Interface engineering in garnet-type solid-state batteries: mitigating lithium dendrite penetration

Abstract

Garnet-type Li7La3Zr2O12 solid electrolytes combine high lithium-ion conductivity, oxidative stability, and mechanical stiffness, but lithium penetration and interfacial failure still limit practical solid-state lithium-metal batteries. We report a multilayer interface-engineering strategy for Ta-doped LLZO pellets and thin separators that combines dry-polished surface preparation, a 12 nm lithiophilic Al2O3-Li3PO4 conversion layer, and a 1.5 um compliant Li-Mg wetting interlayer. Symmetric Li|LLZO|Li cells were tested under controlled stack pressure from 2 to 12 MPa, with operando acoustic emission, impedance spectroscopy, post-mortem X-ray tomography, and cryo-focused-ion-beam cross sections. Compared with untreated LLZO, the engineered interface reduced room-temperature area-specific resistance from 312 to 18 ohm cm2, increased the median critical current density from 0.42 to 1.72 mA cm-2 at 6 MPa, and delayed short-circuit onset during 0.5 mAh cm-2 cycling from 46 to 410 h. The improvement came from three coupled effects: removal of air-formed carbonate/hydroxide species, uniform lithium wetting that eliminated interfacial current constriction, and a compliant interlayer that reduced stress concentration during stripping. Failure was not eliminated. At current densities above 2.2 mA cm-2, tomography showed lithium penetration along pre-existing polishing cracks and high-angle grain boundaries. Thin full cells with LiNi0.8Mn0.1Co0.1O2 composite cathodes retained 83% capacity after 180 cycles at 0.5 C, but impedance growth accelerated when lithium inventory became locally depleted. These results suggest that garnet interface engineering should be evaluated with coupled electrochemical, mechanical, and imaging metrics rather than with a single headline critical current density.

Introduction

Solid-state lithium-metal batteries promise high energy density and improved abuse tolerance, but the electrolyte-lithium interface remains a central obstacle. Garnet-type Li7La3Zr2O12, commonly abbreviated LLZO, is one of the most studied oxide solid electrolytes because it can combine high lithium-ion conductivity with chemical robustness and a wide electrochemical window [1,2,3]. The discovery of fast lithium conduction in garnet LLZO made it a serious battery electrolyte candidate rather than a crystallographic curiosity [1].

The hard part is not only ionic conductivity. A dense garnet pellet with good bulk conductivity can still fail by high interfacial impedance, poor lithium wetting, void formation during stripping, electronic leakage, grain-boundary penetration, or fracture. Early interface studies showed that apparent Li/LLZO resistance is strongly affected by air exposure, surface chemistry, and contact condition [10,11,12,13]. Dendrite-like lithium penetration through inorganic solid electrolytes then made clear that high shear modulus alone does not guarantee safety [5].

The literature now points to several coupled failure mechanisms. Electronic conductivity within the solid electrolyte can support internal lithium deposition [6]. Stripping can create voids that concentrate current during subsequent plating [7]. Microstructure, dopants, grain boundaries, and residual flaws influence critical current density [8,9]. Direct visualisation studies show that lithium can propagate through solid electrolytes under large local stresses and along defects [14,15,16,17]. Interface chemistry can improve performance, but only when it is tested under realistic pressure, current density, areal capacity, and cycling history [18,19,20].

This article reports a fictional but realistic interface-engineering study for Ta-doped LLZO. The goal is not to claim dendrite immunity. It is to identify which interfacial changes reduce current constriction and stress concentration, quantify how much the critical current density improves, and show where failure still initiates.

Materials and surface preparation

Ta-doped LLZO pellets were prepared by solid-state reaction, hot pressing, and oxygen annealing. The nominal composition was Li6.5La3Zr1.5Ta0.5O12 with 10 mol% excess lithium precursor to compensate for volatilisation. Pellets were 12.0 mm in diameter and 620 +/- 35 um thick after polishing. Relative density measured by Archimedes method was 97.8 +/- 0.6%. The room-temperature bulk ionic conductivity from blocking-electrode impedance was 0.78 +/- 0.06 mS cm-1.

The surface preparation protocol was deliberately strict. Pellets were dry polished from 1200 grit to 0.05 um alumina slurry under argon, rinsed with anhydrous dimethyl carbonate, and transferred to the coating chamber without air exposure. Control pellets were polished identically but exposed to laboratory air for 24 h before cell assembly. XPS and time-of-flight secondary-ion mass spectrometry were used to quantify surface carbonate, hydroxide, aluminium residue, and lithium depletion.

