Lithium-metal batteries represent a promising next-generation energy storage technology due to their high theoretical energy density. However, their commercialization faces significant challenges related to degradation mechanisms, including solid electrolyte interphase (SEI) growth, dead lithium formation, and capacity fade. Understanding and modeling these processes is critical for improving battery performance and longevity. This article examines degradation modeling approaches for lithium-metal batteries, focusing on SEI kinetics, dead lithium accumulation, and their impact on capacity fade, while comparing empirical, physics-based, and machine learning methods.

The SEI layer forms on the lithium-metal anode due to reactions between the electrode and electrolyte. While the SEI passivates the anode, its continued growth consumes active lithium and increases cell impedance. SEI growth kinetics are influenced by electrolyte composition, current density, and operating temperature. Physics-based models often describe SEI growth using a parabolic growth law, where SEI thickness increases with the square root of time. The growth rate depends on the diffusion of reactive species through the SEI layer and the activation energy for decomposition reactions. Empirical models, in contrast, fit experimental data to power-law or exponential functions without explicitly accounting for underlying mechanisms. Machine learning approaches can identify complex correlations between SEI growth and operational parameters but require extensive datasets for training.

Dead lithium accumulation is another critical degradation mode in lithium-metal batteries. Dead lithium refers to electrically isolated lithium metal that no longer participates in electrochemical reactions. It forms due to inhomogeneous lithium deposition and stripping, leading to dendrite fragmentation and detachment. Physics-based models simulate dead lithium formation using phase-field or mesoscale models that capture lithium morphology evolution. These models incorporate parameters such as overpotential, exchange current density, and mechanical stress. Empirical models quantify dead lithium accumulation based on coulombic efficiency measurements, assuming a fixed loss per cycle. Machine learning techniques can predict dead lithium formation by analyzing voltage profiles and impedance data but may lack interpretability.

Capacity fade in lithium-metal batteries results from both SEI growth and dead lithium accumulation. SEI growth permanently consumes lithium inventory, while dead lithium reduces the electrochemically active surface area. Physics-based models couple SEI and dead lithium effects to predict capacity fade over cycling. These models solve partial differential equations for mass transport, charge transfer, and side reactions. Empirical models often use semi-empirical equations, such as the Arrhenius rate law, to describe capacity fade as a function of cycle number, temperature, and current rate. Machine learning models can integrate multiple degradation indicators but require careful validation to ensure generalization.

Comparing modeling approaches reveals trade-offs between accuracy, computational cost, and interpretability. Physics-based models provide mechanistic insights but require detailed material parameters and significant computational resources. Empirical models are computationally efficient but may not extrapolate well beyond tested conditions. Machine learning models excel at pattern recognition but need large, high-quality datasets and lack physical transparency. Hybrid approaches that combine physics-based frameworks with data-driven corrections offer a promising middle ground.

SEI growth kinetics in lithium-metal batteries differ from those in lithium-ion batteries due to the reactive nature of lithium metal. The SEI is more dynamic and prone to cracking, exposing fresh lithium to further reactions. Models must account for this self-accelerating degradation process. Dead lithium accumulation is unique to lithium-metal systems and absent in intercalation-based anodes. Its modeling requires explicit consideration of lithium morphology evolution and mechanical effects.

Capacity fade mechanisms in lithium-metal batteries are often more severe than in lithium-ion batteries due to the combined effects of SEI growth and dead lithium. Physics-based models can separate these contributions, while empirical models may conflate them. Machine learning models can detect subtle patterns in degradation data but must be carefully designed to avoid overfitting.

Validation of degradation models for lithium-metal batteries remains challenging due to the complexity of the underlying processes. In-situ and operando characterization techniques provide critical data for model calibration. Cross-validation between different modeling approaches can improve robustness. Future work should focus on integrating multi-scale phenomena, from atomic-scale reactions to cell-level performance, and on developing standardized testing protocols for model comparison.

In summary, degradation modeling for lithium-metal batteries requires careful consideration of SEI growth kinetics, dead lithium accumulation, and their combined impact on capacity fade. Physics-based models offer deep mechanistic understanding, empirical models provide practical fitting tools, and machine learning enables data-driven pattern recognition. The choice of modeling approach depends on the specific application, available data, and desired level of detail. Continued advances in modeling will accelerate the development of durable lithium-metal batteries for real-world applications.