X-CEL: eXtreme Fast Charge Cell Evaluation of Lithium-ion Batteries

February 2019 Report Period: October–December 2018

2

Table of Contents

Project Overview ................................................................................................................................................. 3

XCEL R&D: CAMP, Testing & Post-test Characterization, and Modeling (Argonne National Laboratory) ...................................................................................................... 4

XCEL R&D: Extreme Fast Charging (National Renewable Energy Laboratory)................................ 10 XCEL R&D: Electrochemical, Atomistic, and Techno-Economic (BatPaC)

(Argonne National Laboratory) .................................................................................................... 18 XCEL R&D: Performance Characterization and Post-Test Analysis

(Argonne National Laboratory) .................................................................................................... 28 XCEL R&D: Battery Testing Activities (Idaho National Laboratory) ................................................ 31 XCEL R&D: SLAC National Accelerator Laboratory ......................................................................... 34 XCEL R&D: Lawrence Berkeley National Laboratory ....................................................................... 36

3

Project Overview

Venkat Srinivasan PI (Argonne National Laboratory) and Samuel Gillard (Department of Energy)

DOE-EERE has identified fast charge as a critical challenge in ensuring mass adoption of electric vehicles with a goal of 15-min recharge time. Present day high energy cells with graphite anodes and transition metal cathodes in a liquid electrolyte are unable to achieve this without negatively affecting battery performance. There are numerous challenges that limit such extreme fast charging at the cell level, including Li plating, rapid temperature rise, and possible particle cracking. Of these, Li plating is thought to be the primary culprit. This project aims to gain an understanding of the main limitations during fast charge using a combined approach involving cell builds, testing under various conditions, characterization, and continuum scale mathematical modeling. Expertise from five National Labs is utilized to make progress in the project. Cells are built at the Cell Analysis, Modeling, and Prototyping (CAMP) facility at Argonne National Laboratory (ANL) using various carbons and different cell designs, in both half-cell and full cell configuration and with reference electrodes. Cells are tested at both Idaho National Laboratory (INL) and ANL under various operating conditions (c-rate, temperature) and under different charging protocols with the aim of identifying the onset of plating, quantifying the extent of the problem, and determining parameters and test data for mathematical models. Tested cells are opened and various advanced characterizing performed at ANL to determine the extent of plating and to determine if other failure models, such as particle cracking, also play a role. A critical part of the project is the use of continuum scale mathematical models to understand the limitations at high charge rates and, therefore, suggest possible solutions that can be pursued. Both macro-scale approaches and microstructure-based simulations are pursued and serve to complement each other. Macromodeling at National Renewable Energy Laboratory (NREL) is used to test cell designs, accompanied by microstructure models to provide deeper insights into the phenomenon in the battery. This is complemented with development of models incorporating of new physics, such as phase change and solid-electrolyte interphase growth, at ANL. Two exploratory projects aim to study ways to detect Li in situ during operation. NREL is pursuing the use of microcalorimetry to detect heat signatures during plating. INL is working with Princeton University to examine the use of acoustic methods to determine if plating leads to a signature in the acoustic signal. Finally, SLAC National Acceleratory Laboratory is using synchrotron X-ray methods to guide cell design and charging protocols of extreme fast-charging Li-ion battery cells, and Lawrence Berkeley National Laboratory is investigating the initial onset of Li plating during fast charging and developing a strategy to detect it.

4

XCEL R&D: CAMP, Testing & Post-test Characterization, and Modeling (Argonne National Laboratory)

Contributors: Alison Dunlop, Andrew Jansen, Dave Kim, Bryant Polzin, and Steve Trask (Argonne)

Project Introduction The Cell Analysis, Modeling, and Prototyping (CAMP) Facility continues to support the XCEL Program’s objective of identifying and mitigating causes of lithium plating at fast charge (>4C) in single-layer pouch cells. Efforts in FY18 demonstrated that the choice in graphite did not significantly affect the ability to cycle under fast charge conditions (6C charge, C/2 discharge) - all six of the selected graphites were able to achieve 750 cycles with 80% capacity retention, using a 2 mAh/cm² graphite loading. A decision was made to use SLC1506T graphite from Superior Graphite as the baseline graphite material, and NMC532 as the baseline cathode material. Over 70 single-sided single-layer pouch cells were fabricated in the Round 1 cell build using a 2 mAh/cm² graphite loading and delivered to lab partners (INL, Argonne, and NREL) for fast charge testing with a nominal capacity of 19 mAh capacity. Prescreening of anode-cathode pairs with varying electrode capacity loading indicated that loadings over ~2.5 mAh/cm² were not able to charge at a true 6C rate. Thus, the next pouch cell build was designed with a graphite loading of 3.0 mAh/cm². A total of 48 of these Round 2 pouch cells were delivered to Argonne, INL, and NREL for testing. These two cell builds are now considered the baselines for the XCEL Program.

Objectives The goal of the FY19 work is to explore methods of preventing lithium plating via modifications of the electrode architecture. Ideally, the negative and positive electrodes should have low tortuosity to enable fast lithium ion transport to and from the active material closest to the current collector, while maintaining low porosity to maintain high energy density. In addition, the CAMP Facility will support the DOE-EERE-VTO FOA and Lab Call projects and teams that are focused on developing methods of detecting lithium plating.

