The lithium-ion battery industry demands the same analytical rigour as pharmaceuticals — a few ppm of Fe or Cu in a cathode material can cause internal short circuits, capacity fade, and catastrophic thermal runaway. This technical note adapts the ICH Q2(R2) validation framework for ICP-OES and ICP-MS methods used in battery material characterisation.
From Pharma to Gigafactory
ICH Q2 was written for drug manufacturers, but the core logic is universal: prove your method measures what it claims, at the level it claims, every time it claims. Battery OEMs and cell manufacturers (CATL, Samsung SDI, LG, Northvolt, Panasonic) increasingly require validated analytical methods from their internal testing labs and material suppliers.
Why ppb-level impurities matter in batteries
Metallic impurities (especially Fe, Cu, Cr, Zn) dissolve from cathode particles, migrate through the electrolyte, and deposit as metallic dendrites on the anode. At sufficient concentration, these dendrites pierce the separator and create an internal short circuit — triggering thermal runaway. Most battery-grade specifications set limits of < 1–10 ppm for transition metals in cathode active materials.
Proving You Measure What You Claim
Specificity demonstrates that the signal measured is exclusively from the target analyte — not from the matrix, spectral neighbours, or instrument artefacts. Battery matrices are chemically aggressive (concentrated acids, high dissolved solids from digested oxides) and spectrally crowded.
ICP-OES Interferences
- → Ni 231.604 nm overlapped by Co lines in NMC matrices — use Ni 221.648 nm or apply inter-element correction (IEC)
- → Li 670.784 nm self-absorbed at high concentration — dilute to < 100 mg/L or use Li 610.365 nm
- → Fe 238.204 nm overlapped by Co 238.892 nm wing — validate with Co-only solution to quantify contribution
- → Al 396.153 nm in LFP matrices — strong Fe background requires careful wavelength selection or IEC
ICP-MS Interferences
- → ⁵⁶Fe⁺ masked by ⁴⁰Ar¹⁶O⁺ — use He-KED mode or H₂ reaction mode (m/z 56)
- → ⁶³Cu⁺ interfered by ⁴⁰Ar²³Na⁺ from Na-rich electrolyte matrices — He-KED effective
- → ⁵²Cr⁺ masked by ⁴⁰Ar¹²C⁺ from organic carbon in electrolyte digest — He-KED or NH₃ reaction mode
- → ⁵¹V⁺ interfered by ³⁵Cl¹⁶O⁺ when HCl is used in digestion — avoid HCl or use DRC
How to Demonstrate Specificity
- 1Prepare a reagent blank (matrix acid without sample) and a method blank (full digestion of empty vessel).
- 2Prepare a synthetic matrix solution containing the major elements (Li, Ni, Co, Mn) at expected concentrations — without the trace impurity analytes.
- 3Measure the blank signal at each analyte wavelength/mass. Blank contribution must be < LOD.
- 4For ICP-OES: scan the wavelength window (±0.05 nm) of each analyte line with the matrix-only solution. Identify and document any wing overlap from major element emission lines.
- 5For ICP-MS: run interference check standards (ICS) in He-KED and no-gas mode. Compare recoveries — if the ratio deviates > 20%, a polyatomic is present and KED or DRC mode is required.
- 6Assess matrix effects: compare sensitivity (slope) of matrix-matched calibration vs. aqueous calibration. If > 10% difference, use matrix matching or internal standards.
Fe impurity by ICP-MS (m/z 56) in digested NMC 811: Reagent blank (2% HNO₃) = 45 cps at m/z 56 in He-KED mode. Matrix blank (Li + Ni + Co + Mn at 50 mg/L each, no Fe) = 62 cps. LOD (3σ blank) = 180 cps → both blanks well below LOD. ArO⁺ suppression confirmed: no-gas/He-KED ratio for Fe ICS = 48,200/2,310 = 20.9× → KED mode essential. Matrix effect: Fe sensitivity in NMC digest vs. aqueous = −4.8% → acceptable (< 10%) with Sc internal standard correction.
