Found the culprit! -- Stanford University EES reveals: the fundamental reason for the difference in Coulombic efficiency of high-performance lithium metal battery electrolytes!

                  

Lithium metal batteries are considered ideal energy storage devices due to their high capacity and energy density, but the high activity of lithium limits their commercialization. In recent years, advances in the design of liquid electrolytes have improved the efficiency of lithium metal batteries, but the efficiency improvement has reached a bottleneck, and the reason is still unclear.

Recently, Stacey F. Bent's team at Stanford University discovered that electrochemical corrosion is an important and previously overlooked driver of high CE electrolyte performance by combining data science techniques with a series of characterization measurements. First, an interpretable machine learning algorithm was used to identify the performance drivers of 93 high CE electrolytes (electrolyte properties such as low solvent oxygen content and high fluorine content). Then, new high CE electrolytes were designed based on these key properties, and through spectral and electroanalytical methods, it was found that the properties commonly used to predict CE (such as lithium morphology, Li-electrolyte reactivity, lithium ion transport, and SEI chemical composition) were not strongly correlated with the performance differences of high CE electrolytes. On the contrary, electrochemical corrosion can explain the performance differences of high CE electrolytes, and it is negatively correlated with CE. This study provides important insights into the field of lithium metal battery research and guides people to design electrolytes with Coulombic efficiency exceeding 99.9%.

【Key points】

By combining machine learning with experimental characterization, this paper reveals the key influencing mechanism of galvanic corrosion on Coulombic Efficiency (CE) in high-performance lithium metal battery electrolytes. This discovery breaks through the limitations of performance indicators relied on in traditional electrolyte design.

 1. Limitations of Traditional Performance Indicators

In the design of lithium metal battery electrolytes, commonly used performance indicators include lithium morphology, electrolyte transport properties, chemical composition of the solid electrolyte interface (SEI), and reactivity between lithium and electrolyte. These indicators have certain guiding significance in distinguishing low CE (<98%) and high CE (>98%) electrolytes.

However, these traditional indicators show obvious limitations when further optimizing high CE electrolytes. For example, although lithium morphology differs between low CE and high CE electrolytes, the similarity of lithium morphology in high CE electrolytes makes it unable to effectively explain the small differences in CE. In addition, the chemical composition of SEI and the reactivity of Li-electrolyte also fail to provide sufficient differentiation in high CE electrolytes.

2. Construction and application of machine learning models

Data segmentation and feature selection: The researchers divided the data of 189 electrolytes into two groups: low CE (<98%) and high CE (>98%), and identified the key features that affect the performance of the electrolytes through machine learning algorithms. For example, the performance of low CE electrolytes is mainly affected by the oxygen content (sO) and fluorine-oxygen ratio (F/O) in the solvent, while the performance of high CE electrolytes is closely related to the carbon content (sC) and fluorine content (sF) in the solvent.

Model prediction and experimental verification: Based on the machine learning model, the researchers designed a new high-CE electrolyte and verified its performance through experiments. However, there are significant discrepancies between the experimental results and the model predictions, indicating that traditional performance metrics cannot fully explain the performance differences of high-CE electrolytes.

 3. The key role of electrochemical corrosion

Experimental design and measurement: In order to further explore the performance differences of high CE electrolytes, the researchers measured the electrochemical corrosion current in the electrolyte through a voltage step experiment. Specifically, the lithium copper battery was reduced from the open circuit voltage (OCV) to 15 mV vs Li/Li⁺ and maintained at this voltage for 48 hours, and the corrosion capacity was obtained by integrating the corrosion current.

Results and Analysis: The experimental results showed that the corrosion capacity was significantly negatively correlated with the CE of the electrolyte. For example, the DEE (diethyl ether) electrolyte has the lowest corrosion capacity and the highest CE (99.6%), while the EBE (ethyl butyl ether) electrolyte has the highest corrosion capacity and the lowest CE (98%). This indicates that electrochemical corrosion is the key factor causing the performance differences in high CE electrolytes.

Mechanism explanation: Electrochemical corrosion mainly occurs at the interface between lithium metal and copper current collector, resulting in lithium loss due to electron transfer from lithium to copper. In high CE electrolytes, the driving force of this corrosion reaction is smaller, thereby reducing lithium loss and improving CE. Therefore, electrochemical corrosion is not only a minor capacity loss factor, but also has important guiding significance in the design of high CE electrolytes.

 4. Conclusion and Outlook

   - Conclusion: This paper reveals the key role of electrochemical corrosion in high-performance lithium metal battery electrolytes through a combination of machine learning and experimental characterization. This finding shows the limitations of traditional performance indicators in the design of high CE electrolytes, and electrochemical corrosion is a more sensitive and effective performance indicator.

