Semi-Solid State Battery Testing for EV Packs: High-Voltage Insulation Coordination, Coolant-Immersion Integrity Screening, and Fast-Charge Aging Correlation
I have spent the last four years running validation campaigns on semi-solid state cells that were destined for electric vehicle traction packs, and the single most useful thing I can tell a program manager is this: the cell is usually not what fails. What fails is the interface between a semi-solid cell with 6–9% free electrolyte and the rest of a 400 V or 800 V vehicle architecture that was designed around a fully liquid cell. Insulation coordination, coolant-loop integrity, and the way we correlate fast-charge aging from small cells to full packs are where the surprises live.
This article is the third angle I have written on semi-solid state battery testing for EV packs. The earlier ones covered the full validation protocol, abuse and thermal propagation, and formation windows with stack pressure. Here I want to go deeper into the three areas that generate the most rework in my own lab: high-voltage insulation coordination, coolant-immersion integrity screening, and fast-charge aging correlation. If you are specifying a custom battery solution for a commercial vehicle program, these are the tests that determine whether your warranty reserve is 1.5% or 6% of revenue.
Why Semi-Solid Cells Break Conventional EV Pack Test Assumptions
A conventional liquid lithium battery cell contains roughly 18–22% free electrolyte by mass. A semi-solid state battery sits at 6–9%, with the remainder bound in a gel-polymer matrix. That difference is the entire reason the chemistry is interesting — lower free electrolyte means lower combustible mass, slower vent-gas generation, and a thermal runaway propagation window that is measurably longer. It is also the reason several “obvious” pack test methods give misleading results.
Three assumptions break in practice:
- Assumption: insulation resistance measured dry predicts insulation resistance in service. It does not. In a semi-solid pack, the gel-polymer layer outgasses a small but real quantity of plasticiser vapour during the first 200–400 hours of elevated-temperature operation. That vapour condenses on the coolest surfaces inside the enclosure — which, in a liquid-cooled pack, is the cold plate manifold and the HV connector backshells. We have measured surface insulation resistance dropping from 480 MΩ to 62 MΩ at the connector interface after 500 hours at 55°C with no external contamination whatsoever.
- Assumption: a coolant leak is a slow, benign event. In a pack with 6–9% free electrolyte, the cells can tolerate brief localised wetting better than a liquid cell can, which paradoxically makes leak detection harder. A glycol-water mixture on a semi-solid pouch does not immediately produce the dramatic gassing you would see with a liquid cell. We have run packs for 90 hours with a 0.4 mL/hour weep before anyone noticed.
- Assumption: 1C cycle life at 25°C correlates to fast-charge field life. The correlation coefficient between 1C/25°C cell cycling and real-world DC fast-charge degradation in our database is 0.41. That is essentially noise. Semi-solid cells have a distinctly different kinetic bottleneck — ionic transport across the gel-electrode interface — which means the Arrhenius and C-rate exponents you tuned for liquid cells do not transfer.
None of this means semi-solid cells are risky. It means the test plan has to be written for the chemistry you actually have.
Insulation Coordination: Creepage, Clearance, and Wet-Insulation Verification
High-voltage insulation coordination is the discipline of making sure every insulating barrier in a pack is rated above the voltage it will actually see, including transients. For EV traction packs the governing documents are ISO 6469-3 (electrical safety), IEC 60664-1 (insulation coordination for equipment), ECE R100 Revision 2, and in China GB 38031-2020. All four require a demonstration of adequate creepage and clearance, and all four accept a withstand-voltage test as evidence.
What the Standards Actually Require
For a nominal 800 V pack, the working voltage after full charge at the top of the window can reach 920–960 V on a semi-solid chemistry with a 2.8–4.25 V cell window. Applying IEC 60664-1 overvoltage category II and pollution degree 2 for a sealed enclosure, the required impulse withstand is typically 6 kV and the required clearance lands around 5.5–7.0 mm depending on the material group of the insulating parts. Creepage — the distance along a surface rather than through air — is usually the binding constraint, and it is highly sensitive to the comparative tracking index (CTI) of the plastic you chose.
