zero FP on small normal set	Detectable impedance pattern	Simulated fault; rare-event field rate unknown
Thermal abuse	11.6 Ah NMC cell and 1s4p 46 Ah configuration	22.5 min and 29.2 min warning in overtemperature tests	Potentially actionable lead time	Condition-specific; modality ranking changes
Pack multi-fault	LFP and NCM lab packs; aging, plating, internal short	60 s acquisition; AA error within 5%; ISC within 6.4%	Recent pack-level decoupling result	Energy-storage framing; no road vehicle / field prevalence
Integrated IC metrology	TI BQ79826Z-Q1 preview condition	1% impedance accuracy at 1 A and 200 micro-ohm; 0.01 Hz-3.5 kHz	Concrete chip-level condition	Vendor specification; not diagnostic accuracy

4.1 Interpretation by evidence level

E1-E2 studies answer whether an electrochemical feature contains information. E3 studies answer whether it survives embedded measurement, topology, power electronics or abuse conditions. E4 answers whether an OEM integrated it. E5 must answer whether it works often enough, with sufficiently few false alarms, to improve customer and safety outcomes.

Missing Link: No public source reviewed here combines named high-volume EV deployment, traceable impedance performance across the vehicle envelope, held-out pack diagnostic metrics, and accumulated field outcomes.

4.2 Why a high cell accuracy can fail in a vehicle

• Temperature, SOC, aging and relaxation move overlapping impedance features; a model can learn the laboratory schedule instead of the target.

• A pack contains series/parallel averaging, busbar and contact impedance, common-mode voltage, isolation networks and cell-monitor channel mismatch.

• Fast charging and driving violate stationarity while charger, DC/DC and traction-inverter harmonics overlap useful bands.

• Safety faults are rare. A study with a few healthy cells cannot bound false alarms per fleet-hour, even when classifier accuracy is high.

• Ground truth is often a BMS estimate or simulated shunt rather than destructive confirmation, spontaneous internal short or direct plated-lithium quantification.

• A diagnostic can be accurate but economically useless if it consumes too much rest time, injected energy, service time or calibration effort.

5. EV production and commercial landscape

This section grades public evidence, not supplier competence. Product status is taken from official vendor pages where possible. AEC-Q100 or ASIL capability supports component qualification and safety integration; it does not validate an EIS algorithm or a vehicle safety goal.

Supplier	Public architecture	Public maturity claim	Evidence grade	Still missing
ADI + XPENG	ADBMS6842 cell monitor, ADBMS2970 pack monitor, ADBMS6822, BLISS software	ADI states XPENG is first OEM to adopt ADI EIS for mass-production EV batteries; architecture scales across XPENG EVs	E4 named EV production	Vehicle/model/volume; band; accuracy; false alarms; charge-life, range, safety or warranty outcomes
TI	BQ79826Z-Q1 + BQ79881-Q1; integrated EIS engines; local/global excitation	Preview; preproduction quantities available; production quantities expected by end-2026	Product / reference-design readiness	No named OEM EIS production; condition-wide independent curves; diagnostic performance
NXP	BMA7418 + BMA6402 + BMA8420; hardware sync, on-chip DFT, BJB excitation	Complete solution announced and expected from beginning 2026; active evaluation ecosystem	Automotive chipset availability	No named EV production; public end-to-end impedance or diagnostic error
Marelli	EIS-ready and next-generation Full EIS BMS with cloud/AI claims	Tier-1 says cost-effective solution is ready for large-scale production	Production-readiness claim	No named OEM, shipped population, band, metrology or diagnostic dataset
ST	L9965x application-note system; L9963E peer-reviewed acquisition prototype	5-cell, 32 Ah prismatic demo at 5-520 Hz; L9963E IC is volume production	Vendor demo + E3 cell prototype	Volume production of the monitor is not EIS deployment; no named EV
Infineon	PSoC 4 HVPA-SPM pack monitor supports EIS; cell-monitor ecosystem	Automotive EV/HEV pack-monitor capability and evaluation material	EIS-capable platform	No named EV EIS production; limited public band and algorithm results

5.1 What is genuinely new in the IC generation

The new platforms remove important measurement barriers: synchronized cell and pack sampling, integrated DFT, wider frequency coverage, local or global excitation control, automotive diagnostics and scalable daisy chains. TI publishes a conditional 1% impedance specification; NXP exposes a synchronized BJB excitation path; ADI provides a named production architecture; ST and Infineon provide implementable automotive building blocks [25-35].

