Ionic Interdiffusion at Cathode-Solid-Electrolyte Interface: A Machine Learning-Assisted Multiscale Investigation and Mitigation Strategies

This study combines machine learning-assisted multiscale simulations and continuum modeling to demonstrate that ionic interdiffusion at the LiCoO2|Li10GeP2S12 interface causes rapid capacity fade, while a LiNb0.5Ta0.5O3 interlayer effectively suppresses this diffusion but introduces a risk of delamination due to mechanical stiffness, highlighting the need for interlayers that balance low interdiffusion with low stiffness.

The Gist

The Big Picture: Building a Better Battery

Imagine you are trying to build a super-efficient battery for a future electric car. To make these batteries hold more energy and charge faster, scientists want to swap out the flammable liquid inside current batteries for a solid block of material (a Solid Electrolyte). Think of this like replacing a messy, leaky water pipe with a solid, high-tech highway for electricity.

One of the best "highways" scientists have found is a material called LGPS. However, there's a problem. When you connect this highway to the battery's positive side (the Cathode, made of a material called LCO), they don't get along. It's like trying to park a Ferrari next to a rusting truck; they start to break each other down.

The Problem: The "Chemical Meltdown"

The paper investigates what happens when the Cathode (LCO) touches the Solid Highway (LGPS).

• The Analogy: Imagine the Cathode is a house made of bricks (Cobalt atoms), and the Highway is a garden next to it. When they touch, the bricks from the house start crumbling and falling into the garden. The garden gets clogged with bricks, and the house loses its structure.
• The Science: In the battery, Cobalt atoms from the Cathode diffuse (migrate) into the LGPS electrolyte. This creates a messy, resistive layer (a "gunk" layer) between them. This gunk blocks the flow of electricity, causing the battery to lose its power very quickly, sometimes failing in the very first charge cycle.

The Proposed Fix: The "Buffer Zone"

To stop the bricks from falling into the garden, researchers tried putting a thin, protective wall between the house and the garden. This wall is made of a material called LNTO.

• The Analogy: Think of LNTO as a sturdy, high-quality fence. The researchers hoped this fence would stop the bricks (Cobalt) from leaving the house and entering the garden.
• The Result (Good News): The computer simulations showed that this fence works! The Cobalt atoms cannot easily break through the LNTO fence to get into the LGPS garden. The fence is made of strong metal-oxygen bonds that hold tight, unlike the LGPS material which is more "flexible" and lets the Cobalt slip through.

The Catch: The Fence is Too Rigid

While the LNTO fence stops the chemical mixing, the paper found a new problem: The fence is too stiff.

• The Analogy: Imagine the house (Cathode) and the garden (Electrolyte) are made of soft clay. They expand and shrink slightly when the battery charges and discharges (like breathing). The LNTO fence is made of rock-hard concrete. When the soft clay tries to move, the hard concrete doesn't bend. Eventually, the pressure causes the house to pull away from the fence, creating a gap.
• The Science: Because LNTO is mechanically very stiff, it creates stress at the interface. Over time, this stress can cause the layers to separate (delaminate). Once they separate, the battery stops working well because the electricity can't jump across the gap.

How They Studied This (The "Time Machine")

The scientists used three different tools to figure this out:

1. Super-Computer Simulations (AIMD): They ran tiny, ultra-accurate simulations of atoms. This is like watching a slow-motion video of individual bricks falling, but it's so expensive computationally that they can only watch for a few seconds.
2. Machine Learning (MLMD): They taught a computer to learn from the slow-motion video so it could predict what happens over much longer times (nanoseconds) with millions of atoms. This is like using an AI to predict the outcome of a game after watching just a few plays.
3. Continuum Modeling: They used math to scale this up to the size of a real battery (microns and hours). This is like predicting how a whole city's traffic will behave based on how one car drives.

The Final Verdict

The paper concludes that:

1. LCO + LGPS: A disaster. The materials mix, creating a "gunk" layer that kills the battery.
2. LCO + LNTO + LGPS: A partial success. The LNTO layer successfully stops the chemical mixing (the "gunk").
3. The New Problem: However, because LNTO is so stiff, it might cause the battery layers to peel apart (delaminate) over time, which also hurts performance.

The Takeaway: The paper suggests that to make the perfect battery, we need a new "fence" material that is strong enough to stop the chemical mixing but flexible enough to bend with the battery as it charges and discharges, so it doesn't peel apart.