LOD criterion (blank must be below LOD):
σ_blank (10 replicate blank injections at m/z 56, He-KED) = 60 cps
LOD = 3 × σ_blank = 3 × 60 = 180 cps
Reagent blank signal = 45 cps → 45 < 180 ✓
Matrix blank signal = 62 cps → 62 < 180 ✓
ArO⁺ polyatomic interference check (no-gas vs. He-KED):
Fe ICS (1 µg/L) in no-gas mode = 48,200 cps
Fe ICS (1 µg/L) in He-KED mode = 2,310 cps
Ratio = 48,200 / 2,310 = 20.9×
Interpretation: 95% of the no-gas signal was ArO⁺ polyatomic
→ He-KED mode required for Fe at m/z 56 ✓
Matrix suppression (%):
Sensitivity_aqueous (slope) = 2,310 cps/(µg/L)
Sensitivity_NMC_matrix = 2,199 cps/(µg/L)
Effect = (2,199 − 2,310) / 2,310 × 100
= −111 / 2,310 × 100 = −4.8%
|−4.8%| < 10% threshold ✓
Internal Standard (⁴⁵Sc) recovery:
Expected Sc signal = 125,000 cps
Measured Sc in NMC matrix = 119,375 cps
IS recovery = (119,375 / 125,000) × 100 = 95.5%
85% ≤ 95.5% ≤ 115% ✓
Calibration Integrity
Linearity proves that the detector response is directly proportional to analyte concentration across the working range. Battery methods often span very wide dynamic ranges — from ppb impurities to percent-level major elements.
Main Element Assay (ICP-OES)
Range: 50% to 150% of nominal (e.g., for Ni in NMC 811: ~24–72 wt%)
Levels: Minimum 5 points: 50%, 75%, 100%, 125%, 150% of target
Acceptance: r ≥ 0.999, residuals within ±1.5%, y-intercept not significantly different from zero
Watch for: Self-absorption at high concentration (especially Li, Na, K alkali metals)
Trace Impurity (ICP-MS)
Range: LOQ to 200% of specification limit
Levels: Minimum 5 points: LOQ, 25%, 50%, 100%, 200% of spec
Acceptance: r ≥ 0.999 (for quantitative impurity methods)
Watch for: Dead time correction at high count rates; mass calibration drift affecting low-mass isotopes (Li, B)
Dataset (Y internal-standard-normalised intensity vs. Ni concentration):
x (mg/L): 50.0 75.0 100.0 125.0 150.0
y (IS-norm): 0.482 0.724 0.968 1.209 1.452
Means:
x̄ = (50.0 + 75.0 + 100.0 + 125.0 + 150.0) / 5 = 100.0 mg/L
ȳ = (0.482 + 0.724 + 0.968 + 1.209 + 1.452) / 5 = 0.967
Slope b = Σ(xi − x̄)(yi − ȳ) / Σ(xi − x̄)²:
Σ(xi − x̄)² = (−50)² + (−25)² + 0² + 25² + 50² = 6,250
Σ(xi−x̄)(yi−ȳ) = (−50)(−0.485) + (−25)(−0.243) + 0
+ (25)(0.242) + (50)(0.485)
= 24.250 + 6.075 + 0 + 6.050 + 24.250 = 60.625
b = 60.625 / 6,250 = 0.009700 ✓
Intercept a = ȳ − b · x̄:
a = 0.967 − 0.009700 × 100.0 = 0.967 − 0.970 = −0.003
y-intercept not significantly ≠ 0 (t-test: p = 0.82) ✓
Pearson r:
Σ(yi − ȳ)² = (−0.485)² + (−0.243)² + (0.001)²
+ (0.242)² + (0.485)² = 0.588
r = 60.625 / √(6,250 × 0.588)
= 60.625 / √3,675 = 60.625 / 60.622 = 0.99998 ✓
Residual check at x = 50.0 mg/L:
ŷ = 0.009700 × 50.0 + (−0.003) = 0.482
Residual = (0.482 − 0.482) / 0.482 × 100 = 0.0%
Max residual across all points: +0.4% (at 75 mg/L) within ±1.5% ✓
Regression equation:
y = 0.009700x − 0.003
r = 0.99998, r² = 0.99996 → PASS
Spike Recovery & Reference Materials
Accuracy measures how close the result is to the true value. For battery materials, two approaches are used: spike recovery (adding known amounts of analyte to the sample matrix) and certified reference materials (CRMs) when available.