Approach In FY18, the CAMP Facility developed two single-layer pouch cells to benchmark the fast charge capabilities of a typical lithium-ion battery, with the main difference between the cells being electrode loading. These two electrode/cell designs will be used as the baseline designs for this program in FY19. The details of the two cell designs are as follows: Round 1 Pouch Cells were assembled with 14.1 cm² single-sided cathodes (0.145 grams of NMC532 per pouch cell) and 14.9 cm² single-sided graphite anodes (SLC1506T from Superior Graphite) using Celgard 2320 separator (20 µm, PP/PE/PP) and 0.5 mL of Tomiyama 1.2 M LiPF6 in EC:EMC (3:7 wt%) “Gen2” electrolyte. The n:p ratio is between 1.12 to 1.22 for this voltage window (3.0 to 4.1 V). After assembly, the pouch cells underwent formation cycles at ~4 psi in the 3.0 to 4.1 V window as follows: 1.5 V tap charge and hold for 15 minutes, followed by a 12 hour rest, and then 3 cycles at C/10, followed by 3 cycles at C/2. The cells were then brought to a safe state of charge by constant voltage charging to 3.5 V for 6 hours, and then degassed and prepared for shipping/delivery to the battery test labs. A nominal C/3 capacity of 19 mAh was recommend for future tests.

5

Round 2 Pouch Cells were assembled with 14.1 cm² single-sided cathodes (0.236 grams of NMC532 per pouch cell) and 14.9 cm² single-sided graphite anodes (SLC1506T from Superior Graphite) using Celgard 2320 separator (20 µm, PP/PE/PP) and 0.615 mL of Tomiyama 1.2 M LiPF6 in EC:EMC (3:7 wt%) “Gen2” electrolyte for an electrolyte-to-pore volume factor of 4.20. The n:p ratio is between 1.07 to 1.16 for this voltage window (3.0 to 4.1 V). Figure 1 details the electrode composition and design parameters for the Round 2 electrodes. Formation process was the same as in Round 1 cells. A nominal C/2 capacity of 32 mAh was recommend for future tests.

Figure 1. Electrode composition and design parameters for Round 2 pouch cell design. Round 1 electrodes were with the

same composition but less mass loading (and thickness).

Results Development of Structured Graphite Electrodes – Freeze Casting by LBNL Freeze casting of the negative electrode is a unique method to create an electrode structure with very low, almost ideal, tortuosity. Lawrence Berkeley National Laboratory (LBNL) has this capability and has agreed to create electrode samples using the SLC1506T graphite material, which Argonne then sent to them. LBNL was able to make a trial coating with the SLC1506T graphite onto copper foil (see Figure 2), which they then sent to Argonne for testing. As can be seen in the SEM photo in Figure 2, this structured electrode has near perfect tortuosity with straight electrolyte paths to the copper current collector. The porosity and adhesion to the foil are the main concern for this first trial sample. A thickness measurement was made by slowly and gently lowering the head of the thickness gauge tip. The punched electrodes had a few bare foil spots showing because the coating flaked off near the edges during the punching process. These are issues that future freeze casting development will address. A comparison of the electrode properties is made in Table 1. Coin cells were assembled with 13 mm diameter LBNL freeze-cast graphite electrodes versus lithium metal (15.6 mm diameter) using Celgard 2320 separator (20 µm, PP/PE/PP) and Tomiyama 1.2 M LiPF6 in EC:EMC (3:7 wt%) “Gen2” electrolyte. Two duplicate coin cells were made and were then cycled in the 3.0 to 4.1V window with 3 formation cycles at C/10, followed by a rate performance test at C/10, C/5, C/2, 1C, and 2C delithiation and C/10 or C/5 lithiation cycles.

Cathode: LN3107-189-3 90 wt% Toda NMC532 5 wt% Timcal C45 5 wt% Solvay 5130 PVDF

Matched for 4.1V full cell cycling Prod:NCM-04ST, Lot#:7720301 Single-sided coating, CFF-B36 cathode Al Foil Thickness: 20 µm Al Foil Loading: 5.39 mg/cm2 Total Electrode Thickness: 91 µm Coating Thickness: 71 µm Porosity: 35.4 % Total Coating Loading: 18.63 mg/cm2 Total Coating Density: 2.62 g/cm3

Made by CAMP Facility

Anode: LN3107-190-4A 91.83 wt% Superior Graphite SLC1506T 2 wt% Timcal C45 carbon 6 wt% Kureha 9300 PVDF Binder 0.17 wt% Oxalic Acid

Lot#: 573-824, received 03/11/2016 Single-sided coating, CFF-B36 anode

Cu Foil Thickness: 10 µm Total Electrode Thickness: 80 µm Total Coating Thickness: 70 µm Porosity: 34.5 % Total SS Coating Loading: 9.94 mg/cm2 Total SS Coating Density: 1.42 g/cm3

Made by CAMP Facility

6

Figure 2. SEM cross-section (left) of first trial batch of LBNL freeze-cast electrodes (right) made with SLC1506T graphite.

