Source-linked AI summary
Machine Learning Enabled Computational Screening of Inorganic Solid Electrolytes for Dendrite Suppression with Li Metal Anode
Zeeshan Ahmad, Tian Xie, Chinmay Maheshwari, Jeffrey C. Grossman, Venkatasubramanian Viswanathan
TL;DR
Dendritic electrodeposition and capacity fade motivate assessing solid electrolytes for Li-metal batteries. The paper combines stable-electrodeposition criteria with machine-learning-enabled screening, identifying more than twenty dendrite-stable interfaces involving six solid electrolytes while noting a small candidate list.
Problem
Dendritic electrodeposition on Li metal anodes causes capacity fade, while the capability of solid electrolytes also needs assessment.
Method
The study applies stable-electrodeposition criteria with machine-learning techniques to computationally screen solid electrolytes, reducing reliance on expensive quantum mechanical calculations through structural descriptors.
Results
More than twenty interfaces with six solid electrolytes are predicted to be stable to dendrite initiation, after screening 12,950 solids using isotropic stability criteria.
Takeaways & Limitations
The screening identifies candidate Li-metal/solid-electrolyte interfaces for suppressing dendrite initiation and reveals common features among the predicted candidates.
Takeaways & Limitations
The identified candidate list is small, and some candidates may be toxic or expensive because of low constituent abundance.
Abstract
from arXiv · showhide
Next generation batteries based on lithium (Li) metal anodes have been plagued by the dendritic electrodeposition of Li metal on the anode during cycling, resulting in short circuit and capacity loss. Suppression of dendritic growth through the use of solid electrolytes has emerged as one of the most promising strategies for enabling the use of Li metal anodes. We perform a computational screening of over 12,000 inorganic solids based on their ability to suppress dendrite initiation in contact with Li metal anode. Properties for mechanically isotropic and anisotropic interfaces that can be used in stability criteria for determining the propensity of dendrite initiation are usually obtained from computationally expensive first-principles methods. In order to obtain a large dataset for screening, we use machine learning models to predict the mechanical properties of several new solid electrolytes. We train a convolutional neural network on the shear and bulk moduli purely on structural features of the material. We use AdaBoost, Lasso and Bayesian ridge regression to train the elastic constants, where the choice of the model depended on the size of the training data and the noise that it can handle. Our models give us direct interpretability by revealing the dominant structural features affecting the elastic constants. The stiffness is found to increase with a decrease in volume per atom, increase in minimum anion-anion separation, and increase in sublattice (all but Li) packing fraction. Cross-validation/test performance suggests our models generalize well. We predict over 20 mechanically anisotropic interfaces between Li metal and 6 solid electrolytes which can be used to suppress dendrite growth. Our screened candidates are generally soft and highly anisotropic, and present opportunities for simultaneously obtaining dendrite suppression and high ionic conductivity in solid electrolytes.
Introduction
Li-metal batteries face dendrite-driven short circuits and capacity loss, motivating solid-electrolyte strategies and computational screening focused on dendrite initiation. This work combines machine-learned mechanical properties with stability analyses to identify promising inorganic electrolytes and interfaces.
- Motivation: Dendritic Li electrodeposition causes short circuits and capacity loss, making stable, smooth, dendrite-free deposition important for adoption.The introduction identifies dendrite growth and electrolyte consumption as major commercialization hurdles.
- Research gap: A comprehensive and precise criterion for dendrite suppression remains elusive, while inorganic crystalline materials require softer electrolytes for stability.The cited stability criterion applies specifically to dendrite initiation; propagation may require other suppression approaches.
- Research gap: Dendrite initiation is prioritized because growth becomes extremely difficult to mitigate once initiated.The paper therefore targets prevention of initiation to support smooth electrodeposition throughout cycling.
- Approach: Machine learning using structural descriptors can accelerate high-throughput screening by several orders of magnitude while predicting mechanical properties for stability analysis.The study trains models for mechanical properties and feeds their predictions into a stability-parameter framework for dendrite initiation.
- Approach: The study screens solid electrolytes for dendrite-initiation suppression, ranking isotropic interfaces by the surface roughness wavenumber required for stable electrodeposition.It also analyzes more than 15,000 anisotropic Li-metal/solid-electrolyte interfaces using the Stroh formalism.
- Findings: The anisotropic analysis identifies over twenty candidate interfaces, generally soft and highly anisotropic, offering simultaneous dendrite suppression and fast ion conduction.For isotropic interfaces, most electrolytes are not intrinsically stabilized by interfacial stresses, so surface nanostructuring may be essential.
Results and Discussion
The study combines machine-learning predictions of elastic properties with isotropic and anisotropic stability criteria to screen solid electrolytes for dendrite suppression. Screening shows that surface tension stabilizes isotropic interfaces at sufficiently high roughness wavenumber, while promising anisotropic candidates are generally soft or highly anisotropic.