Main Element Assay
Approach: Spike at 80%, 100%, 120% of nominal into digested matrix. 3 reps × 3 levels = 9 measurements.
Acceptance: Recovery 98.0%–102.0% per level; RSD ≤ 2.0%.
Alternative: Analyse a CRM (e.g., NIST SRM for similar oxide material) — result must fall within certified uncertainty.
Trace Impurity
Approach: Spike at LOQ, 100%, 200% of spec limit into matrix blank. 3 reps × 3 levels = 9 measurements.
Acceptance: Recovery 80%–120% (at LOQ), 90%–110% (at spec level).
Critical: Digest the CRM alongside real samples — incomplete dissolution biases accuracy low (especially for refractory oxides like Al₂O₃ in NMC coatings).
Spec limit: 10 ppm Fe in cathode powder
Spike levels: LOQ (0.5 ppm), 100% (10 ppm), 200% (20 ppm)
LOQ level (n=3): 0.48, 0.51, 0.49 ppm → Mean 0.493 → Recovery 98.7%, RSD 3.1%
100% level (n=3): 9.94, 10.08, 10.12 ppm → Mean 10.05 → Recovery 100.5%, RSD 0.9%
200% level (n=3): 19.87, 20.14, 20.06 ppm → Mean 20.02 → Recovery 100.1%, RSD 0.7%
Overall: Mean recovery 99.8%, Overall RSD 0.9% → PASS
Formula: % Recovery = (C_found / C_spiked) × 100
Level 1 — LOQ (C_spiked = 0.500 ppm):
Rep 1: (0.480 / 0.500) × 100 = 96.0%
Rep 2: (0.510 / 0.500) × 100 = 102.0%
Rep 3: (0.490 / 0.500) × 100 = 98.0%
Mean₁ = (96.0 + 102.0 + 98.0) / 3 = 98.67% → 98.7%
SD₁ = √[((−2.67)² + (3.33)² + (−0.67)²) / 2]
= √[(7.13 + 11.09 + 0.45) / 2] = √9.33 = 3.055
RSD₁ = (3.055 / 98.67) × 100 = 3.1% ✓ (≤ 10% at LOQ)
Level 2 — 100% of spec (C_spiked = 10.00 ppm):
Rep 1: (9.94 / 10.00) × 100 = 99.4%
Rep 2: (10.08 / 10.00) × 100 = 100.8%
Rep 3: (10.12 / 10.00) × 100 = 101.2%
Mean₂ = (99.4 + 100.8 + 101.2) / 3 = 100.47% → 100.5%
RSD₂ = 0.9% ✓ (≤ 5% at spec level)
Level 3 — 200% of spec (C_spiked = 20.00 ppm):
Values: 99.4%, 100.7%, 100.3%
Mean₃ = 100.13% → 100.1%; RSD₃ = 0.7% ✓
Overall (n = 9):
Grand mean = (98.67 + 100.47 + 100.13) / 3 = 99.76% → 99.8%
Overall SD = 0.893
Overall RSD = (0.893 / 99.76) × 100 = 0.9% → PASS ✓
Repeatability & Intermediate Precision
Precision measures how closely repeated measurements agree with each other. For battery material testing labs, the main sources of variability are sample digestion (weighing, acid volume, microwave program), dilution, and instrument drift.