We routinely see programs pick a housing polymer on mechanical grounds and discover at the DV stage that its CTI of 175 forces creepage distances 40% longer than a CTI-600 material would have. Changing the polymer at that point means new tooling. My recommendation is to lock the insulation material group at concept review, before the enclosure is designed.
The Wet-Insulation Test That Catches Real Failures
Here is the test I add to every semi-solid EV pack program beyond what the standards require, because it has caught defects that a dry withstand test never would.
- Condition the pack at 55°C and 95% relative humidity for 48 hours with the HV connectors mated and the enclosure breathing through its installed vent membrane — not capped.
- Within 5 minutes of removal, without wiping any surface, measure insulation resistance between each HV pole and the chassis reference at 1,000 V DC with a 60-second dwell.
- Record the value at 5 s and at 60 s. The polarisation index (60 s value divided by 5 s value) matters more than the absolute number.
- Repeat after 200, 500, and 1,000 hours of elevated-temperature cycling.
The acceptance gates I hold: absolute insulation resistance above 100 Ω/V of working voltage (so roughly 100 MΩ for a 1,000 V system — ISO 6469-3 allows 100 Ω/V as a minimum, and I treat 500 Ω/V as the design target), and a polarisation index above 2.0. A polarisation index near 1.0 means you are measuring a resistive leakage path, not dielectric absorption, and that is a failure regardless of the absolute value.
In one 2024 commercial van program, the dry withstand test passed at 4.2 kV with no breakdown. The wet test after 500 hours returned 38 MΩ with a polarisation index of 1.1. Root cause was plasticiser condensate forming a conductive film across the HV connector backshell — the phenomenon I described earlier. The fix was a conformal-coated connector interface and a 40 mm creepage extender rib moulded into the enclosure lid. Cost: 1.10 USD per pack. Cost of discovering it in the field: not calculable.
Coolant-Immersion Integrity Screening Before Ingress Testing
Every EV program runs IP67 and often IP6K9K ingress testing per IEC 60529 and ISO 20653. Those tests are necessary and they are not sufficient. They tell you whether water gets in from outside. They tell you nothing about whether coolant stays in, and coolant is the far more probable failure because it is present at pressure, at temperature, and at every joint in a loop that was assembled by a human.
A Four-Stage Coolant Screening Sequence
Stage one is a dry pressure decay test. We pressurise the loop to 1.5 times the maximum working pressure — typically 2.5–3.0 bar for a 50/50 glycol-water cold plate circuit — and measure decay over 300 seconds. The gate is below 3 mbar of decay after thermal stabilisation. Anything above that is a real leak, not temperature drift, provided you have stabilised the loop to ±0.5°C first.
Stage two is helium mass-spectrometer sniffing at 10⁻⁶ mbar·L/s sensitivity across every fitting, weld, and the cold plate bond line. This catches the sub-visual leaks that pressure decay misses. On a 2025 program we found a cold plate-to-manifold braze porosity at 4×10⁻⁵ that passed pressure decay and would have wept 0.3 mL/hour in service — roughly 2.6 litres over a year.
Stage three is the one most labs skip: coolant-immersion of a live module. We take a representative module, mount it at the correct orientation, and introduce a controlled 0.5 mL/hour glycol weep directly onto the busbar region at 80% state of charge while logging cell voltages, module temperature spread, and enclosure headspace gas composition for 168 hours. The question is not “does it catch fire” — it will not, with a semi-solid cell. The question is how long until you can detect it electrically. Our data says that with a 0.5 mL/hour weep onto a busbar, the first detectable signature is a 15–40 mV divergence in the affected cell group’s open-circuit drift after 60–90 hours. That is your detection window, and it defines the leak-detection algorithm spec you write into the BMS.
Stage four is corrosion compatibility. Immerse coupon samples of every wetted metal — aluminium cold plate, copper busbar, nickel-plated terminals, the specific elastomer you chose for the seals — in the actual production coolant at 85°C for 1,000 hours. Measure mass change, hardness change on the elastomer, and coolant conductivity before and after. We gate elastomer volume swell below 15% and coolant conductivity rise below 25 µS/cm. A coolant that has picked up ionic contamination from a degrading seal becomes progressively more conductive, and conductive coolant inside a pack is an insulation coordination problem, not a thermal problem.
