# Fileset

[unveiling-electrolyte-design-principles-for-sodium-ion-batteries-using-combinatorial-electrochemistry-and-machine.pdf](https://mdr.nims.go.jp/filesets/6ac4c94d-6131-4992-aa4d-e1dc1c801996/download)

## Creator

[Manai Ono](https://orcid.org/0000-0003-4406-4113), Misato Takahashi, [Ryo Tamura](https://orcid.org/0000-0002-0349-358X), [Shoichi Matsuda](https://orcid.org/0000-0002-0640-3404)

## Rights

[Creative Commons BY-NC-ND Attribution-NonCommercial-NoDerivs 4.0 International](https://creativecommons.org/licenses/by-nc-nd/4.0/)

## Other metadata

[Unveiling Electrolyte Design Principles for Sodium-Ion Batteries Using Combinatorial Electrochemistry and Machine Learning-Assisted Analysis](https://mdr.nims.go.jp/datasets/6f69b8c5-9d9a-4839-8b4e-f9bfee409c7c)

## Fulltext

Unveiling Electrolyte Design Principles for Sodium-Ion Batteries Using Combinatorial Electrochemistry and Machine Learning-Assisted AnalysisUnveiling Electrolyte Design Principles for Sodium-Ion BatteriesUsing Combinatorial Electrochemistry and Machine Learning-Assisted AnalysisPublished as part of ACS Applied Energy Materials special issue “Material and Interface Design in Next-Generation Batteries”.Manai Ono, Misato Takahashi, Ryo Tamura,* and Shoichi Matsuda*Cite This: ACS Appl. Energy Mater. 2026, 9, 1405−1411 Read OnlineACCESS Metrics & More Article Recommendations *sı Supporting InformationABSTRACT: To accelerate the development of high-performance electrolytesfor sodium-ion batteries (SIBs), we systematically investigated the effects ofthree key parameters, NaFSI concentration, DMC/EMC ratio, and FECcontent, on the performance of SIB using NaNi1/3Fe1/3Mn1/3O2 and hardcarbon as the positive and negative electrodes. A total of 132 electrolyteformulations were prepared using automated liquid handling, and theirelectrochemical performance was evaluated using multichannel full-cellmeasurements. Data-driven analysis employing machine learning revealed thatNaFSI concentration plays the most critical role in enabling highly reversiblecharge−discharge behavior. Long-term cycling tests and interfacial compositionanalyses were further conducted to clarify the influence of electrolytecomponents on stability. Detailed studies focusing on high-NaFSI, FEC-containing electrolytes showed that EMC-rich formulationsoutperformed DMC-rich counterparts, maintaining Coulombic efficiencies over 99.6% even after 300 cycles. X-ray photoelectronspectroscopy confirmed that these stable systems promote the formation of NaF-rich interphases on both electrodes. These findingsprovide valuable insights into electrolyte design strategies for durable and efficient SIBs and highlight the utility of high-throughputexperimentation coupled with machine learning for electrolyte discovery.KEYWORDS: high throughput experiment, sodium ion battery, multicomponent electrolyte, machine learning, solid-electrolyte interface■ INTRODUCTIONSodium-ion batteries (SIBs) have garnered growing attentionas a promising alternative to lithium-ion batteries (LiBs) fornext-generation energy storage systems.1,2 Owing to thenatural abundance, low cost, and broad geographic distributionof sodium resources, SIBs offer a more sustainable andeconomically viable solution, particularly for large-scaleapplications such as grid storage. Furthermore, SIBs sharesimilar intercalation chemistry and cell design as LiBs, allowingfor relatively straightforward adaptation of existing technolo-gies. Despite these advantages, SIBs are still at an earlier stageof development, and several key challenges must be addressedbefore their widespread commercialization. One of the mostcritical challenges lies in achieving long-term cycling stability,which is essential for the reliable operation of SIBs overextended periods. Among various factors influencing batteryperformance, the electrolyte plays a central role. It not onlymediates ion transport but also governs the formation andevolution of the solid electrolyte interphase (SEI) on theanode and the cathode electrolyte interphase (CEI) on thecathode.3,4 These interphases critically impact capacityretention, Coulombic efficiency, and overall cell lifespan.Designing a high-performance electrolyte is a highly intricatetask, as it requires careful tuning of multiple interrelatedparameters, including solvent polarity, salt concentration,solvation structure, the electrochemical stability window, andinterfacial compatibility.In practical battery systems, mixed-solvent electrolyteformulations are commonly used to balance the ionicconductivity, thermal and electrochemical stability, andinterfacial properties. However, the rational selection of solventcombinations for SIBs remains largely empirical. The vastcompositional space and the nonlinear interplay betweencomponents make it difficult to predict optimal electrolyteformulations. As a result, electrolyte discovery has traditionallyrelied on trial-and-error approaches, which are time-consumingReceived: September 26, 2025Revised: December 24, 2025Accepted: December 29, 2025Published: January 16, 2026Articlewww.acsaem.org© 2026 The Authors. Published byAmerican Chemical Society1405https://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−1411This article is licensed under CC-BY-NC-ND 4.0Downloaded via NATL INST FOR MATLS SCIENCE (NIMS) on February 12, 2026 at 22:54:48 (UTC).See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.https://pubs.acs.org/curated-content?journal=aaemcq&ref=featurehttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Manai+Ono"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Misato+Takahashi"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Ryo+Tamura"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Shoichi+Matsuda"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://pubs.acs.org/action/showCitFormats?doi=10.1021/acsaem.5c03028&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?goto=articleMetrics&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?goto=recommendations&?ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?goto=supporting-info&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=tgr1&ref=pdfhttps://pubs.acs.org/toc/aaemcq/9/3?ref=pdfhttps://pubs.acs.org/toc/aaemcq/9/3?ref=pdfhttps://pubs.acs.org/toc/aaemcq/9/3?ref=pdfhttps://pubs.acs.org/toc/aaemcq/9/3?ref=pdfwww.acsaem.org?ref=pdfhttps://pubs.acs.org?ref=pdfhttps://pubs.acs.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://www.acsaem.org?ref=pdfhttps://www.acsaem.org?ref=pdfhttps://creativecommons.org/licenses/by-nc-nd/4.0/and resource-intensive. Recent advances in high-throughputexperimentation and data-driven methodologies have demon-strated their effectiveness for transforming the landscape ofmaterials discovery even in the field of rechargeablebatteries.5,6 In the context of solid-state materials, machinelearning and computational screening have successfully guidedthe identification of promising candidates.7,8 For liquidmaterials, such as electrolytes, high-throughput measurementtechniques for fundamental physicochemical properties such asionic conductivity and viscosity are now available.9,10 However,translating these basic metrics into reliable predictions ofelectrochemical performance, particularly under realisticcycling conditions, remains a significant challenge. This isprimarily due to the complexity of electrochemical reactions atinterfaces and the lack of scalable methods for evaluating long-term battery performance in a high-throughput manner.Based on this research background, we recently developed amodular high-throughput experimental platform that integratesautomated liquid handling with parallel electrochemicaltesting.11−13 Using LiB electrolyte systems as a model, wedemonstrated the capability to evaluate hundreds ofcompositions in a systematic and reproducible manner.14This platform enables rapid screening of electrolyte candidatesunder conditions that closely mimic the real battery operation.In this study, we extend our high-throughput methodology toSIBs, aiming to accelerate the discovery of electrolyteformulations that promote long-term cycling stability. Bysystematically exploring a wide array of mixed-solvent systems,we seek to uncover correlations among electrolyte composi-tion, interfacial behavior, and electrochemical performance.Our results provide valuable insights into the design principlesgoverning electrolyte optimization for SIBs and demonstratethe power of high-throughput approaches in advancing next-generation battery technologies.■ EXPERIMENTAL METHODBattery Performance EvaluationThe battery-grade chemicals, including NaFSI, EC, PC, DMC, EMC,and FEC, were purchased from Kishida Chemical Co., Ltd., withpurities exceeding 99.9% for NaFSI and 99.5% for the solvent. Thediameter of the interior of the cell was 9 mm. The glass fibermembrane (thickness: 100 μm, diameter: 6.4 mm) was employed as aseparator. The hard carbon electrode and the NaNi1/3Fe1/3Mn1/3O2electrode were used after drying in a vacuum.For fabrication of MCE cells, electrode and separator sheets arepunched into the desired size and inserted into MCE cells by usingsemiautomated punching equipment. One MCE cell module isprepared within 1 h and introduced into an Ar-filled Gl-box.Multicomponent electrolyte preparation and injection into theMCE cell were performed by a liquid handling dispenser (Gyger,CERTUS), which is installed inside of an Ar-filled Gl-box. Theequipment realizes contactless dispensing of various kinds of liquidsamples with different physical properties by utilizing eightindividually controllable channels, using microvalve technology andair pressure control. The mother solution is supplied to the dispenserfrom 50 mL sized bottles to the dispensing channel through the tube.Calibration is performed for each mother solution by adjustingmicrovalve and air pressure control conditions, which enables theelectrolyte injection with accuracy at the scale of 50 nL. Mothersolutions were directly injected into each MCE cell with anappropriate solution volume ratio, resulting in the total volume ofsolution injected into each cell being controlled to be 40 μL. Notably,as the electrolyte injection is done by a dispensing system, the injectedelectrolyte is spontaneously mixed inside of the cell. An Au-coatedSUS-based spring was introduced into each well, electricallyconnected, and controlling the confining pressure. Then, the toppart of the MCE cell was tightly sealed by capping its top with bolts,thereby largely suppressing the volatilization of electrolyte.Fabricated MCE cell modules were taken out from the Gl box andwere subjected to a charge/discharge test using a battery cycler(Hokuto Denko, HJ1001SD8). During the battery performance test,the MCE cell was installed into a temperature-controlled chamber(ESPEC, LU-124). When the battery performance test wascompleted, the MCE cells were disassembled. The parts of boltswere opened, and electrodes, separators, and electric connection partswere taken out. Although the electrodes and separators were disposedof, the