Repeatability
Same day, same analyst, same instrument
6 independent sample preparations from the same lot
Main elements: RSD ≤ 2.0%
Trace impurities: RSD ≤ 5.0% (at spec level); ≤ 10% (at LOQ)
Intermediate Precision
Different days, analysts, or instruments
6 reps × 2 conditions = 12 results
Main elements: Combined RSD ≤ 3.0%
Trace impurities: Combined RSD ≤ 10%
Reproducibility
Inter-laboratory (if transferring method)
Each lab runs 6 reps independently
Compare means with t-test or equivalence test
Required for method transfers between supplier and OEM labs
Values (wt%): 48.12 48.08 48.19 48.04 48.11 48.06
Mean: 48.10 wt%
SD: 0.053 wt%
RSD: 0.11% → well within ≤ 2.0% criterion
Formulas: SD = √[Σ(xi − x̄)² / (n − 1)]
RSD (%) = (SD / x̄) × 100
Repeatability — Day 1, Analyst A (n = 6):
Values (wt%): 48.12 48.08 48.19 48.04 48.11 48.06
x̄ = 288.60 / 6 = 48.100 wt%
Deviations²: (+0.02)² + (−0.02)² + (+0.09)² + (−0.06)²
+ (+0.01)² + (−0.04)²
= 0.0004 + 0.0004 + 0.0081 + 0.0036
+ 0.0001 + 0.0016 = 0.0142
SD = √(0.0142 / 5) = √0.00284 = 0.0533 wt%
RSD = (0.0533 / 48.100) × 100 = 0.11% ✓ (≤ 2.0%)
Intermediate Precision — Day 3, Analyst B (n = 6):
Values (wt%): 48.05 48.14 48.09 48.18 48.02 48.12
x̄ = 48.100 wt%; SD = 0.0602 wt%; RSD = 0.13%
Combined Intermediate Precision (n = 12):
Grand mean x̄_G = (6 × 48.100 + 6 × 48.100) / 12 = 48.100 wt%
SS_A = (n−1) × SD_A² = 5 × 0.0533² = 0.01420
SS_B = (n−1) × SD_B² = 5 × 0.0602² = 0.01812
SS_between = 6 × (48.100−48.100)² + 6 × (48.100−48.100)² = 0
Pooled variance = (0.01420 + 0.01812 + 0) / 11 = 0.002938
Combined SD = √0.002938 = 0.0542 wt%
Combined RSD = (0.0542 / 48.100) × 100 = 0.11% ✓ (≤ 3.0%)
Detection & Quantitation Limits
For impurity methods, LOD and LOQ define the lowest concentrations detectable and quantifiable with confidence. In battery QC, the LOQ must be well below the specification limit — typically ≤ 10% of the spec to provide sufficient dynamic range.
Method 1 — Signal-to-Noise
LOD: S/N ≥ 3:1 (detect the signal above the noise)
LOQ: S/N ≥ 10:1 (quantify with acceptable precision)
Prepare serial dilutions from 100% spec → sub-ppb. Measure S/N at each level.
Best for ICP-MS where peak/background separation is clear.
Method 2 — Calibration Residuals
LOD = 3.3 × σ / S
LOQ = 10 × σ / S
Where σ = standard deviation of residuals from regression, S = slope.
Practical for ICP-OES where peak/background measurement is less defined.
LOQ Confirmation (Required)
The estimated LOQ must be confirmed experimentally: prepare 6 independent replicates at the LOQ concentration. Criteria: RSD ≤ 10% and recovery 80–120%. If the LOQ fails confirmation, raise it and re-test.