   - Outlook: Future research can further explore the microscopic mechanism of electrochemical corrosion, for example, through molecular dynamics simulation and in-situ characterization technology, to deeply understand the kinetics and thermodynamics of corrosion reactions. In addition, combining more physicochemical properties (such as molecular orbital energy, ion solvation sheath properties, etc.) can further optimize the machine learning model and provide more comprehensive guidance for the design of high-performance electrolytes.

 Figure 1: Machine learning process used to understand the performance drivers of liquid electrolytes

a. Electrolyte properties encoded by machine learning features. F%, O% and C% represent the mole fractions of fluorine, oxygen and carbon in the entire electrolyte volume, respectively. F/C, F/O and O/C represent the mole fraction ratios of the corresponding elements. sO, aC and sC represent the mole fraction of oxygen in the solvent, the mole fraction of carbon in the anion and the mole fraction of carbon in the solvent, respectively. A complete description of the abbreviated features is given in Table 1.  

b. Experimentally measured plot of electrolyte efficiency versus the mole fraction of oxygen in the solvent (sO).  

c. Overall machine learning workflow for electrolytes and feature discovery.

 Figure 2: Model selection and feature importance ranking for low- and high-performance electrolytes

a. Cross-validation error and PAC bounds for low-performing electrolytes.  

b. Cross-validation error and PAC bounds for high-performance electrolytes. Error bars represent the standard deviation of the cross-validation error calculated using 10 different data splits.  

c. The percentage increase in mean squared error for each missing feature in the four-feature model for low-performing electrolytes.  

d. The percentage increase in mean squared error for each missing feature in the four-feature model for high-performance electrolytes.

 Figure 3: Identification and testing of high-performance electrolytes guided by machine learning

a. sO and F/O values ​​of the new solvents selected in this study.  

b. Predicted and measured average CE values ​​for high-performance electrolytes. The error bars for measured CE represent the standard deviation calculated using three cells.

 Figure 4: Relationship between common performance indicators and Coulombic efficiency in high-performance electrolytes

a. Morphology of 0.5 mAh lithium deposited at 0.5 mA cm⁻² in a high-performance electrolyte.  

b. Lithium stripping overpotential in high-performance electrolyte at the end of the Aurbach test at 0.5 mA cm⁻².  

c. Nyquist plot of lithium ion conductivity of high-performance electrolyte.  

d. Relationship between electrolyte CE and lithium stripping overpotential and ionic conductivity.  

e. Atomic ratios of SEI components formed when 0.5 mAh lithium is deposited at 0.5 mA cm⁻² in a high-performance electrolyte versus average CE.  

f. Time evolution of SEI resistance of high-performance electrolyte when tested using Li|Li battery.  

g. Correlation coefficients of common performance indicators and CE, where ΔR(t) represents the ratio of SEI impedance measured at time t to SEI impedance measured at t = 0h. Cyan markers represent measured performance indicators, and gray markers represent calculated machine learning design parameters.

 Figure 5: Relationship between electrochemical corrosion and Coulombic efficiency of high-performance electrolyte

a. Corrosion capacity versus time in high-performance electrolyte.  

b. Average corrosion capacity of the high performance electrolyte calculated using two cells.  

c. Correlation coefficients of corrosion capacity and related derivative characteristics associated with CE.

【in conclusion】

In this study, we revealed that electrochemical corrosion can explain the Coulombic efficiency (CE) differences of high-performance lithium metal battery electrolytes better than common performance indicators through machine learning models combined with common performance indicators. Using machine learning models that include common performance indicators, insights for electrolyte design were extracted and new high-performance electrolytes were synthesized based on them. The Coulombic efficiency of these new high-performance electrolytes ranged from 98% to 99.6%, and although the machine learning model predicted that they would be high-performance (CE >98%), there was a discrepancy between the measured and predicted values. This discrepancy suggests that the common performance indicators encoded in the machine learning model cannot explain the CE differences between high-performance electrolytes. Common performance indicators such as lithium morphology, lithium ion conductivity, SEI chemistry, and Li-electrolyte reactivity were measured by spectral and electroanalytical methods, and correlation coefficients were used to show that there was only a weak correlation between the CE of high-performance electrolytes and common performance indicators. Subsequently, electrochemical corrosion was presented as a key driver of the CE of these high-performance electrolytes, revealing the importance of factors that lead to small capacity losses when designing high-performance CE electrolytes. Overall, this study calls for the introduction of new metrics and considerations in the design of high-performance CE electrolytes.

-End-

Canrd Brief Introduce

Canrd use high battery R&D technology(core members are from CATL) and strong Chinese supply chain to help many foreign companies with fast R&D. We provide lab materials,electrodes, custom dry cells, material evaluation, perfomance and test, coin/pouch/cylindrical cell equipment line, and other R&D services.

Email:janice@canrd.com    

Phone/Wechat/WhatsApp/Skype:+86 18928276992

Website : www.canrud.com