They do not remove the system problem. Excitation amplitude at each cell, fixture/busbar de-embedding, nonstationarity, temperature/SOC compensation, feature transfer, uncertainty, out-of-domain detection, safety concept and closed-loop benefit remain OEM responsibilities unless the supplier contract explicitly includes them.

5.2 Supplier diligence questions

• Show Re/Im and magnitude/phase error distributions versus frequency, cell impedance, excitation, temperature and common-mode voltage - not only a nominal point.

• Define the usable frequency band for each topology and vehicle mode, including acquisition time and invalid-spectrum rate.

• Provide module/pack comparison against a traceable reference with charger, inverter, DC/DC, balancing and communications active.

• Separate EIS measurement specifications from estimator or detector performance; provide held-out packs and confidence calibration.

• Identify field exposure, diagnostic prevalence, false alarms per vehicle-hour, intervention logic and customer/warranty outcome.

• Provide ISO 26262 work products for the implemented safety mechanism, freedom-from-interference analysis and fallback behavior.

• Clarify data ownership, calibration persistence after service, cell replacement and over-the-air software updates.

6. EV implementation architectures

6.1 Functional blocks

Block	Typical EV realization	Engineering purpose
Excitation	Onboard charger, DC-link capacitor / BJB switch, DC/DC, traction inverter, balancing path or local exciter	Generate known small signal without violating voltage/current, EMI or driver-comfort limits
Cell response	Synchronous per-cell ADCs / monitor ICs	Preserve relative phase and localize a weak cell or group
Pack reference	Shunt/Hall/fluxgate current + pack voltage in BJB	Measure actual excitation and reject parasitic or unsynchronized records
Edge compute	DFT/FRA, feature extraction, quality metrics, compact model	Avoid raw-data bandwidth; create spectrum with traceability
BMS application	State fusion, baseline, uncertainty, decision manager	Convert impedance into a bounded request and fallback
Cloud / service	Trend store, fleet calibration, post-event analysis	Improve model and evidence while respecting privacy and safety partitioning

6.2 Architecture tradeoffs

Architecture	Benefit	Main risk	Best EV fit
Global pack/BJB excitation	One source; coherent comparison of all monitored cells; natural fit for new chipsets	Tiny per-cell response; common mode; pack parasitics; parallel-cell masking	Scheduled parked/charging diagnostics
Charger or DC/DC injection	Reuses high-power electronics; controllable while charging	Loop interaction, small-signal fidelity, EMI, unavailable when disconnected	SOH/model update; operando plating
Traction-inverter / passive	No dedicated energy; can use real drive current	Uncontrolled spectrum; road-load nonstationarity; switching harmonics	Opportunistic ECM/temperature trend
Balancing-path excitation	Low incremental BOM; per-cell location	Small current, heat, limited band and ADC precision	Parked selected-point checks
Local cell/module exciter	Strong localization and smaller high-voltage loop	BOM, quiescent power, calibration and safety per channel	Premium / safety-critical module design
External service instrument	Traceable, broad band and controlled fixture	Not continuous; labor and safe pack access	Service, warranty, second-life grading

6.3 Frequency selection by decision

A broad spectrum is valuable for model discovery and service. Production control usually needs a sparse feature set. High-frequency regions emphasize inductance, connections and ohmic effects; mid-frequency regions include interfacial/charge-transfer behavior and temperature-sensitive phase; low-frequency regions include diffusion and can improve aging or abuse sensitivity but demand long windows and are most vulnerable to drift.

Frequency must be selected from a sensitivity-to-uncertainty budget. For example, moving above traction noise may reduce interference but also reduce temperature sensitivity [5]. The correct design is multi-frequency redundancy with quality scoring, not a universal single frequency.

6.4 Measurement-quality gate

Gate	Pass criteria
Preconditions	SOC/T/vehicle mode in calibrated domain; no hard protection event; adequate time and excitation authority.
Signal checks	No clipping; current amplitude achieved; timestamp/sync health; SNR/coherence; harmonic distortion; channel agreement.
State checks	SOC and temperature drift within limits; relaxation/prior-current state known; contactor and load state recorded.
Physics checks	Residual or Kramers-Kronig-type consistency where applicable; plausible spectrum; stable feature across repeats.
Decision checks	Uncertainty below action threshold; persistence/corroboration satisfied; action bounded; fallback available.