MCE cell itself and electric connection parts were washed anddried. For fabrication of coin-type cells, the same kinds of electrolyteand electrode materials that were utilized for MCE cells were alsoused.XPS AnalysisThe chemical composition of the electrode was analyzed by using anX-ray photoelectron spectrometer (VersaProbe II Scanning XPSMicroprobe, ULVAC-PHY). The analytical samples of electrochemi-cally measured electrodes were obtained by disassembling cells andwashing and drying them in an Ar-filled glovebox. THF (tetrahy-drofuran, wako) was utilized for washing electrodes. The sampleswere cut into pieces and directly subjected to analysis of XPS withoutexposure to the atmosphere by using a transfer vessel.Machine LearningThe classification model was implemented using the Random ForestClassifier model available in scikit-learn.15 We tuned the max depthhyperparameter using Grid Search CV with 10-fold cross-validation.Three different scoring metrics were tested: accuracy, recall, and F1-score. For the regression task, the Linear Regression modelimplemented in scikit-learn was employed.■ RESULTS AND DISCUSSIONIn the present study, we developed a systematic workflow toexplore optimal electrolyte compositions for SIBs using a high-throughput experimental and data-driven approach. First, weevaluated the electrochemical performance of 132 electrolyteformulations comprising multiple organic solvents. Automatedliquid handling dispensing and multichannel electrochemicalmeasurement systems were employed to efficiently conductthis experiment (Figure 1). Subsequently, we applied machinelearning techniques to the acquired data set in order to identifykey parameters that contribute to achieving superior batteryperformance. After that, X-ray photoelectron spectroscopy(XPS) analysis of the electrode/electrolyte interface isperformed for selected samples to further deepen ourunderstanding of the reaction inside of the battery.Figure 1. Schematic illustration of the overall workflow of the present study.ACS Applied Energy Materials www.acsaem.org Articlehttps://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−14111406https://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig1&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig1&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig1&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig1&ref=pdfwww.acsaem.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-asThe battery performance evaluation was conducted usingmultichannel electrochemical cells (Figure S1), using Na-Ni1/3Fe1/3Mn1/3O2 and hard carbon as the positive andnegative electrodes. The details of the design of multichannelelectrochemical cells are described in our previous report.14 Arepresentative voltage profile is shown in Figure 2a. In thisexperiment, we use 1 M NaPF6 in EC/EMC (50:50 vol %) asan electrolyte, and the measurement was performed under acurrent density of 0.2 mA cm2 for three cycles. During theinitial charging process, there can be a gradual increase ofvoltage and reached a cutoff voltage of 4.2 V. In thedischarging process, the cell exhibited a stable profile,exhibiting a capacity of 0.78 mAh/cm2. Figure 2b presentsthe value of Coulombic efficiency (CE) during the first to thirdcycles measured across three replicates (n = 3), with averagevalues of 59.5, 88.0, and 90.0%, and standard deviations of 0.5,0.4, and 0.3%, respectively. These results revealed the highreproducibility of the battery performance evaluation in thesetup utilized in the present study.In the present study, we focused on the multisolvent-basedelectrolyte system composed of ethylene carbonate (EC),propylene carbonate (PC), dimethyl carbonate (DMC), ethylmethyl carbonate (EMC), and fluoroethylene carbonate(FEC). NaFSI was selected as the sodium salt because of itshigh solubility in a wide range of solvents. By changing theamount of NaFSI and the ratio of each solvent, the variouskinds of electrolyte with different compositions can beprepared. Electrolyte compositions were described based onmolar ratios, based on our previous study.14 In multi-component electrolytes, using mol/L introduces problemsbecause the total solution volume changes when solvents mixand large amounts of Li salt dissolve, making the final volumeunpredictable and requiring measurement after preparation.Therefore, expressing all electrolyte components in terms ofmolar ratios offers a more precise and consistent method fordescribing the electrolyte composition.While the roles of cyclic carbonates like EC and PC inbattery performance and SEI formation are well investi-gated,16,17 the effects of linear carbonates have been lessexplored. Yet, practical electrolytes often contain mixtures oflinear carbonates such as DMC and EMC to optimize iontransport and interfacial stability. Therefore, this studyspecifically investigates the impact of the DMC/EMC ratio,with the cyclic carbonate content kept constant. Based onthese considerations, we defined the electrolyte design spaceusing three key compositional parameters: y (inverse value ofNaFSI concentration), b (mol ratio of DMC in linearcarbonate), and d (FEC concentration), which are defined asfollows.