The engineered interface used two layers. First, a nominal 12 nm Al2O3-Li3PO4 film was deposited by sequential atomic layer deposition and converted by contact with molten lithium at 180 deg C. Second, a 1.5 um Li-Mg alloy foil was pressed onto each side of the LLZO as a compliant wetting layer. The alloy contains 4 at.% Mg, enough to alter wetting and mechanical response without becoming a separate high-resistance electrode.

Four comparison groups were tested: untreated air-exposed LLZO, dry-polished argon-handled LLZO, ALD-coated LLZO without Li-Mg interlayer, and the full engineered stack. This design separates surface cleaning, chemical conversion, and mechanical compliance.

Cell assembly and diagnostics

Symmetric Li|LLZO|Li cells were assembled in stainless-steel fixtures with calibrated stack pressure of 2, 6, or 12 MPa. Lithium electrodes were 100 um thick for baseline tests and 20 um for selected inventory-limited tests. Electrochemical impedance spectroscopy was measured from 7 MHz to 0.1 Hz at open circuit and after selected cycling intervals. Galvanostatic cycling used areal capacities of 0.1, 0.25, and 0.5 mAh cm-2 per half-cycle.

Critical current density was measured using a step protocol, but we treat it as a distribution rather than as a single material constant. Current density was increased every five cycles until voltage noise, impedance jump, or hard short appeared. The protocol follows recent calls to standardise critical-current measurements in lithium garnets [19]. Cells that survived the final step for less than ten cycles were counted as failed at that current.

Operando acoustic emission sensors were mounted on the compression fixture to detect cracking or abrupt interfacial events. The acoustic system was not sensitive enough to image crack location, but it identified event clusters preceding hard shorts. X-ray microtomography was performed before cycling and after failure using 0.7 um voxel size. Selected regions were then cross-sectioned by cryo-FIB/SEM to preserve lithium morphology.

Full cells used an NMC811 composite cathode containing 70 wt% active material, 20 wt% LLZO powder, 5 wt% carbon, and 5 wt% binder, pressed against 80 um LLZO separators. The cathode loading was 2.1 mAh cm-2. These cells test compatibility with a practical cathode composite but are not presented as optimised high-energy prototypes.

Interfacial chemistry

Air-exposed LLZO showed strong Li2CO3 and LiOH signatures in XPS. The C 1s carbonate peak and O 1s hydroxide shoulder were reduced by dry polishing under argon and were nearly absent after ALD transfer. This agrees with prior work showing that surface chemistry and air exposure strongly affect Li/LLZO interfacial resistance [10,11,13].

After contact with molten lithium, the Al2O3-Li3PO4 layer converted into a mixed Li-Al-O and Li-P-O interphase. Depth profiling indicated that the layer remained thinner than 35 nm after cycling at 60 deg C. It did not behave as a thick polymer-like buffer; its role was to promote wetting and create a chemically stable contact. The Li-Mg interlayer filled residual polishing grooves and reduced contact voids visible in cross-section.

The untreated air-exposed interface had discontinuous lithium contact covering only 61 +/- 8% of the apparent area after stack assembly. Dry polishing increased contact coverage to 79 +/- 5%. The full engineered interface reached 96 +/- 3% coverage. This contact uniformity is crucial because local current density scales inversely with active contact area. A nominal 0.5 mA cm-2 test can locally exceed 2 mA cm-2 if contact is patchy.

The interface chemistry changed little after 100 h of low-current cycling, but high-current failed cells showed reduced phosphorus signal and local aluminium enrichment near lithium filaments. We interpret that change as a consequence of failure rather than its initial cause.

Impedance and wetting

Room-temperature area-specific resistance fell from 312 ohm cm2 for air-exposed LLZO to 96 ohm cm2 for dry-polished argon-handled LLZO, 41 ohm cm2 for ALD-coated LLZO, and 18 ohm cm2 for the full engineered interface. At 60 deg C, the full-interface resistance was 7 ohm cm2. These values are in the range expected when carbonate contamination and poor lithium wetting are removed [4,10,12,13].

Pressure dependence revealed two regimes. Untreated cells improved strongly as pressure increased from 2 to 12 MPa, indicating that mechanical contact was limiting. Engineered cells improved only modestly above 6 MPa, suggesting that the Li-Mg layer already established conformal contact. Excessive pressure was not beneficial: at 12 MPa, several cells failed earlier during stripping, consistent with pressure-enhanced stress at flaws.