7. Functional safety, cybersecurity and vehicle integration

Safety Position: EIS is a supporting diagnostic channel. It is not an intrinsically ASIL-D function and should not be the only path for overvoltage, overtemperature, isolation, overcurrent or contactor protection.

ISO 26262 governs the vehicle safety lifecycle, hazard analysis, safety goals, hardware/software development and verification [36]. ISO 6469-1 addresses rechargeable energy storage system safety, and UNECE GTR No. 20 addresses electric-vehicle safety at the regulatory level [37,38]. None makes an EIS-capable IC, spectrum or algorithm safe by itself.

Failure mode	Example	Required behavior
Measurement unavailable	No stable window, excitation fault, sleep budget	Keep conventional state estimate; no more aggressive limit; reschedule.
Bad spectrum	EMI, clipping, timing, drift, model residual	Mark invalid; retain data for diagnosis; do not actuate from it.
Estimator disagreement	EIS conflicts with V/I/T/model	Reduce confidence; conservative envelope; repeat or service check.
Safety anomaly	EIS + voltage/temp/gas/isolation concern	Execute predefined safe state; contactors/warning are not delayed by analysis.
Software/calibration update	OTA model or threshold change	Versioned calibration, rollback, regression corpus and safety impact analysis.
Spoofed or corrupted data	Bus, clock or cloud integrity failure	Authenticate, plausibility check, isolate safety partition, use local fallback.

For post-crash use, define who owns the decision after propulsion shutdown: onboard BMS, diagnostic service tool, emergency responder workflow or storage operator. Repeat measurements may help determine whether a damaged pack remains quarantined, but thermal, voltage, gas, isolation and mechanical inspection remain primary.

8. EV validation program and acceptance gates

8.1 Stage-gated program

Gate	Minimum test	Exit evidence
V0 - metrology	R/C networks, shunts and stable cells; common-mode, temperature and cable matrix	Traceable uncertainty; calibration/de-embedding; repeatability; sync and drift
V1 - cell	Multiple lots of target LFP/NMC formats; SOC/T/rest/current/aging matrix	Held-out cell error; feature stability; destructive/thermal/capacity ground truth
V2 - module	Series/parallel production topology; busbars, cooling, compression and cell imbalance	Localization, masking, channel variation and embedded excitation performance
V3 - pack/HIL	Full common mode; charger/inverter/DC/DC/communications; faults and timing injections	Valid-spectrum rate; EMI immunity; FTTI; fallback and safety partition
V4 - vehicle	Cold soak, AC/DC charging, WLTP/UDDS/US06, mountain/regen, hot soak, key-off	Action benefit and drivability; energy/time cost; service behavior
V5 - fleet	Seasonal vehicles, variants and software versions; real prevalence	False alarms / exposure; interventions; lifetime, charging, warranty and safety outcomes

8.2 EV environmental matrix

Dimension	Minimum coverage
Chemistry / format	Target LFP and/or NMC; prismatic, pouch or cylindrical; production lots; service replacements.
SOC	5-95% intended domain; dense points near charge taper and chemistry-specific transitions.
Temperature	At least cold soak, 0 deg C region, nominal, hot operation; imposed core-surface gradients and cooling faults.
Aging	Calendar + cycle; fast/slow charge; depth-of-discharge; high/low temperature; storage SOC; connection aging.
Vehicle power states	Sleep, key-on, AC charge, DC fast charge, acceleration, cruise, regeneration, HVAC and preconditioning.
Electrical interference	Inverter speed/load map, OBC/DC-DC switching, balancing, contactors, communications, conducted/radiated immunity.
Fault truth	Plating quantified chemically/post-mortem; instrumented core T; calibrated shorts; physical connection defects; abuse timing.
Service lifecycle	Module replacement, firmware/calibration update, cell supplier change, crash/quarantine, storage and second-life handoff.