= + + +yEC PC DMC EMCNaFSI=+bDMCDMC EMC=+ + +dFECEC PC DMC EMCEach parameter varied across the levels shown in Table S1.As a result, a total of 132 electrolytes are described by settingthe specific values for these three parameters. These multi-solvent electrolytes were experimentally prepared by using aliquid handling dispenser installed in an Ar-filled Gl-box(Figure 1b,c). In particular, 132 kinds of electrolytes weregenerated by combinatorial mixing of 9 kinds of mothersolutions listed in Table S2.To efficiently and accurately prepare electrolytes withdiverse and complex compositions, we developed an integratedworkflow that unifies the electrolyte formulation, mixing, andcell assembly within a single streamlined process. In thismethod, preassembled parallel electrochemical cells, in whicheach preloaded with a positive electrode, separator, andnegative electrode, serve as standardized platforms forelectrolyte introduction. A set of predefined mother solutionsis dispensed directly into each cell in precise volumes by usingan automated liquid handling system. After the electrolyteinjection process is finished, the electric connection parts areinstalled in each cell. Then, the multichannel electrochemicalcells are tightly sealed by capping their tops with bolts, whichlargely suppresses electrolyte volatilization. The multichannelelectrochemical cells are then connected to a suitablemultichannel battery tester via an electrochemical connection,which allows for the parallel measurement of batteryperformance for each cell (Figure 1d). In our experiments,the current density was set to be 0.2 mA/cm2 for the first cycleand 0.5 mA/cm2 for the next 10 cycles, with a cutoff voltage of4.2 V/2.0 V condition. The first cycle was intentionallyoperated at a lower current density to serve as a conditioningstep, allowing stable SEI formation before evaluating theintrinsic electrochemical behavior.To apply machine learning techniques to analyze these datasets, we focused on the CE at the 11th cycle as the objectiveparameter. The 11th cycle was selected as a representativepoint because the initial cycles typically involve SEI formationand capacity stabilization, while later cycles exhibit smallervariations. Thus, the 11th cycle reflects the steady-stateelectrochemical behavior across all formulations. Although avariety of electrochemical parameters can be derived from thedata set, we intentionally simplified the analysis by focusing onthis single, widely recognized indicator of cell stability andreversibility to ensure clear and interpretable correlations. Thesummarized results are presented in Figure 3. We classified thebattery performance into two categories based on the 11th CEvalues as follows:Cyclable: 80% < 11th CE <100%Noncyclable: 11th CE < 80 or ≥100%By focusing on these data sets, a binary classification modelusing a random forest classifier was constructed to predictcyclability from electrolyte composition. Hyperparameterswere initially tuned to maximize the accuracy score, resultingin a model accuracy of 0.9 via 10-fold cross-validation. Featureimportance analysis revealed that the most influentialparameters for achieving a stable CE were in order of y, d,Figure 2. (a) Voltage profile of the sodium-ion battery cell operatedat a current density of 0.2 mA/cm2. (b) CE values in the first andthird cycles, measured across three replicates (n = 3).ACS Applied Energy Materials www.acsaem.org Articlehttps://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−14111407https://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig2&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig2&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig2&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig2&ref=pdfwww.acsaem.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-asand b (Figure 3a). Although the resulting precision was high(0.9375), the confusion matrix (Table 1) showed a largenumber of false negatives. To address this, the tuning objectivewas shifted to maximizing recall, which effectively reduced falsenegatives to 3. However, this trade-off led to an increased rateof false positives and a reduced precision of 0.808. Thishighlights the necessity of carefully considering the objectivewhen performing hyperparameter tuning. Note that theconfusion matrices for maximizing recall and maximizing theF1 score are summarized in Tables S3 and S4. In Figure 4, thehistogram of samples showing higher CE is summarized,revealing that a low value of y (high concentration of NaFSI) isan effective guide for designing an electrolyte to achieve stablecharge−discharge cycles.Furthermore, we attempted to construct predictive modelsfor the CE value based on three explanatory variables: y, d, andb. Specifically, we focused on the subset of samples classified asCyclable (11th CE values between 80 and 100%) and appliedthe linear regression technique (Figure 3b). The resultrevealed that the coefficient of determination (R2) evaluatedby the 10-fold cross-validation remained below 0.5, indicating alimited predictive performance. This result suggests that theCE values cannot be simply predicted from these threevariables alone, indicating the complex and multifactorialnature of this system.While our initial discussion just focused on the CE value ofthe 11th cycle, we extended our research scope for the longcycle stability. For this purpose, we exhibited 300th cycle testson eight kinds of electrolyte compositions with two levels ofNaFSI concentration (y = 0.6 or 1.0), in the presence andabsence of FEC (d = 0 or 0.2), and the choice of solvent ofEMC or DMC (b = 0 or 1). The 11th CE and the profile of CEduring the cycling test and the representative voltage profileare summarized in Figure 5. For the EMC-based electrolyte,the cell without FEC (y = 6, d = 0, b = 0) showed a similarprofile to that of the DMC-based