Wetting tests with molten lithium droplets on treated LLZO showed contact angle decreasing from 132 deg for air-exposed surfaces to 38 deg after ALD conversion and Li-Mg contact. The measurement is crude because molten lithium reacts with the surface, but it correlates with impedance and cross-sectional contact. Interface engineering should therefore be read as a combined chemical and mechanical treatment, not as a single coating effect.

The impedance semicircle remained stable for 300 h at open circuit. This indicates that the engineered interphase is not rapidly consumed by lithium under rest conditions. Cycling under current is a more severe test because it creates local flux, voids, and stress.

Critical current density

The untreated air-exposed cells had median critical current density of 0.42 mA cm-2 at 6 MPa and 25 deg C. Dry-polished cells reached 0.76 mA cm-2. ALD-coated cells reached 1.18 mA cm-2, and full engineered cells reached 1.72 mA cm-2. At 60 deg C, the full engineered cells reached 2.35 mA cm-2, but with wider scatter because higher temperature accelerated lithium redistribution after stripping.

The critical current density was not monotonic with stack pressure. At 2 MPa, contact was incomplete and voltage noise appeared early. At 6 MPa, contact and mechanical stability were balanced. At 12 MPa, the median critical current density fell to 1.41 mA cm-2 because failure initiated at pre-existing surface flaws under higher compressive stress. This pressure dependence supports a mechanical contribution to failure, consistent with lithium penetration and stripping-instability studies [5,7,21].

Dopant and microstructure effects were visible even within the same interface treatment. Pellets with larger abnormal grains failed earlier when a grain boundary connected the two electrodes through the thickness. This aligns with work showing that dopants and microstructure affect critical current density in garnet electrolytes [8,9]. Interface engineering cannot fully compensate for a defective ceramic body.

We report critical current density as a median with interquartile range rather than as a maximum surviving cell. The best full-interface cell survived 2.6 mA cm-2 at 60 deg C, but using that value as a headline would overstate reproducibility.

Cycling stability

At 0.25 mA cm-2 and 0.25 mAh cm-2 per half-cycle, full engineered symmetric cells cycled for 1,200 h without hard short at 25 deg C. Voltage hysteresis stabilised at 38-45 mV after formation. Untreated cells shorted or developed erratic voltage within 140 h. At 0.5 mA cm-2 and 0.5 mAh cm-2, engineered cells had median lifetime of 410 h, compared with 46 h for untreated cells.

Failure was usually preceded by increased high-frequency impedance and intermittent voltage spikes during stripping. This is consistent with void formation at the lithium interface followed by localised plating during the next half-cycle, the mechanism emphasised in critical-stripping-current work [7]. The Li-Mg layer delayed but did not eliminate this sequence.

Increasing areal capacity from 0.25 to 0.5 mAh cm-2 reduced lifetime more strongly than doubling current at fixed capacity. That result highlights the role of lithium inventory and stripping depth. A cell can pass short critical-current pulses and still fail during deeper cycling. Interface qualification should therefore include current density, areal capacity, pressure, and cycle count.

Full cells with NMC811 composite cathodes retained 83% capacity after 180 cycles at 0.5 C and 60 deg C. The average coulombic efficiency after formation was 99.2%. Impedance growth was dominated by the lithium side for the first 120 cycles and by cathode composite resistance thereafter. Cathode optimisation remains outside the scope of this paper.

Imaging of failure paths

X-ray tomography of failed untreated cells showed bright lithium-filled cracks crossing part or all of the electrolyte thickness. In several cases, the crack initiated at a surface pit left by polishing. Full engineered cells that failed above 2.2 mA cm-2 showed fewer but sharper penetration paths, usually along high-angle grain boundaries or pre-existing microcracks. The interface treatment reduced current constriction but did not heal structural defects.

Cryo-FIB cross sections showed lithium penetrating along grain-boundary regions enriched in aluminium and silicon impurities in two failed pellets. These observations are consistent with reports that lithium can exploit grain-boundary softening, local electronic conduction, or stress concentration [5,6,14,15]. We cannot assign a single universal mechanism from post-mortem images alone.

Operando acoustic emission detected event clusters 10-80 min before hard short in 11 of 15 failed high-current cells. The events were absent during low-current cycling of surviving cells. We interpret them as crack extension or sudden interfacial delamination. The acoustic signatures were not unique enough for automated failure prediction, but they provided useful timing for interrupted tomography.

The most important imaging result is negative: cells with low interfacial resistance still failed if the ceramic contained connected defects. This argues against relying on impedance alone as a dendrite-resistance metric.

Mechanistic interpretation

The engineered interface improves performance by reducing three failure drivers. First, removing carbonate and hydroxide species lowers interfacial resistance and increases true contact area. Second, the lithiophilic conversion layer and Li-Mg interlayer reduce wetting heterogeneity, lowering local current constriction. Third, the compliant layer buffers small stripping-induced gaps and reduces stress concentration at the ceramic surface.