Table 1: Comparison of Anode Formulations.

Superior Graphite SLC1506T Electrodes

Total Electrode Thickness

(µm)

Bare Cu Thickness

(µm)

Coating Thickness

(µm)

Total Coating Loading

(mg/cm2)

Total Coating Density (g/cm3)

Porosities (%)

Formation Voltage Window

LBNL Freeze Cast 300 15 285 9.79 0.34 ~80.0% 0.01-1.5V

A-A015 (Round 1) 57 10 47 6.38 1.36 ~37.4% 0.00-1.5V

LN3107-190-4A (Round 2) 80 10 70 9.94 1.42 ~34.5% 0.00-1.5V

An analysis of the delithiation capacity and coulombic efficiency during the formation process for the freeze-cast electrode compared to the Round 1 and Round 2 graphite electrode data (Figure 3) reveals no significant difference between these different electrode designs. However, there was significant loss of capacity at the faster delithiation rates, as can be seen in Figure 4. This is most likely due to the high porosity (less particle-to-particle contact) and the weak adhesion to the copper foil. LBNL is using this feedback to make improvements to their freeze-cast development. One area they are exploring is the use of graphite particles that have higher aspect ratio. The CAMP Facility sent LBNL several more graphite samples to aid in this study.

7

Figure 3. Delithiation capacity and Coulombic efficiency during formation process for LBNL freeze-cast graphite electrode compared to Round 1 and Round 2 graphite electrodes. Electrodes were cycled in coin cells (2320 format) versus lithium

at 30°C.

Figure 4. Delithiation capacity for LBNL freeze-cast graphite electrode compared to Round 1 and Round 2 graphite

electrodes at various current rates. Electrodes were cycled in coin cells (2320 format) versus lithium at 30°C.

Support of Electrolyte Development at INL Lithium plating can be minimized with the use of electrolytes with high ionic conductivity. INL is exploring several electrolyte formulations that are predicted to have ionic conductivity better than the baseline Gen2 electrolyte. INL is currently scoping out these electrolytes in coin cells, with the plan that the electrolytes that pass the initial tests be explored further in pouch cells. The CAMP Facility has supported this effort by fabricating 11 Round 2 pouch cells and shipping them to INL without electrolyte added. INL will add their candidate electrolyte when ready.

8

Support of Lithium Plating Studies at NREL NREL (Donal Finegan) is developing methods of detecting lithium plating on graphite using the synchrotron at the University of College London. In this approach, an ideal electrode would be over 100 microns thick using large graphite particles (larger than the SLC1506T baseline graphite). The CAMP Facility supported this effort by fabricating a 101 micron graphite electrode using SLC1520P graphite from Superior Graphite. This graphite has an average particle diameter near 20 microns versus the 6 microns of the baseline graphite. Also needed was a thick capacity-matched cathode, which the CAMP Facility also developed using NMC622. These electrode sheets were delivered to NREL in time for their beamtime in London. A summary of the electrode properties is provided in Figure 5.

Figure 5. Electrode composition and design parameters of thick matching electrodes for NREL lithium plating detection

study using SLC1520P graphite and NMC622.

Support of Lithium Plating Detection Teams – SLAC/Stanford University The CAMP Facility supported the lithium-plating detection activities at Stanford Linear Accelerator Center (SLAC) and Stanford University by supplying several graphite powder samples and electrode sheets, and fabricating several pouch cells in time for their beamtime experiments at Argonne’s Advanced Photon Source (APS). A list of items supplied in this quarter include:

· 200 grams each of SLC1520P, SLC1506T, and MagE graphites · 5 sheets each of Round 1 positive and negative electrodes · 5 sheets each of Round 2 positive and negative electrodes · 2 formed Round 2 pouch cells · 1 dry Round 2 pouch cell · 1 tap-charge only Round 2 pouch cell · 1 lithium foil in pouch

Support of Lithium Plating Detection Teams – Princeton University The CAMP Facility supported the lithium-plating detection activities at Princeton University by fabricating and supplying 9 formed Round 2 pouch cells. Support of Lithium Plating Detection Teams – UC Berkeley The CAMP Facility supported the lithium-plating detection activities at the University of California – Berkeley by supplying 1 kg of SLC1506T graphite powder, 10 sheets of Round 1 and 10 sheets of Round 2 negative electrode sheets, and 5 tap-charge-only low n:p ratio pouch cells. The low n:p ratio pouch cells consisted of

9

pairing the Round 1 negative electrode with the Round 2 positive electrode, which results in a n:p ratio between 0.65 and 0.77. These cells were designed with the idea to guarantee lithium plating within a few cycles. Coin cells were also made by the CAMP Facility with this electrode couple to confirm that lithium plating is present. One of these coin cells was cycled with one C/10 formation cycle and then opened up in a discharge state – no visual signs of lithium plating were observed. Another of these coin cells was cycled with one C/10 formation cycle, five 6C charge cycles, and it was then opened up in a discharge state – there was ample evidence of lithium plating on the graphite electrode (see Figure 6).