electrolyte (y = 6, d = 0, b =1). Initially, the cell exhibited the higher CE over 99%, and thevalue gradually decreased and reached approximately 98% atthe 80th cycle. In sharp contrast, the cell with FEC-containingelectrolytes (y = 6, d = 0.2, b = 0) exhibited CE above 99%over extended cycling. Notably, even at 300th cycles, the CE ishigher than 99.6%. In the case of DMC-based electrolyte, thecell without FEC (y = 6, d = 0, b = 1) exhibited the gradualincrease of CE and reached 99.2% at around the 20th cycle.However, with the progress of the cycle, the CE quicklydecreased and reached below 98% by the 50th cycle. Incontrast, the cell with FEC (y = 6, d = 0.2, b = 1) showed thehigher CE over 99.4% even up to the 200th cycle, although theCE gradually decreased with the progress of the cycle andreached to 99% at the 300th cycle. The capacity retention ofthe series of SIB cells is summarized in Figure S2, which showsa trend consistent with the CE profiles. These results highlightthat increasing the FEC concentration consistently enhanceslong-term CE retention regardless of the choice of mainFigure 3. (a) Importance of parameters in the random forest classifierpredicting cyclable/noncyclable. All data are used as training data, andthe accuracy of the training data is 0.90. (b) Linear regression analysiswith leave-one-out cross-validation.Table 1. Confusion Matrix for Random Forest Classifier-Based Predictions of (Non) Cyclability Obtained Using theLeave-One-Out Cross-Validationpredictioncyclable noncyclablereal cyclable 75 8noncyclable 5 44Figure 4. (a) Pie charts illustrating the relative proportions of cyclable and noncyclable outcomes for each y value. (b) Frequency tablesummarizing the distribution of CE values.ACS Applied Energy Materials www.acsaem.org Articlehttps://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−14111408https://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig3&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig3&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig3&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig3&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig4&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig4&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig4&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig4&ref=pdfwww.acsaem.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-assolvent (EMC or DMC). In addition, the EMC-basedelectrolyte formulation appears more promising for durability.To elucidate the physicochemical origins of superiorperformance exhibited in FEC containing an EMC-basedelectrolyte, we analyzed the SEI formed on the hard carbonanode after the first charge using XPS. Figure 6 displays theelemental composition of the SEI formed in EMC-basedelectrolytes. High-concentration NaFSI with FEC yielded anSEI with ∼35% fluorine and negligible oxygen content,suggesting the formation of NaF-rich SEI layers. In contrast,electrolytes lacking FEC or with lower salt concentrationsshowed <3% fluorine and 10−20% oxygen, indicative ofNa2CO3 or Na2O-rich SEI layers.Similar trends were observed in DMC-based systems,although the NaF content was generally lower than that ofEMC-based counterparts. These findings suggest that EMC-based, high-NaFSI, FEC-containing electrolytes are mosteffective in forming NaF-rich SEI, which correlates with stablelong-term CE. Actually, the several previous studies reportedthe positive effect of FEC and high NaFSI concentration onimproving the performance of hard carbon electrodes inSIBs.18−20 According to a previous computational study, DMCis more readily reduced than EMC.21 Therefore, in DMC-richelectrolytes, DMC undergoes preferential decompositionduring the early stages of SEI formation, yielding predom-inantly organic, solvent-derived SEI components. In contrast,in EMC-rich electrolytes, the higher reductive stability of EMCsuppresses solvent decomposition. As a result, the reduction ofFEC and, more importantly, the FSI anion proceeds morecompetitively than EMC reduction. This preferential anion-derived decomposition promotes the formation of NaF-richSEI layers. While EIS measurements were not included in theFigure 5. (a, c) Summary of 11th CE value for eight kinds of electrolyte. (b, d) Value of CE plotted against cycle number in repeated charge/discharge test.Figure 6. Chemical composition of (a, b) negative and (c, d) positive electrodes revealed by XPS analysis.ACS Applied Energy Materials www.acsaem.org Articlehttps://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−14111409https://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig5&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig5&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig5&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig5&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig6&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig6&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig6&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?fig=fig6&ref=pdfwww.acsaem.