The improvement does not contradict lithium penetration studies in stiff inorganic electrolytes [5]. Those studies showed that defects, current focusing, and fracture can defeat the simple modulus argument. Our results support that view. The interface can move the operating point away from catastrophic focusing, but once current density and areal capacity are high enough, a flaw-controlled penetration path still wins.

Electronic conductivity remains a concern. The measured DC electronic conductivity of the Ta-doped LLZO was below 10^-9 S cm-1 at room temperature, but local electronic pathways at reduced grain boundaries could still support nucleation under high overpotential. The high-electronic-conductivity mechanism proposed for solid electrolytes [6] is therefore best read as a local risk factor rather than as a bulk-only property.

Mechanical deformation also matters. Monroe and Newman showed that elastic effects influence lithium deposition stability at solid interfaces [21]. In garnets, the problem is amplified by ceramic flaws and imperfect contact. A good interface is not the stiffest possible interface; it is one that distributes current and stress without sacrificing chemical stability.

Comparison with recent interface strategies

Recent garnet interface strategies include ultraclean surfaces, alloy wetting layers, transformed interphases, graphite-based contact layers, and ultrathin lithium inventory control [13,18,20,22]. The present multilayer approach is not meant to be the only answer. Its value is that each contribution is separated experimentally: cleaning, chemical conversion, and compliance all help, and none alone is sufficient.

The 2023 transformed-interface chemistry work is especially relevant because it treats the electrolyte surface as a reactive volume rather than as a passive plane [18]. Our ALD conversion layer is thinner and simpler, but the same principle applies: the interphase must control lithium activity and wetting throughout the contact region.

The 2024 thin-lithium-electrode work shows another route: limiting lithium inventory can reduce the severity of filament propagation and improve practical cell design [20]. Our inventory-limited tests agree qualitatively. Thin lithium reduces catastrophic excess but makes stripping voids more important, so pressure and compliance become even more critical.

No single metric captures these trade-offs. A coating can reduce initial impedance but fail under deep stripping. A high critical current can be obtained with shallow pulses but disappear under larger areal capacity. A smooth pellet can pass symmetric-cell tests but fail in full cells because cathode-side pressure and lithium inventory evolve differently.

Limitations

The first limitation is scale. Most symmetric-cell tests used pellets hundreds of micrometres thick, whereas practical solid-state batteries need thin separators. Thinner LLZO reduces ohmic loss but can increase defect sensitivity and handling damage. The thin-separator full cells here are informative but not yet manufacturable pouch cells.

The second limitation is cathode integration. The NMC811 composite cathodes were intentionally simple and contain excess LLZO powder. High-loading cathodes require percolating ion and electron networks, stable high-voltage interfaces, and controlled stack pressure. Lithium-side improvements alone do not guarantee full-cell success.

The third limitation is diagnostics. X-ray tomography and cryo-FIB reveal failure paths after or near failure, but they cannot track every lithium filament from nucleation. Visualisation studies show the value of direct observation [15,17], and future work should combine optical, acoustic, and X-ray approaches in the same cell geometry.

Finally, the study uses one Ta-doped LLZO composition. Other dopants, grain-boundary chemistries, and sintering histories may respond differently. Interface engineering and ceramic processing must be co-optimised.

Conclusion

A multilayer interface combining dry-polished LLZO, an Al2O3-Li3PO4 conversion layer, and a compliant Li-Mg wetting interlayer substantially reduced Li/LLZO impedance and increased critical current density in garnet-type solid-state cells. The best improvement came from uniform contact and reduced stress concentration rather than from a magic dendrite-blocking coating. Symmetric cells cycled much longer than untreated controls, and full cells retained useful capacity over 180 cycles.

Dendrite penetration was mitigated but not eliminated. At high current density and areal capacity, lithium still followed polishing cracks and high-angle grain boundaries. The central design lesson is therefore coupled: interface chemistry, lithium inventory, stack pressure, ceramic microstructure, and diagnostic protocol must be specified together. Garnet solid-state batteries will advance through reproducible failure control, not through isolated headline critical-current numbers.

Data and code availability

The supplementary archive contains impedance spectra, galvanostatic cycling files, critical-current scripts, XPS peak fits, tomography reconstructions, cryo-FIB image stacks, acoustic-emission event tables, and full-cell cycling data. Raw tomography volumes are downsampled for web distribution, with full-resolution files available in the local supplementary bundle.