Example — Cu by ICP-MS in digested NMC 811 (spec: 5 ppm):
Estimated LOQ (S/N = 10:1): 0.05 ppm (1% of spec) ✓
Confirmation (n=6 at 0.05 ppm): Mean 0.049 ppm, RSD 4.2%, Recovery 98.0% → PASS
Method 1 — Signal-to-Noise:
RMS baseline noise (10-point blank region) = 8.4 cps
Formula: S/N = peak height / (2 × RMS noise)
At 0.02 µg/L: peak height = 50.4 cps
S/N = 50.4 / (2 × 8.4) = 50.4 / 16.8 = 3.0 → LOD ✓
At 0.05 µg/L: peak height = 168.0 cps
S/N = 168.0 / 16.8 = 10.0 → LOQ ✓
Method 2 — Calibration residuals (cross-check):
σ_residuals = 18.5 cps; Slope S = 3,360 cps/(µg/L)
LOD = (3.3 × 18.5) / 3,360 = 61.05 / 3,360 = 0.018 µg/L ≈ 0.02 µg/L ✓
LOQ = (10 × 18.5) / 3,360 = 185.0 / 3,360 = 0.055 µg/L ≈ 0.05 µg/L ✓
(consistent with S/N approach)
LOQ confirmation (n = 6 at 0.05 µg/L in NMC matrix):
Values (µg/L): 0.048 0.051 0.049 0.052 0.047 0.050
Mean = 0.0495 µg/L
SD = 0.00187 µg/L
RSD = (0.00187 / 0.0495) × 100 = 3.8% < 10% ✓
Recovery = (0.0495 / 0.0500) × 100 = 99.0% within 80%–120% ✓
LOQ in sample units:
LOQ_solution = 0.05 µg/L
Sample weight = 0.100 g; Final volume = 50 mL
Dilution factor = 50 / 0.100 = 500×
LOQ_sample = 0.05 × 500 / 1000 = 0.025 ppm
Spec limit = 5 ppm → LOQ is 0.5% of spec ✓ (well below 10%)
Validated Operating Window
The range is the interval where linearity, accuracy, and precision have all been demonstrated simultaneously. It defines the valid reporting window — any result outside this range requires dilution or concentration before reporting.
Main Element Assay
80%–120% of nominal concentration
Example: Ni in NMC 811 (nominal ~48 wt%) → validated from 38.4 to 57.6 wt%
Must pass linearity, accuracy (98–102%), and precision (RSD ≤ 2%) at both boundaries
Trace Impurity
LOQ to 200% of spec limit
Example: Fe spec ≤ 10 ppm → validated from 0.5 ppm (LOQ) to 20 ppm
Must pass linearity, accuracy (80–120% at LOQ; 90–110% at spec), and precision at boundaries
Range accepted when linearity + accuracy + precision all pass at each boundary.
Lower boundary — 0.5 ppm (= confirmed LOQ):
Linearity: predicted ŷ = 0.009700 × 0.5 + (−0.003) = 0.00185
measured = 0.00188 → residual +1.6% within ±2% ✓
Accuracy: Spike recovery (n=3) = 98.7% within 80%–120% ✓
Precision: RSD (n=6 at 0.5 ppm) = 3.1% < 10% (LOQ criterion) ✓
Upper boundary — 20.0 ppm (= 200% of 10 ppm spec):
Linearity: residual = +0.3% within ±2% ✓
Detector: IS-norm signal at 20 ppm not saturated (pulse counting mode) ✓
Accuracy: Spike recovery (n=3) = 100.1% within 90%–110% ✓
Precision: RSD (n=6 at 20 ppm) = 0.7% < 5.0% ✓
Accepted range:
0.5 ppm (LOQ) to 20.0 ppm (200% spec)
All three parameters pass at both boundaries ✓
Stress-Testing the Method
Robustness evaluates whether the method withstands small, deliberate variations in operating parameters — the kind of drift that happens in real-world lab conditions. For battery material analysis, the critical variables span both sample preparation and instrument tuning.