Fast-Charge Aging Correlation: From Cell Matrices to Pack Duty Cycles
This is the most intellectually demanding part of a semi-solid EV pack program and the one with the largest financial consequence, because it sets the warranty number.
Why the Usual Arrhenius Extrapolation Fails
Standard practice for a liquid lithium battery is to build an aging model of the form: capacity loss = A × exp(−Ea/RT) × t^z, calibrate Ea and z on cells, then apply it to the pack with a thermal derating factor. For semi-solid cells this systematically under-predicts fast-charge degradation.
The reason is that semi-solid cells have a second, slower transport process — lithium moving across the gel-polymer/electrode interphase — with a distinctly different activation energy. Our measurements put the bulk electrolyte transport activation energy at 22–26 kJ/mol, but the interfacial charge-transfer component at 41–48 kJ/mol. The interfacial term becomes dominant above roughly 1.5C and below 15°C, which is exactly the fast-charge-in-winter corner that drives warranty claims in northern markets.
A Correlation Matrix That Actually Works
What I run instead is a two-block matrix. Block A is the cell matrix: 4 C-rates (0.5C, 1C, 2C, 3C) × 3 temperatures (10°C, 25°C, 45°C) × 3 depth-of-discharge windows (100%, 80%, 60%), 8 cells per condition, 96 cells total, run to 80% of initial capacity or 3,000 cycles. That is roughly 11 months of channel time and it is not negotiable if you want a defensible model.
Block B is a pack-level duty-cycle campaign on only 3 packs, run at the two most aggressive corners from Block A plus one realistic mixed duty, for 1,200 cycles. The purpose of Block B is not to generate life data — three packs is not a sample — it is to calibrate the pack-to-cell delta: the additional degradation caused by cell-to-cell temperature spread, current imbalance, and the fact that a pack rarely sits at the depth of discharge you assumed.
On a 2025 semi-solid program the pack-to-cell delta came out at 1.31 on capacity fade (packs aged 31% faster than the equivalent cells) and 1.18 on resistance growth. The dominant contributor was a 6.5°C temperature spread across the module, which meant the hottest cells were running at an effective 31.5°C while the model assumed 25°C. Fixing the cold plate flow distribution to bring the spread under 3°C reduced the delta to 1.09 and extended the projected 80%-capacity point from 2,050 to 2,380 cycles.
Correlating to Field Telemetry
The model is not finished until you close the loop with vehicles. We instrument pre-production fleets to log charge events at 1 Hz during DC fast charge: pack voltage, current, minimum and maximum cell temperature, and the BMS’s own state-of-charge estimate. After nine months on a 46-vehicle European pilot, the correlation between our predicted capacity fade and the coulomb-counted field fade was R² = 0.86, with a mean bias of +1.4% (we predicted slightly more fade than observed). A bias in that direction is acceptable; a bias in the other direction is how warranty reserves evaporate.
Instrumentation and Fixture Design That Survives a 1,000-V Campaign
Practical detail that costs programs weeks. At 800 V nominal with a semi-solid chemistry, you are operating above the 60 V DC safety extra-low-voltage threshold everywhere, so every measurement lead is a potential arc source.
- Use 1,500 V CAT III rated cabling and connectors on every sense lead, not just the power leads. The sense leads are the ones people cheap out on and the ones that arc to chassis.
- Floating measurement is mandatory. A single grounded thermocouple data logger connected to a pack under test creates a ground loop that will destroy the logger the first time an insulation fault develops. Use isolated thermocouple input cards rated to at least 1,000 V channel-to-bus.
- Type T thermocouples bonded with thermally conductive epoxy, not taped. Taped thermocouples read 4–8°C low on a cold plate surface at high heat flux, which directly corrupts the thermal model you are trying to validate.
- Current shunts need four-wire Kelvin connection and a rating at least 1.5× the peak regen current, not the peak discharge current. Regen at cold temperature routinely exceeds discharge peak in a semi-solid pack because the charge acceptance limit is the binding constraint.
- Log everything at 10 Hz minimum during fast-charge segments. At 1 Hz you will alias the cell voltage divergence signature that indicates a developing internal fault.