Figure 6. Photos of graphite electrode from low n:p ratio coin cells with one C/10 formation cycle (left) and one C/10

formation followed by five 6C cycles (right).

Conclusions The CAMP Facility devoted a large portion of its effort to supporting several fast charge teams this quarter. Numerous graphite powders, electrode sheets, and pouch cells were delivered to XCEL team members and to teams in the DOE Lab Call that are developing Lithium Plating Detection Methods. Many of the cell builds were fabricated “just-in-time” for beamtime experiments at Argonne’s Advanced Photon Source and other synchrotrons in the United States and Europe. Technical data and electrochemical results were provided to all team members as needed to aid in their experiments and modeling efforts. Several of these activities required new prototype cell developments such as fabricating capacity-matched high-loading (thickness > 100 µm) electrodes for NREL, and low n:p ratio cells for UC-Berkeley. Testing of the first trial experiments in freeze-cast graphite electrode from LBNL indicated that the resulting tortuosity is nearly ideal, but the porosity and adhesion to the copper foil need improvement. In the next quarter, the CAMP Facility will complete fabrication of the Round 2 pouch cells with varying anode porosity. These pouch cells will be distributed to INL, NREL, and Argonne for testing and analysis.

Key Publications 2 papers submitted with INL and NREL

Acknowledgements Key contributors to this work include: Alison Dunlop, Andrew Jansen, Dave Kim, Bryant Polzin, and Steve Trask (all from Argonne National Laboratory).

10

XCEL R&D: Extreme Fast Charging (National Renewable Energy Laboratory)

Matthew Keyser, Kandler Smith, Shriram Santhanagopalan, Francois Usseglio-Viretta, Weijie Mai, and Andrew Colclasure (NREL)

Background This quarter the NREL team improved macro-homogeneous model predictions with fast charge experimental test data from Idaho and Argonne National Laboratories by including non-isothermal behavior in the model. For charge rates greater than 3C, temperature rise coupled with more facile transport/reaction rates at elevated temperature improved model-to-data agreement. A final uncertainty in modeling high-rate current/voltage performance is the graphite solid-state diffusivity at high rates, which shows an apparent increase in diffusivity with rate. Argonne is performing first-principles modeling to understand this behavior. Experimental and modeling analysis in the previous fiscal year showed electrolyte transport to be the dominant contributor to voltage polarization and onset of the lithium plating side reaction when Li-ions in the electrolyte become depleted. To alleviate these transport limitations, this quarter the NREL team investigated electrode architectures that incorporate a secondary pore network (SPN) to reduce effective tortuosity of electrolyte transport across both electrodes’ thickness. Unlike experimental studies in which porosity and, hence, energy density varies across baseline and pore-channel prototypes, the modeling study provides a clean prediction of benefits. The modeling investigation shows up to 40% improved charge-rate acceptance with optimized secondary pores in both the anode and cathode. If only one electrode is to incorporate secondary pore channels, the cathode appears to be more advantageous. This is due to sluggish electrolyte transport at the high electrolyte concentrations that occur in the cathode. Experimental data typically show earlier onset of lithium plating than 1D macro-homogeneous models would predict. To begin to understand this phenomena, NREL reviewed literature on heterogeneities not considered in the macro-homogeneous model. NREL and Purdue University further performed modeling studies at the meso- and micro-length scales to predict early Li plating onset due to electrode coating heterogeneity and particle size/morphology heterogeneity. Each is predicted to cause Li-plating onset 2-5% state-of-charge earlier than the macro model would predict. Other early-onset heterogeneities not yet investigated by the team include graphite sub-particle crystallographic heterogeneity and 3D effects in large cells due to temperature and current collector potential gradients. Battery designers must carefully consider all of these length scales to effectively suppress the unwanted Li plating side reaction during fast charging.

11

Results Figure 1 illustrates three main transport limitations that can occur during XFC and their consequences. During charging lithium ions de-intercalate from the NMC positive electrode active particles and transfer into the electrolyte phase. These lithium ions migrate and diffuse from the positive through the separator and into the negative electrode where they intercalate into the active graphite particles. At the high current densities required to support XFC of >10 mA/cm2, large concentration and electrolyte potential gradients develop to drive the necessary flux of lithium ions within the electrolyte. Sustained fast charge can result in electrolyte Li-ion depletion within the negative and/or saturation within the positive electrode. These electrolyte transport limits can also occur at lower charge rates at low temperature and/or for cell designs when electrolyte conductivity is low, electrolyte diffusivity is low, electrode tortuosity is high, porosity is low, and/or electrode thickness is high. Due to limited solid-state diffusivity for graphite and NMC, lithium ions can become saturated at the graphite surfaces and depleted at the NMC surfaces. Saturation of Li ions at the graphite particle surface along with sluggish kinetics can result in lithium plating, causing cell degradation and safety concerns. Macro-homogeneous Model Enhanced to Capture Non-isothermal Impact of Fast Charge

Figure 2. (Left) Comparison of model predictions (solid lines) vs. experimental measurements (symbols) for cell potential of Round 2 cells during charging at rates from 1 to 9C. Round 2 cells have a cell loading of 2.5 mAh/cm2 with approximately 70-micron thick anode and cathode. (Right) Model predictions for lumped temperature of Round 2 single layer pouch cell

during charging.