- Method: The workflow predicts electrolyte mechanical properties with machine-learning models and feeds them into a stability-parameter framework for screening.The framework evaluates isotropic and anisotropic interfaces separately.
- Stability criteria: The stability parameter χ determines electrodeposition behavior: positive χ promotes dendrite growth, whereas negative χ indicates stabilization or suppression.The parameter incorporates surface tension and hydrostatic and deviatoric stresses at the metal–electrolyte interface.
- Isotropic screening: Surface tension increasingly dominates as surface roughness wavenumber k rises, stabilizing interfaces beyond the critical wavenumber kcrit.For stress-destabilized materials, χ changes sign at an intermediate k because χ → −∞ as k → ∞.
- Machine-learning models: The elastic-constant models use AdaBoost for C11, Lasso for C12, and Bayesian ridge for C44, achieving overall R2 values of 0.98, 0.85, and 0.69, respectively.Three-fold cross-validation R2 values are 0.68, 0.85, and 0.71; the dominant structural features are volume per atom, packing fraction, and anion–anion separation.
- Anisotropic screening: The top anisotropic interfaces are stabilized by stresses rather than surface tension and are generally mechanically soft or highly anisotropic.All screened candidates have universal anisotropy index greater than 10, and anisotropy may permit compliance along selected crystallographic directions.
Conclusions
The study combines machine-learning property prediction with isotropic and anisotropic stability criteria to screen inorganic solid electrolytes for Li dendrite suppression. Anisotropic screening identifies over twenty predicted-stable interfaces, while isotropic screening finds no stress-only stabilized materials.
- Machine-learning screening: Machine-learning models accelerate computational screening by predicting solid-electrolyte mechanical properties from structure–property relationships.The approach includes a convolutional neural network for shear and bulk moduli and regression models for selected elastic constants.
- Screening results: 12,950 solids were screened using isotropic stability criteria, and over 15,000 Li–electrolyte interfaces were screened using anisotropic criteria.
- Isotropic stability: No isotropically screened material was stabilized solely by interface-generated stresses, although surface-tension stabilization was predicted at high surface-roughness wavenumbers.
- Anisotropic stability: Over twenty interfaces involving six solid electrolytes were predicted stable against dendrite initiation when crystallographic orientation was included.
- Candidate characteristics: The screened candidates are mechanically soft and highly anisotropic, offering an opportunity to combine dendrite suppression with fast ion conduction.
- Outlook: Doping and defect generation may be needed to satisfy other solid-electrolyte requirements, and experiments could clarify the dendrite-suppression mechanism.
Supporting Information Available
The supporting information provides model details, predicted mechanical-property datasets, and isotropic or anisotropic stability data for Li-containing compounds.
- Machine-learning data: The supporting files include details of the machine-learning models and predicted shear and bulk moduli for 60,648 compounds.
- Stability data: Additional files provide anisotropic stability parameters for 482 Li-containing compounds with database properties and 548 cubic Li-containing compounds with model-predicted properties.
Supporting Information: Machine Learning
The supporting-information section identifies the paper and its authors and affiliations.
- Paper identification: The paper is titled “Enabled Computational Screening of Inorganic Solid Electrolytes for Dendrite Suppression.”
- Authors: The authors are Zeeshan Ahmad, Tian Xie, Chinmay Maheshwari, Jeffrey C. Grossman, and Venkatasubramanian Viswanathan.
- Affiliations: The affiliations include Carnegie Mellon University in Pittsburgh and the Massachusetts Institute of Technology in Cambridge.
Interface between Li metal and solid electrolyte
Figure S1 represents the Li-metal–solid-electrolyte boundary during electrodeposition as a general two-dimensional interface z = f(x).
- Interface geometry: Figure S1 depicts a general 2D interface z = f(x) between Li metal and an inorganic solid electrolyte during electrodeposition.
Details of the machine learning models
The supplementary workflow specifies the CGCNN and regression-model resources, interface-orientation search, and stability-parameter inputs used for screening.
- Figure S2 schematically presents the architecture of the CGCNN used in the modeling workflow.
- The supplementary tables document CGCNN hyperparameters, elastic-tensor descriptors, cubic-class regression models, training-data availability, and surface energies.
- Each model was trained for 1000 epochs, with the learning rate reduced tenfold.
- The anisotropic interface search combines low-index Li and solid-electrolyte facets, including seven specified electrolyte directions normal to the interface.
- Elastic tensors for arbitrary crystallographic orientations were obtained by rotating the axes using tensors from conventional or primitive unit cells.
- 16,180 total interfaces were screened from DFT and predicted surfaces using the stated surface-count calculation.The calculation is 2401 × 4 + 548 × 3 × 4 = 16180.
- For cubic crystals, the stability parameter χ increases with the elastic-tensor eigenvalues and decreases as the material becomes softer.The eigenvalues are A = C11 − C12, B = (C11 + 2C12)/3, and C44.