8.3 Acceptance metrics by use case

Use case	Report at minimum	Ground truth
Core temperature	MAE/RMSE/bias, 95th percentile, valid-domain coverage, lag during gradients, false cold/hot decisions	Instrumented core and surface sensors
SOH / SOP	Capacity/resistance error, uncertainty calibration, range/power prediction error, seasonal bias	Controlled energy/capacity and HPPC/dyno reference
Plating	Detection limit, sensitivity/specificity, charge-energy throughput, time saved, irreversible plated Li and cycle life	Titration/post-mortem/reference electrode plus lifetime
Short / TR	Lead time distribution, sensitivity, false alarms / 1,000 vehicle-h, localization, intervention success	Reproducible and naturalistic faults; healthy fleet exposure
Instrument	Re/Im and magnitude/phase error, repeatability, invalid rate, sync, drift, acquisition time and injected energy	Traceable standards and lab FRA/potentiostat
Factory/service	False accept/reject, gauge R&R, cycle time, downstream yield/life, technician repeatability and cost	Disposition and long-term outcome

8.4 Comparator requirement

Every EIS function must be compared with the best credible lower-cost baseline: V/I/T plus observer, DC resistance/HPPC, relaxation voltage, incremental capacity/differential voltage, gas/pressure/acoustic sensing, or a short pulse. Report incremental benefit, not only standalone accuracy. Full EIS is unjustified when a five-point pulse or one temperature-compensated resistance value gives the same action quality.

9. Research that would close the EV evidence gap

Study A - vehicle core-temperature observer

Instrument cells in production-representative LFP and NMC modules with core and surface temperature. Calibrate 100 Hz-1 kHz phase/complex impedance on bench, then validate in a pack/HIL and vehicle across cold soak, preconditioning, DC fast charge, highway, hill climb and regeneration. Compare active selected-point and passive estimates against a thermal observer. Primary endpoints: 95th-percentile core/max error, valid-spectrum rate, EMI sensitivity and incremental thermal/power-control benefit.

Study B - closed-loop plating-aware fast charge

Randomize matched cells/modules to (1) production V/I/T charge map and (2) the same map with confidence-gated impedance feedback. Use low-temperature and aged-cell conditions, quantify plating with titration/post-mortem subsets, and continue cycle life after intervention. Report charge time and accepted energy together with irreversible plated lithium, capacity retention, false interventions and energy used for preheating/excitation.

Study C - pack safety and post-crash quarantine

Test production topology with calibrated soft shorts, connection defects, abnormal aging, controlled heating and mechanically damaged modules. Combine EIS with voltage, temperature, isolation, gas/pressure and crash signals. Pre-register decision thresholds. Report lead-time distributions, localization, false alarms on long healthy pack/vehicle exposure, intervention success and repeatability during post-event cooling/storage.

Study D - passive versus active EIS in real vehicles

Collect synchronized high-bandwidth cell/pack data across parked, AC/DC charge and representative drive cycles. At matched states, compare passive drive-current estimates, BJB/charger active excitation and a service reference. Map usable frequency coverage, bias, invalid rate, acquisition time and energy by vehicle mode. The output is a scheduler and uncertainty model, not only a mean spectrum.

Study E - retired EV pack routing

Acquire packs/modules with known vehicle history and a broad safety/degradation distribution. At standardized SOC/T, compare a 30-180 s EIS screen with full capacity/power tests, inspection, isolation and selected teardown. Evaluate routing to high-power reuse, energy reuse, restricted use or recycling. Report false-safe and false-reject rates, throughput, technician variability, downstream second-life performance and net recovered value.

9.1 Publication package

• Raw synchronized current and per-cell voltage windows, not only fitted parameters or Nyquist images.

• Cell/pack pedigree, topology, thermal boundary, fixture, SOC-setting method, rest and prior current direction.

• Calibration and uncertainty budget; exact quality rejection rules and invalid-measurement rate.

• Training/validation split by physical cell, lot, pack and vehicle; no leakage through repeated spectra.

• Predefined comparator and action thresholds; confidence calibration and out-of-domain behavior.

• Outcome data: charge, life, range, safety, warranty/service time and false alarms normalized by exposure.

10. Conclusions and engineering recommendation

Embedded EIS now has credible automotive hardware and the first named mass-production OEM integration. The research base demonstrates that impedance contains useful information about internal temperature, aging, plating and safety anomalies, and that module/pack acquisition is feasible. Therefore the correct conclusion is not that EV EIS has no use case.

The correct conclusion is narrower: public evidence has not yet closed the loop from calibrated impedance to a proven fleet outcome. Most performance numbers remain cell or controlled-module results. Supplier product pages show capability; they rarely show diagnostic generalization, false alarms or customer benefit. XPENG/ADI establishes production adoption, not disclosed performance.

For an EV program, prioritize one bounded use case and one natural measurement window. The most defensible pilots are parked health/model updates, service or retired-pack screening, and internal-temperature correction. Plating-aware fast charge and early safety warning may have larger value, but require stronger pack-scale ground truth, sensor fusion and rare-event validation.