NREL’s macro-homogeneous electrochemical battery model was enhanced to account for temperature rise during high rate charging by solving the lumped energy equation,

𝐶𝐶𝑝𝑝𝑑𝑑𝑇𝑇𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑑𝑑𝑑𝑑

= 𝑄𝑄 − 𝑈𝑈𝑈𝑈(𝑇𝑇𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 − 𝑇𝑇∞),

Figure 1. Schematic of the various transport limitations during XFC that can result in poor charge acceptance,

heat generation, lithium plating, and capacity loss.

12

where 𝐶𝐶𝑝𝑝 is the total heat capacity of the cell, 𝑇𝑇𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 is the lumped/average temperature of the cell, 𝑈𝑈𝑈𝑈 is the effective heat transfer coefficient for the cell in W/K, and 𝑇𝑇∞ is the ambient temperature. The total heat produced in watts 𝑄𝑄 accounts for lithium ion movement within the electrolyte phase, electron conduction within the solid electrode phase, any contact resistance, and overpotential from charge transfer chemistry. To relate physical properties to temperature, the activation energy for solid-state diffusion and exchange current density in both the negative and positive electrode were taken to be 30 kJ and 70 kJ, respectively. Electrolyte conductivity and diffusivity were also expressed as functions of temperature using previously identified relationships. The non-isothermal macro model was fitted to the Round 2 pouch cell data and is illustrated in. The model fit is improved by considering effects from temperature rise during high rate charging. Previously, the isothermal model severely under-predicted the initial charge voltage at high charge rates. Further, to match capacities for high rate charging, inputs for the isothermal model’s solid-state diffusion and exchange current density previously had to be set significantly higher than values estimated from GITT experiments. Temperature rise fully explains the observed improved transport at high C-rates with one exception. The one exception is the solid-state diffusion coefficient for graphite must be set to 5-10 times higher than the expected value in order to match charge rates above 4C. ANL is investigating the physical phenomena that result in the apparent increase in solid state diffusion of graphite with applied rate. The model was found to fit the single layer pouch cell data best with a heat transfer coefficient U of 3.5 W/m2/K. At typical charging rates of 1-2C, a single layer pouch cell remains close to isothermal even with poor heat transfer. At 6C and above, however, cell performance is strongly tied to heat transfer. Cell performance may actually be better if there is poor heat transfer resulting in significant temperature rise during fast charging. This cannot be at the detriment of safety or lifetime, however, both of which could be an issue at temperatures above 45-50oC. The non-isothermal model is important to explain cell performance for different cell sizes and heat transfer configurations. Model predictions of heat generation will be validated with NREL isothermal calorimeter data later this year. A 2D model was developed to investigate electrode architecture optimization to enable fast charging by reducing the overall electrodes’ effective tortuosity impeding electrolyte transport. Figure 3(a) shows a schematic of SPN channel in both electrodes, aligned to provide a fast through-plane transportation channel. To ensure a fair comparison, the total porosity 𝜀𝜀𝑐𝑐 of each electrode is fixed, such that after the introduction of SPN of width 𝑙𝑙, the porosity of the primary section of width 𝐿𝐿 can be determined by 𝜀𝜀𝑐𝑐′ = (1 + 𝑐𝑐

𝐿𝐿)𝜀𝜀𝑐𝑐. As the

width of SPN increases, the through-plane diffusion of Li+ is improved while the in-plane diffusion is hindered due to the reduction of porosity in the primary section. This implies an optimal configuration of SPN that allows fast in-plane and through-plane diffusions. Figure 3(b) shows that when the primary section is thin (𝐿𝐿𝑁𝑁 = 17.5 um), the in-plane diffusion is not limiting, and thus wider SPN can be used to boost through-plane diffusion, allowing the 6C capacity of the cell to be improved by ~47% compared to the baseline case. In separate results when SPN is introduced into only the anode or only the cathode, the 6C capacities are improved by ~25% and ~35%, respectively. This indicates that it is more important to improve Li+ transport in the cathode. This is because the electrolyte transport becomes sluggish at the high Li-ion concentrations which occur in the cathode during fast charge. Figure 3(c) shows cell capacity and minimum anode-interface potential versus C rate for the baseline case and optimal SPN case with cell loading is 3 mAh/cm2. Simulations are also presented for elevated temperatures, which further enhance electrolyte transport. The combination of advanced electrode architecture and 40oC – a reasonable and safe cell temperature – would allow 6C charging without lithium plating or undue aging of the cell.

13

Electrode Architecture Optimization with Secondary Pore Network (SPN)

Figure 3. (a) Schematics of the baseline and secondary pore network (SPN) models. (b) Improvement of the 6C capacity of the SPN model compared to the baseline when the thickness of anode primary section LN is 17.5 um. (c) Variations of cell

capacity and minimum anode interface potential versus C rate at different temperatures when the cell loading is 3 mAh/cm2.