Technical Summary

Technical Summary: Ionic Interdiffusion at Cathode|Solid-Electrolyte Interface

Problem Statement
All-solid-state lithium batteries (ASSLBs) utilizing sulfide-based solid electrolytes (SSEs), such as Li10GeP2S12 (LGPS), offer high ionic conductivity and safety but face significant interfacial instability when paired with high-voltage layered oxide cathodes like LiCoO2 (LCO). Specifically, thermodynamic instability leads to ionic interdiffusion, particularly the migration of Cobalt (Co) from the cathode into the electrolyte. This process forms a resistive interphase layer, causing rapid capacity fade and performance degradation, even within the first cycle. While experimental studies have suggested that introducing a thin LiNb0.5Ta0.5O3 (LNTO) interlayer can mitigate this, the underlying atomic-scale chemistry and the long-term mechanical implications of such protective layers remain unclear.

Methodology
This study employs a multiscale computational framework integrating three distinct levels of simulation to investigate the LCO|LGPS and LCO|LNTO interfaces:

1. Ab Initio Molecular Dynamics (AIMD): Used to generate accurate training data for interatomic potentials and to simulate initial interfacial stability. These simulations were limited to systems of a few hundred atoms and timescales of picoseconds.
2. Machine Learning Molecular Dynamics (MLMD): To overcome the computational limitations of AIMD, a Deep Learning Potential (DLP) based on a Deep Neural Network (DNN) was developed using the DeePMD-kit framework. Trained on AIMD and DFT data, this potential allowed for large-scale simulations (systems >6,000 atoms) over extended timescales (up to 3.6 nanoseconds) at 300 K. This enabled the observation of slow diffusion processes and interphase growth.
3. Continuum Modeling: Diffusion coefficients extracted from the atomistic simulations were fed into continuum models (1D and 2D frameworks). These models simulated the time-dependent growth of the interphase layer, species concentration profiles, mechanical stress evolution, and overall cell performance (voltage vs. capacity) over macroscopic timescales (hours) and length scales (microns).

Key Results

• LCO|LGPS Instability: MLMD simulations confirmed that the LCO|LGPS interface is unstable, permitting the interdiffusion of Co and O atoms. A resistive interphase layer forms and grows steadily, reaching a saturation point after approximately 2 nanoseconds. The diffusion coefficients at the interface follow the trend O > Li > P > S > Ge > Co.
• Interphase Growth Mechanism: Continuum modeling predicted that this interdiffusion leads to a rapid initial increase in interphase thickness, which slows over time due to the buildup of compressive stress that retards mass transport. The growth follows a damped diffusion mechanism (power law exponent ~0.155). The model suggests that a detrimental interphase layer exceeding 1 µm could form within 24 hours, leading to significant loss of active material and increased charge transfer resistance.
• LNTO as a Mitigation Strategy: Simulations of the LCO|LNTO interface showed no significant Co interdiffusion over 2 nanoseconds. Thermodynamic analysis revealed that substituting Li with Co in LNTO is energetically unfavorable (substitution energy = +0.144 eV) due to the rigid, stable Nb5+/Ta5+ oxide framework and charge imbalance issues. In contrast, substitution in LGPS is favorable (-1.109 eV). Additionally, the work of adhesion for LCO|LNTO (1.56 eV/Ų) is significantly higher than for LCO|LGPS (0.73 eV/Ų), indicating stronger interfacial binding.
• Mechanical Trade-offs: While LNTO prevents chemical interdiffusion, continuum analysis highlighted a mechanical drawback. LNTO possesses a higher elastic modulus (stiffness) compared to LGPS. This stiffness mismatch can generate significant tensile stresses at the interface during cycling, potentially leading to interfacial delamination. The model suggests that even with a protective layer, delamination (estimated at ~30% in the study) can degrade performance by increasing charge transfer resistance and reducing effective contact area.

Significance and Claims
The paper claims to provide a comprehensive, multiscale explanation for the experimental observations regarding LCO|LGPS instability and the partial success of LNTO interlayers.

• Mechanistic Insight: The study elucidates the atomic-scale reasons why LNTO inhibits Co diffusion (thermodynamic unfavorability of substitution and rigid framework) while LGPS does not.
• Multiscale Validation: By bridging atomistic simulations with continuum models, the work successfully correlates nanoscale diffusion events with macroscale performance metrics, such as capacity fade and voltage profiles.
• Design Guidelines: The authors conclude that while LNTO effectively mitigates ionic interdiffusion, its mechanical stiffness introduces a new failure mode (delamination). Therefore, the development of novel interlayers must balance low interdiffusion properties with low mechanical stiffness to ensure long-term stability. The paper posits that this delamination mechanism explains why experimental studies observe Co migration after long-term cycling (e.g., 300 cycles) despite the initial effectiveness of the LNTO layer.

The study does not propose new experimental protocols but rather offers a theoretical framework to interpret existing data and guide the design of future interfacial materials for solid-state batteries.