Go / No-Go: Proceed to vehicle pilot only when the embedded system meets a traceable impedance uncertainty budget, produces calibrated diagnostic uncertainty, survives EV EMI/topology, fails conservatively, and demonstrates incremental action benefit over the production V/I/T baseline.

Appendix A. Supplier request-for-information template

Topic	Ask for
Architecture	Excitation source/path; per-cell/pack measurement; frequency list; amplitude; simultaneous channels; common-mode range.
Timing	Acquisition and compute time by band; synchronization error; communication load; vehicle modes and preconditions.
Metrology	Reference method; Re/Im and phase/magnitude error surfaces; repeatability; drift; calibration and de-embedding.
Vehicle robustness	Inverter/OBC/DC-DC/balancing interference; EMC results; invalid rate; temperature and aging envelope.
Algorithm	Target definition; feature/model; training population; held-out vehicle split; uncertainty and out-of-domain logic.
Safety	Safety goal allocation; ASIL decomposition; FTTI; diagnostic coverage; freedom from interference; fallback and fault injection.
Production	Named program or anonymized SOP evidence; vehicle population; software versions; calibration after service / OTA.
Field outcome	Sensitivity/specificity, false alarms per exposure, interventions, charging/life/range/warranty benefit and confidence interval.

Appendix B. Reporting checklist

• [ ] Cell/module/pack model, chemistry, form factor, topology, lot, sample count and prior use.

• [ ] SOC method/error; chamber, surface and core temperature; rest; pressure/fixture; prior current direction.

• [ ] Excitation waveform, amplitude, DC bias, frequency points, cycles/settling, injected Ah/Wh and vehicle operating state.

• [ ] Current/voltage ranges, sampling, clocks, synchronization, filtering, isolation, bit depth and data reduction.

• [ ] Calibration standards, four-wire plane, open/short/load or de-embedding, uncertainty and calibration interval.

• [ ] SNR/coherence, harmonics, drift, residual/physics consistency, rejection rule and invalid rate.

• [ ] Independent ground truth and timing: capacity, core T, chemical titration, post-mortem, fault resistance or event marker.

• [ ] Training split by physical unit; held-out lots/packs/vehicles; baseline comparator; confidence and OOD performance.

• [ ] Action logic, fallback, safety analysis, intervention outcome and exposure-normalized false alarms.

Appendix C. Glossary

Term	Meaning
AFE	Analog front end.
BJB	Battery junction box; location for pack current/voltage, contactor, fuse and isolation functions.
BMS	Battery management system.
DFT / FRA	Discrete Fourier transform / frequency-response analysis.
DVA / ICA	Differential voltage / incremental capacity analysis.
ECM	Equivalent circuit model.
EIS	Electrochemical impedance spectroscopy; complex response versus frequency.
FTTI	Fault-tolerant time interval.
OOD	Out of distribution; operating data outside the validated model domain.
SOC / SOH / SOP	State of charge / state of health / state of power.
RUL	Remaining useful life.
TR	Thermal runaway.

References

URLs and product pages were accessed on 2 September 2026. Supplier sources substantiate product capability and deployment claims; peer-reviewed or government sources substantiate research results.

[1] McCarthy et al., Review - Use of Impedance Spectroscopy for the Estimation of Li-Ion Battery SOC, SOH and Internal Temperature, JES 168 (2021) 080517. Source

[2] Wang et al., Electrochemical impedance measurement and analysis for automotive batteries: review and perspective, Current Opinion in Electrochemistry (2025) 101768. Source

[3] Hallemans et al., Measurement, interpretation, and application of EIS to lithium-ion batteries, EES Batteries 2 (2026) 80-102. Source

[4] Richardson, Ireland and Howey, Battery internal temperature estimation by combined impedance and surface temperature measurement, JPS 265 (2014) 254-261. Source

[5] Sailer, Steffan and Schmidt, Practical considerations and limitations of online EIS-based battery internal temperature estimation in traction applications, JPS 667 (2026) 239111. Source

[6] Haussmann and Melbert, Internal Cell Temperature Measurement and Thermal Modeling for Automotive Applications by EIS, SAE 2017-01-1215. Source

[7] Chen et al., Comparison of methodologies to estimate SOH of commercial Li-ion cells from electrochemical frequency response data, JPS 542 (2022) 231814. Source