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashigh-throughput platform, we have performed additional EIStests for two representative electrolytes (high-NaFSI EMC-richand DMC-rich). The results are summarized in Figure S3,confirming that EMC-rich formulations yield lower interfacialresistance, consistent with XPS findings. Taken together, thesedifferences in the intrinsic reductive stability between DMCand EMC rationalize why EMC-rich electrolytes moreeffectively facilitate NaF-rich SEI formation.The chemical composition of CEI was analyzed in a similarmanner. In the absence of FEC (d = 0), CEI layers showed∼15% fluorine and oxygen with no major differences betweenconditions. However, in FEC-containing systems (NaFSI: w/FEC and High-con NaFSI: w/FEC), the fluorine contentincreased to ∼25%, again suggesting NaF-rich CEI formation.These results confirm that the presence of FEC facilitates thegeneration of NaF-rich interphases on both electrodes,contributing to enhanced electrochemical stability. Takingtogether, these results suggest the following interfacialformation mechanisms governing SIB performance: (1) Onthe cathode side, the presence of FEC is sufficient to promoteNaF-rich CEI formation, regardless of the solvent or saltconcentration. (2) On the anode side, the formation of a NaF-rich SEI requires the synergistic combination of high NaFSIconcentration, the presence of FEC, and EMC as the primarysolvent. Considering the fact that the LiF-rich SEI and CEIhave beneficial effects on the performance of LiBs,22−26 theNaF-rich SEI/CEI is also considered to play a crucial role insuppressing side reactions, stabilizing electrode interfaces, andenabling high CE over long-term cycling in SIBs.In optimizing electrolyte formulations, it is essential toconsider not only SEI/CEI formation but also additionalfactors, such as ionic conductivity and material cost. Althoughionic conductivity measurements were not included in thehigh-throughput platform, we conducted ionic conductivitytests for eight representative electrolyte compositions. Theresults show that DMC-rich electrolytes exhibit higher ionicconductivity than EMC-rich electrolytes, which can beattributed to the lower viscosity of DMC compared withEMC (Table S5). The introduction of FEC slightly decreasesthe ionic conductivity, consistent with the higher viscosity ofFEC relative to DMC and EMC. Increasing the NaFSIconcentration also leads to reduced conductivity. Conse-quently, the electrolyte composition that demonstrated thebest cycling performance actually exhibited the lowest ionicconductivity among the tested formulations. This reflects atypical trade-off between long-term cycle stability and ratecapability. Therefore, an EMC-rich electrolyte offers advan-tages not only in electrochemical performance but also in cost-effectiveness, providing an additional practical benefit.In this study, our high-throughput experimental platformcombined with data-driven analysis successfully enabled therapid screening of a broad range of mixed-solvent electrolytecompositions for SIBs. We identified key compositionalparameters that influence cycling stability, particularly thecrucial role of high NaFSI concentrations and FEC additions inpromoting stable Coulombic efficiency and long-term perform-ance. The formation of NaF-rich SEI and CEI was revealed as apivotal factor underpinning the enhanced electrochemicalstability, especially in EMC-based electrolytes. These findingsprovide valuable design guidelines for future electrolyteoptimization in SIBs, demonstrating the strength of integratingautomated experimentation with machine learning foraccelerating materials discovery.■ CONCLUSIONSIn this study, we established a high-throughput experimentalworkflow for the systematic evaluation of electrolytecompositions in SIBs, enabling the rapid screening of 132mixed-solvent formulations. By combining automated liquidhandling with parallel electrochemical testing, we efficientlyidentified key compositional factors, namely, NaFSI concen-tration, FEC content, and the choice of cosolvent (DMC orEMC) that significantly influence Coulombic efficiency andlong-term cycling stability. Machine learning analysis revealedthat higher salt concentrations and the inclusion of FEC arecritical for achieving a stable cycling performance. Furtherelectrochemical and XPS analyses showed that the superiorperformance of FEC-containing EMC-based electrolytes isattributable to the formation of NaF-rich SEI and CEI layers,which enhance interfacial stability. Our findings providepractical guidelines for designing high-performance electrolytesystems for SIBs and demonstrate the power of data-driven,high-throughput approaches in accelerating electrolyte devel-opment for next-generation energy storage technologies.■ ASSOCIATED CONTENT*sı Supporting InformationThe Supporting Information is available free of charge athttps://pubs.acs.org/doi/10.1021/acsaem.5c03028.Photographic image of experimental setup; compositionof mother solution; confusion matrix for random forestclassifier-based predictions; profile of capacity retention;Nyquist plots; and ionic conductivity of electrolyte(PDF)■ AUTHOR INFORMATIONCorresponding AuthorsRyo Tamura − Center for Basic Research on Materials,National Institute for Materials Science, Tsukuba, Ibaraki305-0044, Japan; orcid.org/0000-0002-0349-358X;Email: TAMURA.Ryo@nims.go.jpShoichi Matsuda − Center for Green Research on Energy andEnvironmental Materials, National Institute for MaterialScience, Tsukuba, Ibaraki 305-0044, Japan; orcid.org/0000-0002-0640-3404; Email: MATSUDA.Shoichi@nims.go.jpAuthorsManai Ono − Center for Green Research on Energy andEnvironmental Materials, National Institute for MaterialScience, Tsukuba, Ibaraki 305-0044, JapanMisato Takahashi − Center for Green Research on Energy andEnvironmental Materials, National Institute for MaterialScience, Tsukuba, Ibaraki 305-0044, JapanComplete contact information is available at:https://pubs.acs.org/10.1021/acsaem.5c03028NotesThe authors declare no competing financial interest.