Review of Heterogeneities Leading to Early Li Plating Onset While battery average response is well described by the 1D macro-homogeneous model, certain heterogeneities cause localized degradation not predicted by that model. All heterogeneities cause earlier onset of Li plating than the 1D model would predict. Wherever Li plating occurs, no matter how localized, it impacts the whole battery’s capacity performance due to reduction of Li inventory. To begin to better understand capacity fade and Li plating patterns observed in experiments, NREL performed a review of heterogeneities at various length scales. Ranging from sub-particle to cell length scale, relevant heterogeneities are: Crystal anisotropy. Active material particles are generally an agglomeration of individual grains, each grain oriented in a different, generally random direction. Crystal grains in graphite range from 0.02 to 0.3 μm [1]. Transport within graphite particles takes place between graphene sheets and is thus 2D. Lithium preferentially intercalates into edge planes compared to basal planes. Hence, part of the graphite particle surface is shut off to intercalation and is thought to cause a hotspot where saturation first occurs, and metallic Li can nucleate [1,2].

14

1) Particle size and morphology. Small particles and particles with cracks have shorter solid-state diffusion lengths and are thus preferentially used and may saturate and plate first. This effect is explored below with NREL’s microstructure model and found to cause Li plating 8-9 mV and 2.5% SOC earlier than the 1D macro model predicts.

2) Particle-to-particle contact resistance. To capture observed particle-to-particle differences in optical cell experiments, Thomas-Alyea [1] modeled the graphite electrode with 5 particles assigned contact resistances of 2 ± 1 mΩ·m2. In addition to representing particle-particle contact, the resistance represented the SEI film between particles, particle size and multi-phase behavior.

3) Electrode thickness. As predicted by the 1D macro model, the front of the graphite electrode is preferentially used due to electronic conductivity being much higher than ionic conductivity. The preferential use of the front of the electrode is greatly exacerbated if/once Li ions are depleted in the back of the electrode during fast charge.

4) Electrode coating heterogeneity. Slight variations in coating thickness or density of active material can cause heterogeneous electrode utilization. Electronic conductivity is expected to be uniform in the graphite electrode. Analysis by the Purdue University team (Prof. Partha Mukherjee) [3] based on electrode tomographic images found ~50-μm local regions of high active material density caused locally high tortuosity, and led to 5 mV early Li plating onset relative to the 1D macro model.

5) Electrode edge effects. In regions where there is excess cathode, the anode can become overcharged. To provide a margin for slight possible misalignment between anode and cathode, the anode sheet is designed to be larger, and thus to overhang the cathode sheet by 0.5-1.0 mm. To prevent ionic shorting, the separator further overhangs both anode and cathode.

6) Defects and delamination. Again, regions with excess cathode can cause the anode to become overcharged. Arnold [4] showed ~400 μm pore-closed regions of separator (or blocked electrode surface) to cause local plating at the 0.67C rate, as blocked cathode ions converged on the nearest open region of anode. Even smaller defects would presumably cause plating at the higher 6C rate. Delaminated or damaged anode regions could similarly become overcharged.

7) Large-format 3D cell design. Jellyroll temperature, pressure, and current collector potential gradients all can cause non-uniform utilization of large cells [5]. Higher temperature regions of a cell have faster kinetics and transport and thus are preferentially used. Pressure may also encourage preferential utilization to some high-pressure inflection point where pores become closed off. Potential drop as electrons travel in current collectors causes regions close to the tabs to be over-utilized and regions far from the tabs to be underutilized. The middle (hottest) region of a jellyroll and/or the region closest to the tabs may thus plate Li first.

Figure 4 summarizes the various heterogeneities leading to early plating onset. The section below further describes recent simulations investigating particle size and morphology effects.

15

Figure 4. Summary of heterogeneities that can lead to early, localized onset of Li plating in an Li-ion cell during fast charge.

Over the past year, NREL’s 3D microstructure electrochemical battery model has been numerically enhanced (~100 times faster) to allow investigation of larger fields of view. The simulation shown in Figure 5 has 4.5M degrees of freedom and simulates 6C charge in 3 days computation time on parallel processors. The 3D microstructure geometry was obtained via X-ray computed tomography performed at University College of London. Particle size and morphological variations in an electrode result in electrochemical heterogeneities not quantifiable though 1D-macroscale models. Particle size and crack density have been found to be first-order parameters which control the local lithiation, while C-rate and tortuosity factor control the magnitude of in-plane heterogeneities. Figure 6 (left) shows model predictions of heterogeneous in-plane lithium plating. Figure 6 (right) shows an earlier plating compared with the macromodel, which can be attributed to the particle heterogeneity. Local plating onset is partly correlated with local tortuosity factor. Future work will focus on (i) establishing quantitative correlation, if any, between lithium plating and local morphology features such as cracks, (ii) investigating the impact of the carbon-binder domain (CBD) on lithium plating with a heterogeneous description of the CBD rather than use of an effective parameter as done in macromodels, (iii) considering full-cell configuration and thicker electrodes, (iv) exploring sub-particle heterogeneity (crystal orientation) and its impact on lithium plating, and (v) providing guidance for Argonne fabrication of graded electrodes and electrodes utilizing bi-modal particle size distributions.

