[Paper Review] Machine learning-assisted discovery of many new solid Li-ion conducting materials

Austin D. Sendek, Ekin D. Cubuk+4 more|arXiv (Cornell University)|Aug 7, 2018

Machine Learning in Materials Science11 references4 citations

TL;DR

This study leverages a machine learning (ML)-guided search combined with density functional theory molecular dynamics (DFT-MD) to identify over 12,000 new crystalline solid Li-ion conductors, achieving a 45x improvement in average room-temperature Li-ion conductivity and a 2.7x higher success rate than random search. The ML model outperforms six expert Ph.D. students, doubling the F1 score and accelerating discovery by 1,000-fold while identifying materials with high stability and low electronic conductivity.

ABSTRACT

We discover many new crystalline solid materials with fast single crystal Li ion conductivity at room temperature, discovered through density functional theory simulations guided by machine learning-based methods. The discovery of new solid Li superionic conductors is of critical importance to the development of safe all-solid-state Li-ion batteries. With a predictive universal structure-property relationship for fast ion conduction not well understood, the search for new solid Li ion conductors has relied largely on trial-and-error computational and experimental searches over the last several decades. In this work, we perform a guided search of materials space with a machine learning (ML)-based prediction model for material selection and density functional theory molecular dynamics (DFT-MD) simulations for calculating ionic conductivity. These materials are screened from over 12,000 experimentally synthesized and characterized candidates with very diverse structures and compositions. When compared to a random search of materials space, we find that the ML-guided search is 2.7 times more likely to identify fast Li ion conductors, with at least a 45x improvement in the log-average of room temperature Li ion conductivity. The F1 score of the ML-based model is 0.50, 3.5 times better than the F1 score expected from completely random guesswork. In a head-to-head competition against six Ph.D. students working in the field, we find that the ML-based model doubles the F1 score of human experts in its ability to identify fast Li-ion conductors from atomistic structure with a thousand- fold increase in speed, clearly demonstrating the utility of this model for the research community. All conducting materials reported here lack transition metals and are predicted to exhibit low electronic conduction, high stability against oxidation, and high thermodynamic stability.

Motivation & Objective

• To overcome the limitations of trial-and-error methods in discovering solid-state Li-ion conductors with fast ionic conductivity.
• To develop a machine learning model capable of predicting fast ion conduction in crystalline materials based on atomic structure.
• To guide a high-throughput search across over 12,000 experimentally synthesized materials using ML predictions and DFT-MD validation.
• To identify new solid Li-ion conductors that are stable, non-transition-metal-based, and exhibit low electronic conductivity.
• To demonstrate the superiority of ML-guided discovery over human expert intuition and random search in materials discovery.

Proposed method

• A machine learning model was trained on structural and compositional features of known materials to predict fast Li-ion conduction.
• The model ranked over 12,000 experimentally synthesized materials based on predicted ionic conductivity, prioritizing candidates for further study.
• Density functional theory molecular dynamics (DFT-MD) simulations were used to calculate the actual room-temperature Li-ion conductivity of top-ranked candidates.
• The model’s predictive performance was evaluated using F1 score and compared against random search and expert human performance.
• Materials were screened for thermodynamic stability, oxidation resistance, and low electronic conductivity to ensure practical viability.
• The ML model was validated in a head-to-head competition against six Ph.D. students in the field, measuring F1 score and discovery speed.

Experimental results

Research questions

• RQ1 Can a machine learning model effectively predict fast Li-ion conduction in crystalline materials without relying on prior understanding of structure-property relationships?
• RQ2 How does ML-guided discovery compare to random search in identifying high-conductivity solid Li-ion conductors?
• RQ3 Can an ML model outperform expert human intuition in identifying promising Li-ion conductor candidates from atomic structure?
• RQ4 What is the impact of ML guidance on the average room-temperature Li-ion conductivity of discovered materials?
• RQ5 What are the stability and electronic conductivity characteristics of the newly discovered Li-ion conductors?

Key findings

• The ML-guided search identified materials with a 45x higher log-average room-temperature Li-ion conductivity compared to random search.
• The ML model achieved an F1 score of 0.50, representing a 3.5-fold improvement over random guesswork.
• The ML model doubled the F1 score of six expert Ph.D. students in identifying fast Li-ion conductors while operating 1,000 times faster.
• The model increased the likelihood of discovering fast Li-ion conductors by 2.7 times compared to random search.
• All identified materials are free of transition metals, exhibit high thermodynamic stability, and are predicted to have low electronic conductivity.
• The study discovered many new solid Li-ion conductors with promising properties for all-solid-state battery applications.