■ ACKNOWLEDGMENTSThe present work was supported by JST COI-NEXT(JPMJPF2016) and the Ministry of Education, Culture, Sports,Science, and Technology (MEXT) Program: Data Creationand Utilization Type Materials Research and DevelopmentACS Applied Energy Materials www.acsaem.org Articlehttps://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−14111410https://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?goto=supporting-infohttps://pubs.acs.org/doi/suppl/10.1021/acsaem.5c03028/suppl_file/ae5c03028_si_001.pdfhttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Ryo+Tamura"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://orcid.org/0000-0002-0349-358Xmailto:TAMURA.Ryo@nims.go.jphttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Shoichi+Matsuda"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://orcid.org/0000-0002-0640-3404https://orcid.org/0000-0002-0640-3404mailto:MATSUDA.Shoichi@nims.go.jpmailto:MATSUDA.Shoichi@nims.go.jphttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Manai+Ono"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://pubs.acs.org/action/doSearch?field1=Contrib&text1="Misato+Takahashi"&field2=AllField&text2=&publication=&accessType=allContent&Earliest=&ref=pdfhttps://pubs.acs.org/doi/10.1021/acsaem.5c03028?ref=pdfwww.acsaem.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-asProject (JPMXP1121467561). The author also acknowledgesthe support from the National Institute for Materials Science(NIMS) Battery Research Platform.■ REFERENCES(1) Yabuuchi, N.; Kubota, K.; Dahbi, M.; Komaba, S. ResearchDevelopment on Sodium-Ion Batteries. Chem. Rev. 2014, 114 (23),11636−11682.(2) Zhang, F.; He, B.; Xin, Y.; Zhu, T.; Zhang, Y.; Wang, S.; Li, W.;Yang, Y.; Tian, H. Emerging Chemistry for Wide-TemperatureSodium-Ion Batteries. Chem. Rev. 2024, 124 (8), 4778−4821.(3) Pei, B.; Yu, H.; Zhang, L.; Fang, G.; Zhou, J.; Cao, X.; Liang, S.Hard Carbon for Sodium-Ion Batteries: From Fundamental Researchto Practical Applications. Adv. Mater. 2025, 37 (39), No. 2504574.(4) Zhang, J.; Ma, S.; Zhang, J.; Zhang, J.; Wang, X.; Wen, L.; Tang,G.; Hu, M.; Wang, E.; Chen, W. Critical Review on CathodeElectrolyte Interphase towards Stabilization for Sodium-Ion Batteries.Nano Energy 2024, 128, No. 109814.(5) Ng, M.-F.; Zhao, J.; Yan, Q.; Conduit, G. J.; Seh, Z. W.Predicting the State of Charge and Health of Batteries Using Data-Driven Machine Learning. Nat. Mach. Intell. 2020, 2 (3), 161−170.(6) Si, Q.; Matsuda, S.; Yamaji, Y.; Momma, T.; Tateyama, Y. Data-Driven Cycle Life Prediction of Lithium Metal-Based RechargeableBattery Based on Discharge/Charge Capacity and RelaxationFeatures. Adv. Sci. 2024, 11 (33), 2402608.(7) Nishijima, M.; Ootani, T.; Kamimura, Y.; Sueki, T.; Esaki, S.;Murai, S.; Fujita, K.; Tanaka, K.; Ohira, K.; Koyama, Y.; Tanaka, I.Accelerated Discovery of Cathode Materials with Prolonged CycleLife for Lithium-Ion Battery. Nat. Commun. 2014, 5 (1), 4553.(8) Jalem, R.; Kanamori, K.; Takeuchi, I.; Nakayama, M.; Yamasaki,H.; Saito, T. Bayesian-Driven First-Principles Calculations forAccelerating Exploration of Fast Ion Conductors for RechargeableBattery Application. Sci. Rep. 2018, 8 (1), 5845.(9) Yan, P.; Fischer, M.; Martin, H.; Wölke, C.; Krishnamoorthy, A.N.; Cekic-Laskovic, I.; Diddens, D.; Winter, M.; Heuer, A. Non-Aqueous Battery Electrolytes: High-Throughput Experimentation andMachine Learning-Aided Optimization of Ionic Conductivity. J.Mater. Chem. A 2024, 12 (30), 19123−19136.(10) Rahmanian, F.; Vogler, M.; Wölke, C.; Yan, P.; Fuchs, S.;Winter, M.; Cekic-Laskovic, I.; Stein, H. S. Conductivity Experimentsfor Electrolyte Formulations and Their Automated Analysis. Sci. Data2023, 10 (1), 43.(11) Matsuda, S.; Nishioka, K.; Nakanishi, S. High-ThroughputCombinatorial Screening of Multi-Component Electrolyte Additivesto Improve the Performance of Li Metal Secondary Batteries. Sci. Rep.2019, 9 (1), 6211.(12) Matsuda, S.; Lambard, G.; Sodeyama, K. Data-DrivenAutomated Robotic Experiments Accelerate Discovery of Multi-Component Electrolyte for Rechargeable Li&#x2013;O2 Batteries.Cell Rep. Phys. Sci. 2022, 3 (4), No. 100832.(13) Tamura, R.; Tsuda, K.; Matsuda, S. NIMS-OS: An AutomationSoftware to Implement a Closed Loop between Artificial Intelligenceand Robotic Experiments in Materials Science. Sci. Technol. Adv.Mater. Methods 2023, 3 (1), 2232297.(14) Matsuda, S.; Takahashi, M. Multichannel Electrochemical Celland Liquid-Handling Dispenser for High-Throughput CombinatorialScreening of Multicomponent Electrolytes for Advanced Lithium-IonBatteries. Batter. Supercaps 2025, 8 (8), No. e202400777.(15) Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.;Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.;Dubourg, V.; Vanderplas, J.; Passos, A.; Cournapeau, D.; Brucher, M.;Perrot, M.; Duchesnay, É. Scikit-Learn: Machine Learning in Python.J. Mach. Learn. Res. 2011, 12 (85), 2825−2830.(16) Tasaki, K. Solvent Decompositions and Physical Properties ofDecomposition Compounds in Li-Ion Battery Electrolytes Studied byDFT Calculations and Molecular Dynamics Simulations. J. Phys.Chem. B 2005, 109 (7), 2920−2933.(17) Hirsh, H. S.; Sayahpour, B.; Shen, A.; Li, W.; Lu, B.; Zhao, E.;Zhang, M.; Meng, Y. S. Role of Electrolyte in Stabilizing Hard Carbonas an Anode for Rechargeable Sodium-Ion Batteries with Long CycleLife. Energy Storage Mater. 2021, 42, 78−87.