16

Particle Heterogeneity Leading to Early Onset of Lithium Plating during Fast Charge

Figure 5. Simulated intercalation fraction, x in LixC6, within SLC1506T anode nearing end of 6C charge.

Model shows particle in-plane state of charge is heterogeneous, with large particles having a lower

concentration compared with smaller particles.

Figure 6. Left: Ratio of the electrochemical active interface area with thermodynamically favorable lithium plating potential (фs – фe <0) during a 6C charge with the geometry shown in Figure 5. Model shows lithium plating propagation front is not a 2D plane but is heterogeneous. Right: Li plating potential at the separator/anode interface calculated with

the micromodel (solid lines) compared with the macromodel (dashed lines) during 6C and 4C charges. Micromodel predicts an earlier lithium plating, 8-9 mV before the macromodel.

Conclusions Non-isothermal electrochemical model quantified performance was found to increase in experiments due to temperature rise. There is still an apparent increase in graphite solid-state diffusivity with charge rate that is unexplained. The non-isothermal model is also important to explain cell performance for different cell sizes and heat transfer configurations. 2D model predictions show that secondary pore networks can increase charge capacity acceptance by around 20% when incorporated into the anode, 30% in the cathode, and 40% when incorporated in both anode and cathode. The feature size of the secondary pores and periodicity must be small, however, to achieve the optimal high energy density and fast-charge capability. This modeling study is helping to inform experimental efforts in freeze casting and magnetic alignment ongoing at LBL and Argonne.

17

There are numerous heterogeneities that can lead to early onset of Li plating. Preliminary modeling analyses were conducted at the micro- and meso-scales. Locally high tortuosity can lead to early plating at the electrode surface on a ~50-μm length scale. Particle size heterogeneity leads to 8-9 mV / 2.5% SOC early onset of Li plating relative to a 1D macro model prediction. Synchrotron experimental studies at various length scales are ongoing by multiple team members and will be used to validate these model predictions. The ultimate goal is to provide design guidelines across all length scales of Li-ion batteries to best avoid Li plating during fast charge.

Milestones and Deliverables

NREL FY19Q1 Milestone to DOE, “Simulation-based comparison of electrochemical response of Argonne graphite electrode library using 3D geometry obtained from tomography. Correlate heterogeneity of electrode utilization with electrode morphology,” December 31, 2018.

References [1] K.E. Thomas-Alyea, C. Jung, R.B. Smith, M.Z. Bazant, “In situ observation and mathematical modeling

of lithium distribution within graphite,” J. Electrochem. Soc. 164 (11) E3063-E3072, 2017. [2] S.J. Harris, E.K.Rahani, V.B. Shenoy, “Direct in situ observation and numerical simulations of non-

shrinking-core behavior in an MCMB graphite composite electrode,” J. Electrochem. Soc. 159 (9) A1501-A1507 (2012).

[3] F. Usseglio-Viretta, A. Colclasure, K. Smith, M. Keyser, A. Mistry, P.P. Mukherjee, “Simulation-based comparison of electrochemical response of Argonne graphite electrode library using 3D geometry obtained from tomography. Correlate heterogeneity of electrode utilization with electrode morphology,” NREL/DOE XCEL Program Milestone Report, December 31, 2018.

[4] X.M. Liu, A. Fang, M.P. Haataja, C.B. Arnold, “Size dependence of transport non-uniformities on localized plating in lithium-ion batteries,” J. Electrochem. Soc. 165 (5) A1147-A1155 (2018).

[5] G.-H, Kim, K. Smith, K.-J. Lee, S. Santhanagopalan, A. Pesaran, “Multi-Domain Modeling of Lithium-Ion Batteries Encompassing Multi-Physics in Varied Length Scales,” J. Electrochem, Soc. 158(8), A955-A969 (2011).

18

XCEL R&D: Electrochemical, Atomistic, and Techno-Economic (BatPaC) (Argonne National Laboratory)

Dennis Dees, Hakim Iddir, Juan Garcia, and Shabbir Ahmed (Argonne National Laboratory)

Background Electrochemical modeling uses continuum-based transport equations combined with kinetic and thermodynamic expressions to allow the potential, concentration, and current distributions to be determined throughout the cell. The recent focus of the electrochemical modeling effort is to improve and better quantify diffusion and phase change in graphite active materials based on previous modeling efforts [1]. The previous model treats graphite active materials as multiple phases, also referred to as stages for graphite, where the well-known Avrami equation was introduced to describe the phase changes as a function of lithium concentration. Further, the model effectively correlated lithium diffusion and phase change during galvanostatic intermittent titration technique (GITT) studies. However, based on limited half-cell (i.e., graphite//lithium metal cell) data with an MCMB graphite electrode, the model tended to underestimate the performance of the graphite at high current rates. Work in this program has identified that the diffusion coefficient increases significantly with applied current rate for all phases. In order to examine an increasing diffusion coefficient with charging rate, Li-C phases and Li diffusion in graphite are modeled at the atomic scale to characterize the structure of the starting material and its changes during fast charging. Further, bulk defects and surface and edge effects on Li diffusion will be investigated for select conditions. The Battery Performance and Cost model (BatPaC) was developed for lithium-ion battery packs used in automotive transportation. The model designs the battery for a specified power, energy, and type of vehicle battery. The cost of the designed battery is then calculated by accounting for every step in the lithium-ion battery manufacturing process.