Sample Preparation Factors
- → Digestion temperature ± 10 °C
- → Digestion time ± 5 min
- → Acid ratio (HNO₃:HCl) ± 10%
- → Sample weight ± 5 mg
- → Final dilution volume ± 1 mL
Instrument Factors
- → RF power ± 50 W
- → Nebulizer gas flow ± 0.05 L/min
- → Spray chamber temperature ± 2 °C
- → Plasma viewing height ± 1 mm (OES radial)
- → He-KED flow ± 0.5 mL/min (MS)
Recommended Approach: Plackett-Burman Screening
For 7 factors, a Plackett-Burman design requires only 8 experimental runs (vs. 128 for full factorial). Each factor is tested at its high (+) and low (−) level. The main effect of each factor is calculated — any factor with |effect| > 2% relative is flagged as critical and requires tighter SOP control.
Acceptance: No single factor shifts the result by > ±2% relative. Internal standard recovery stays within 85–115% across all runs.
Factor assignments:
A = RF power (±50 W) B = Nebulizer flow (±0.05 L/min)
C = Spray chamber T (±2 °C) D = Digestion temp (±10 °C)
E = Digestion time (±5 min) F = Sample weight (±5 mg)
G = Viewing height (±1 mm)
Ni assay result per run (wt%):
Run: 1(+) 2(+) 3(+) 4(+) 5(−) 6(−) 7(−) 8(−) [A level]
wt%: 48.02 48.08 47.98 48.05 48.42 48.38 48.35 48.45
Main effect of A — RF power:
Mean at A(+1600W) = (48.02+48.08+47.98+48.05) / 4 = 192.13/4 = 48.033%
Mean at A(−1500W) = (48.42+48.38+48.35+48.45) / 4 = 193.60/4 = 48.400%
Effect_A = 48.033 − 48.400 = −0.367 wt%
% shift = (−0.367 / 48.22) × 100 = −0.76%
|−0.76%| < 2.0% tolerance ✓
Main effect of D — Digestion temperature:
Mean at D(+) = 48.18%; Mean at D(−) = 48.25%
Effect_D = −0.07 wt% → −0.15% relative ✓ (not significant)
All other factors (B, C, E, F, G): |effect| < 0.3% relative ✓
IS (⁴⁵Sc) recovery across all 8 runs:
Values (%): 96.2 97.1 95.8 96.9 98.4 97.6 99.1 98.2
Range: 95.8%–99.1%; all within 85%–115% ✓
Conclusion:
No factor exceeds ±2% tolerance → method is ROBUST
RF power is the largest contributor (−0.76%)
→ SOP specifies RF power as 1550 ± 25 W (tightened from ± 50 W)
Quick Reference — Acceptance Criteria
| Parameter | Main Element Assay | Trace Impurity |
|---|---|---|
| Specificity | Blank < LOD, matrix effect < 10% | Blank < LOD, ICS recovery 80–120% |
| Linearity | r ≥ 0.999, residuals ±1.5% | r ≥ 0.999, residuals ±2% |
| Accuracy | Recovery 98–102% | 80–120% (LOQ), 90–110% (spec) |
| Repeatability | RSD ≤ 2.0% | RSD ≤ 5.0% (spec); ≤ 10% (LOQ) |
| Intermediate Precision | Combined RSD ≤ 3.0% | Combined RSD ≤ 10% |
| LOD / LOQ | N/A (assay method) | LOQ ≤ 10% of spec, confirmed n=6 |
| Range | 80–120% of nominal | LOQ to 200% of spec |
| Robustness | No factor shifts result > ±2% relative; IS recovery 85–115% | |
A validated method is not a one-time exercise — it's a living document. Revalidation is triggered by changes in sample matrix (new cathode chemistry), instrument platform (OES → MS), or specification limits. Ongoing system suitability (CCV, ISTD recovery, calibration verification) provides the daily evidence that the method continues to perform as validated.