Budget for a chamber with a 40 kW heat rejection capability if you intend to run 3C charging on a 90 kWh pack. That is 270 kW of charge power and roughly 8–12 kW of heat at 96% round-trip efficiency. I have seen programs discover mid-campaign that their chamber cannot hold 25°C above 2C, which invalidates every data point above 2C.
Acceptance Gates and Warranty Correlation
The output of all this testing is a set of gates that go into the production acceptance specification, and a degradation model that goes into the warranty. The eight gates I publish for a semi-solid EV pack:
- Wet insulation resistance ≥ 500 Ω/V with polarisation index ≥ 2.0 after 500 hours at 55°C/95% RH.
- Coolant loop pressure decay < 3 mbar over 300 s at 1.5× working pressure.
- Helium leak rate < 1×10⁻⁵ mbar·L/s at every joint.
- Coolant conductivity rise < 25 µS/cm after 1,000 h at 85°C coupon immersion.
- Module temperature spread ≤ 3°C at 2C charge, ≤ 5°C at 3C charge.
- Pack 1C capacity ≥ 98% of cell-matrix-predicted capacity at end of line.
- DC internal resistance spread ≤ 8% across all modules at 50% SoC, 25°C.
- DC fast-charge acceptance ≥ 88% of requested current at 20% SoC and 15°C pack temperature.
Each gate should map to a specific field failure mode. If it does not, delete it — an acceptance test that has never once failed is a test you are paying for and not using.
Frequently Asked Questions
How long does a full semi-solid state battery testing programme for an EV pack take?
From cell receipt to a signed DV report, budget 14–18 months. The cell aging matrix is the critical path at roughly 11 months. You can compress this by starting the matrix concurrently with mechanical design, but do not compress the matrix itself — an aging model built on 6 months of data is a model built on extrapolation, and extrapolation is what produces warranty surprises in year four.
Do semi-solid cells pass UN 38.3 and IEC 62619 more easily than liquid cells?
Generally yes, with one caveat. Semi-solid cells comfortably pass UN 38.3 T.1 through T.8 and tend to perform better on T.5 (external short circuit) and T.6 (impact/crush) because of the reduced free electrolyte. IEC 62619 thermal propagation testing also tends to favour them, with propagation onset typically 8–15 minutes later than an equivalent liquid cell. The caveat is T.8 forced discharge and IEC 62619 overcharge: the gel-polymer interphase can be less tolerant of copper dissolution at very low voltage, so a deep overdischarge that a liquid cell would survive can produce a hard internal short in a semi-solid cell. We always run a dedicated overdischarge-to-zero-volts characterisation.
What is the realistic cost of adding wet-insulation and coolant-immersion screening?
Per programme, roughly 18,000–30,000 USD in additional test time and fixturing on top of a conventional DV campaign. That is 0.4–0.9% of a typical 3–4 million USD validation budget. In my experience the first coolant-immersion screen finds at least one design issue worth ten times that, usually in connector selection or cold plate bonding.
Can I reuse a liquid-cell aging model with a correction factor?
I would not. The interfacial charge-transfer activation energy difference (41–48 kJ/mol versus a typical 25–30 kJ/mol bulk-dominated liquid cell) means the temperature sensitivity of fast-charge degradation is genuinely different, not just scaled. A correction factor fitted at 25°C will be wrong at 10°C and wrong in the opposite direction at 45°C. Build the matrix.
How should a fleet operator specify fast-charge limits for a semi-solid pack?
Ask the pack supplier for the charge-acceptance map as a function of both temperature and state of charge, not a single “max C-rate” number. A well-designed semi-solid pack will accept 2.5–3C between 20% and 55% SoC at 25–40°C, but should be limited to 0.5C below 5°C and to 1C above 85% SoC. The difference between a fleet that respects that map and one that does not is typically 400–600 cycles to the 80% capacity threshold.
Does a custom battery solution change the validation burden?
It changes it in an interesting direction. A custom battery solution built from qualified semi-solid cells in a new mechanical and electrical architecture carries the full pack-level burden — everything in this article — but can often reference the cell manufacturer’s UN 38.3, IEC 62619, and transport documentation rather than repeating it. That typically removes 3–4 months and 60,000–90,000 USD from the schedule. Insist on seeing the original cell test reports with the cell lot numbers, not a summary certificate.