(18) Komaba, S.; Ishikawa, T.; Yabuuchi, N.; Murata, W.; Ito, A.;Ohsawa, Y. Fluorinated Ethylene Carbonate as Electrolyte Additivefor Rechargeable Na Batteries. ACS Appl. Mater. Interfaces 2011, 3(11), 4165−4168.(19) Fondard, J.; Irisarri, E.; Courreg̀es, C.; Palacin, M. R.;Ponrouch, A.; Dedryver̀e, R. SEI Composition on Hard Carbon inNa-Ion Batteries After Long Cycling: Influence of Salts (NaPF6,NaTFSI) and Additives (FEC, DMCF). J. Electrochem. Soc. 2020, 167(7), 70526.(20) Takada, K.; Yamada, Y.; Watanabe, E.; Wang, J.; Sodeyama, K.;Tateyama, Y.; Hirata, K.; Kawase, T.; Yamada, A. Unusual PassivationAbility of Superconcentrated Electrolytes toward Hard CarbonNegative Electrodes in Sodium-Ion Batteries. ACS Appl. Mater.Interfaces 2017, 9 (39), 33802−33809.(21) Li, J.; Huang, S.; Yu, P.; Lv, Z.; Wu, K.; Li, J.; Ding, J.; Zhu, Q.;Xiao, X.; Nan, J.; Zuo, X. Unraveling the Underlying Mechanism ofGood Electrochemical Performance of Hard Carbon in PC/EC−Based Electrolyte. J. Colloid Interface Sci. 2024, 657, 653−663.(22) Chen, S.; Huang, L.; Wen, X.; Chen, Q.; Xia, Z.; Li, S.; Wang,H.; Xu, M.; Li, W. Formation Mechanism and Regulation of LiF in aSolid Electrolyte Interphase on Graphite Anodes in CarbonateElectrolytes. J. Phys. Chem. C 2023, 127 (24), 11462−11471.(23) Baek, M.; Kim, J.; Jin, J.; Choi, J. W. Photochemically DrivenSolid Electrolyte Interphase for Extremely Fast-Charging Lithium-IonBatteries. Nat. Commun. 2021, 12 (1), 6807.(24) Bai, P.; Ji, X.; Zhang, J.; Zhang, W.; Hou, S.; Su, H.; Li, M.;Deng, T.; Cao, L.; Liu, S.; He, X.; Xu, Y.; Wang, C. Formation of LiF-Rich Cathode-Electrolyte Interphase by Electrolyte Reduction. Angew.Chemie Int. Ed. 2022, 61 (26), No. e202202731.(25) Llanos, P. S.; Ahaliabadeh, Z.; Miikkulainen, V.; Lahtinen, J.;Yao, L.; Jiang, H.; Kankaanpää, T.; Kallio, T. M. High Voltage CyclingStability of LiF-Coated NMC811 Electrode. ACS Appl. Mater.Interfaces 2024, 16 (2), 2216−2230.(26) Chen, L.; Nian, Q.; Ruan, D.; Fan, J.; Li, Y.; Chen, S.; Tan, L.;Luo, X.; Cui, Z.; Cheng, Y.; Li, C.; Ren, X. High-Safety and High-Efficiency Electrolyte Design for 4.6 V-Class Lithium-Ion Batterieswith a Non-Solvating Flame-Retardant. Chem. Sci. 2023, 14 (5),1184−1193.ACS Applied Energy Materials www.acsaem.org Articlehttps://doi.org/10.1021/acsaem.5c03028ACS Appl. Energy Mater. 2026, 9, 1405−14111411https://doi.org/10.1021/cr500192f?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/cr500192f?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acs.chemrev.3c00728?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acs.chemrev.3c00728?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1002/adma.202504574https://doi.org/10.1002/adma.202504574https://doi.org/10.1016/j.nanoen.2024.109814https://doi.org/10.1016/j.nanoen.2024.109814https://doi.org/10.1038/s42256-020-0156-7https://doi.org/10.1038/s42256-020-0156-7https://doi.org/10.1002/advs.202402608https://doi.org/10.1002/advs.202402608https://doi.org/10.1002/advs.202402608https://doi.org/10.1002/advs.202402608https://doi.org/10.1038/ncomms5553https://doi.org/10.1038/ncomms5553https://doi.org/10.1038/s41598-018-23852-yhttps://doi.org/10.1038/s41598-018-23852-yhttps://doi.org/10.1038/s41598-018-23852-yhttps://doi.org/10.1039/D3TA06249Jhttps://doi.org/10.1039/D3TA06249Jhttps://doi.org/10.1039/D3TA06249Jhttps://doi.org/10.1038/s41597-023-01936-3https://doi.org/10.1038/s41597-023-01936-3https://doi.org/10.1038/s41598-019-42766-xhttps://doi.org/10.1038/s41598-019-42766-xhttps://doi.org/10.1038/s41598-019-42766-xhttps://doi.org/10.1016/j.xcrp.2022.100832https://doi.org/10.1016/j.xcrp.2022.100832https://doi.org/10.1016/j.xcrp.2022.100832https://doi.org/10.1080/27660400.2023.2232297https://doi.org/10.1080/27660400.2023.2232297https://doi.org/10.1080/27660400.2023.2232297https://doi.org/10.1002/batt.202400777https://doi.org/10.1002/batt.202400777https://doi.org/10.1002/batt.202400777https://doi.org/10.1002/batt.202400777https://doi.org/10.1021/jp047240b?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/jp047240b?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/jp047240b?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1016/j.ensm.2021.07.021https://doi.org/10.1016/j.ensm.2021.07.021https://doi.org/10.1016/j.ensm.2021.07.021https://doi.org/10.1021/am200973k?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/am200973k?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1149/1945-7111/ab75fdhttps://doi.org/10.1149/1945-7111/ab75fdhttps://doi.org/10.1149/1945-7111/ab75fdhttps://doi.org/10.1021/acsami.7b08414?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acsami.7b08414?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acsami.7b08414?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1016/j.jcis.2023.12.022https://doi.org/10.1016/j.jcis.2023.12.022https://doi.org/10.1016/j.jcis.2023.12.022https://doi.org/10.1021/acs.jpcc.3c02731?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acs.jpcc.3c02731?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acs.jpcc.3c02731?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1038/s41467-021-27095-whttps://doi.org/10.1038/s41467-021-27095-whttps://doi.org/10.1038/s41467-021-27095-whttps://doi.org/10.1002/anie.202202731https://doi.org/10.1002/anie.202202731https://doi.org/10.1021/acsami.3c14394?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1021/acsami.3c14394?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-ashttps://doi.org/10.1039/D2SC05723Ahttps://doi.org/10.1039/D2SC05723Ahttps://doi.org/10.1039/D2SC05723Awww.acsaem.org?ref=pdfhttps://doi.org/10.1021/acsaem.5c03028?urlappend=%3Fref%3DPDF&jav=VoR&rel=cite-as