Beyond Fluorination: A Golden Criterion Guided by Chemical Coordination-Informed Machine Learning for High-Voltage...

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry

Background:

• Fluorine chemistry enhances electrolyte voltage limits via interfacial passivation due to high electronegativity.
• Conventional high-voltage electrolyte design faces trade-offs between oxidation stability and ion transport kinetics due to complex component interactions.

Purpose of the Study:

• To develop a machine learning approach for designing high-voltage electrolytes by parsing chemical coordination features.
• To overcome the limitations of traditional electrolyte design and reconcile conflicting performance targets.

Main Methods:

• A Chemical Coordination-Informed Molarity feature parsing approach was developed and embedded into machine learning models.
• Gradient boosting regression was trained to predict oxidation potential based on component molarities.
• A dataset of 2808 ternary-solvent blend candidates was analyzed.

Main Results:

• The trained model achieved a prediction of oxidation potential with Mean Absolute Error (MAE) below 0.36 V.
• The ratio of mono-coordinated fluorine (F1) to double-bonded oxygen (O1) molarity (F1/O1) was identified as crucial for enhancing oxidative stability.
• A design criterion (F1(≥8.19)/O1(≥13.39) ∈ [0.55, 1.10]) was defined for O1-involved recipes.
• Two promising low-fluoride electrolyte recipes with oxidation potentials around 6.3 V vs. Li+/Li and high ion-transport kinetics were identified.

Conclusions:

• The study demonstrates a customizable feature engineering strategy for intelligent materials design.
• The approach enables the reconciliation of mutually exclusive performance targets in electrolyte development.
• This work provides a data-driven method for designing advanced electrolytes for high-voltage applications.