[8] Rastegarpanah et al., Rapid Model-Free SOH Estimation for End-of-First-Life EV Batteries Using Impedance Spectroscopy, Energies 14 (2021) 2597. Source

[9] Kasper et al., Calibrated EIS and Time-Domain Measurements of a 7 kWh Automotive Module with 396 Cells, Batteries & Supercaps (2023) e202200415. Source

[10] Urban et al., Potential of DVA and EIS for EVs based on real-world data, eTransportation 29 (2026) 100619. Source

[11] Brown et al., Detecting onset of lithium plating during fast charging using operando EIS, Cell Reports Physical Science 2 (2021) 100589. Source

[12] Strasser, Adam and Li, In operando detection of lithium plating via EIS for automotive batteries, JPS 580 (2023) 233366. Source

[13] Shen et al., Impedance Evolution for Lithium Plating Detection during Multi-Stage Constant Current Fast Charging, SAE 2025-01-8124. Source

[14] Zhang et al., Equivalent sampling-enabled module-level battery impedance measurement for in-situ lithium plating diagnostic, JPS 600 (2024) 234239. Source

[15] Cui et al., Internal short-circuit early detection from impedance spectroscopy using deep learning, JPS 563 (2023) 232824. Source

[16] Torres-Castro et al., Early Detection of Li-Ion Battery Thermal Runaway Using Commercial Diagnostic Technologies, JES 171 (2024) 020520. Source

[17] NHTSA/Sandia, Early Detection of Thermal Runaway With Advanced Diagnostics, DOT HS 813 671. Source

[18] Perez et al., Impedance-based thermal-runaway early detection methodology for lithium-ion batteries, IJEPES 171 (2025) 111053. Source

[19] MRS Bulletin, Challenges and needs for system-level electrochemical lithium-ion battery management and diagnostics 46 (2021) 420-428. Source

[20] Sandschulte and Ferrero, Multi-Cell and Wide-Frequency In-Situ Battery Impedance Spectroscopy, IEEE OJIM 2 (2023) 1-11. Source

[21] Multi-cell testing topologies for defect detection using EIS, Batteries 9 (2023) 415. Source

[22] Mattia et al., A low-cost approach to on-board EIS for a lithium-ion battery, Journal of Energy Storage 81 (2024) 110330. Source

[23] Liebhart, Diehl and Endisch, Passive impedance spectroscopy for monitoring lithium-ion cells during vehicle operation, JPS 449 (2020) 227297. Source

[24] Schmidt et al., Employing Real Automotive Driving Data for EIS on Lithium-Ion Cells, SAE 2015-01-1187. Source

[25] Analog Devices, XPENG: First OEM to Implement EIS Battery Intelligence, 15 June 2026. Source

[26] Analog Devices, ADBMS2970 Battery Pack Monitor with EIS support. Source

[27] Texas Instruments, BQ79826Z-Q1 Automotive 26-S Battery Monitor with Smart EIS Engine. Source

[28] Texas Instruments, BQ79881-Q1 BJB Pack Monitor with EIS. Source

[29] Texas Instruments, EIS-enabled BQ79826Z-Q1 announcement, 9 June 2026. Source

[30] NXP, EIS-capable BMS chipset announcement, 29 October 2025. Source

[31] NXP, BMA8420 Battery Junction Box Monitor IC with EIS Capability. Source

[32] STMicroelectronics, AN6310: EIS implementation with L9965x battery management chipset. Source

[33] STMicroelectronics, L9963E automotive battery-monitor product page. Source

[34] Infineon, PSoC 4 HVPA-SPM 1.0 smart battery pack monitor with EIS support. Source

[35] Marelli, EIS-based Battery Management Systems solution announcement, CTI Symposium Berlin 2024. Source

[36] ISO 26262-1:2018, Road vehicles - Functional safety - Part 1: Vocabulary. Source

[37] ISO 6469-1:2019, Electrically propelled road vehicles - Safety specifications - Rechargeable energy storage system. Source

[38] UNECE, Global Technical Regulation No. 20: Electric Vehicle Safety. Source

[39] Multi-fault diagnosis of lithium-ion battery packs based on fast EIS under aging, plating and internal short, Journal of Energy Storage 169 (2026) 122749. Source

[40] Architectural Pathways and Integration Constraints for Feasible Onboard EIS for Battery Electric Vehicles, World Electric Vehicle Journal 17 (2026) 315. Source