Results Electrochemical Modeling One of the strengths of electrochemical modeling is that it allows the distributions of concentrations, temperature, current, and potentials within the cell to be examined during operation. Actually, measuring these quantities within the cell sandwich is quite difficult. Recent work by Yao measuring the lithium and phase concentrations in a thick graphite electrode (i.e., 114 µm) during a 1C charge affords an excellent opportunity to compare and help calibrate the model [2]. In the study, an analysis of operando energy dispersive X-ray diffraction studies was used to establish the stage and lithium concentrations. The thick negative electrode was split into five regions (i.e., L0 to L4) determined by the gaussian-shaped X-ray beam profile paths through the cell. L0 is next to the separator, and L4 is next to the current collector. The study includes a constant charge, followed by a constant voltage hold at top-of-charge, and then a discharge. The initial comparison is focused on the constant charge region. Further, as the model assumes only three phase, stages I and II are the most easily directly compared. In the model, the volume fractions of the sum of the phases always add up to one. However, in the X-ray study the phase concentrations do not necessarily sum to one. Also, the positive electrode (i.e., 111 µm) was not calendared, giving it a higher porosity (i.e., 37% for the negative vs. 47% for the positive) to lessen any electrolyte transport effects. The beam profiles penetrate regions of lithium and stage concentration gradients across the electrode and within particles. It is assumed that values reported in the article correspond to volume averaged quantities.

19

Point values from model simulation, identified as reasonable estimates for volume averaged quantities, are reported here for comparison. Figure 1 shows that the original model parameters simulation, based on earlier development of thinner CAMP electrodes, results in a much better performing cell than the data from the actual cell indicates. Using NREL-estimated tortuosities only incrementally improves the fit of data. Positive electrode parameters of the uncalendared electrode were modified to improve the fit to the data, while the negative electrode parameters were not changed. The resulting very good fit is shown in Figure 1.

Figure 1. 1C charge cell voltage data and indicated simulations from NMC532 // A12 Graphite cell.

Model comparisons to the experimental lithium concentration and phase data for each layer during the 1C charge are shown in Figures 2-4. The overall fit of the lithium concentration (Figure 2) and Stage I (Figure 3) data is very good, while the Stage II data (Figure 4) are clearly off. The timing of the initiation of conversion to Stage II is excellent, and the position of the maxima is also good. It is unclear the reason for the variation of the absolute quantity of Stage II. The fact that the sum of the phase data does not sum to one complicates the comparison. Also, it is not clear how the amount of stage II can be significantly off and still get good agreement with the lithium concentration. However, it is clear that the overall good agreement with no fitted negative electrode parameters is a positive. Examining the differences more closely would require an extensive effort teaming with the experimentalist to study the approaches and assumptions connected with the model and the data analysis.

20

Figure 2. Graphite lithium concentration data and simulation for a 1C charge of a NMC532 / A12 graphite cell.

Figure 3. Graphite Stage I data and simulation for a 1C charge of a NMC532 // A12 graphite cell.

21

Figure 4. Graphite stage II data and simulation for a 1C charge of a NMC532 // A12 graphite cell.

Atomistic Level Modeling We use atomistic level analysis of Li-C phases and Li diffusion in graphite to understand the effect of fast charging conditions on the thermodynamics and kinetics behavior of the battery. Given the disparity between the electronic and ionic conductivities, an extra electronic charge in the graphite during the initial charging process might be present, particularly near the surface (graphite/electrolyte). The presence of an uncompensated charge (during a very short time scale) could play a role in the diffusion and staging processes. First, we studied the effect of extra charge on the interlayer spacing in graphite for several LiCx phases. We assume that under fast charging conditions and at the very beginning of the charge cycle there are more electrons than Li+ ions. In order to simulate such conditions, we built a supercell able to represent all the range of compositions between LiC36 stage IV and LiC6 stage I. Figure 5a shows a schematic representation of the supercell. Figure 5a shows the change in interplanar distances between layers with the number of extra electrons in the system for LiC36 (the model includes 72 C and 2 Li atoms). The red line represents pure graphite. As expected, the interplanar distances increase when extra electrons are present in the system due to electrostatic repulsion. In Figure 5b, d1 and d3 overlap because the cell is symmetric, and those interplanar spaces are next to the lithiated channel. For a neutral system (at charge = 0), the interplanar distance in the lithiated channel is larger than the rest of the other channels due to the presence of Li. The effect of extra negative charge on the d-spacing for LiC36 case is shown in Figure 5b. When the system is charged with extra electrons (up to three extra electrons: 3e- per 72C and 2Li), the interplanar distances of the Li-depleted channels d1, d3, and d4 increase faster than d2. In this case the electrostatic repulsion is compensated in channel d2 by the Li ions, while no charge compensation is possible for Li-depleted channels. However, the interplanar distance d2 starts to decrease when more than three extra

22

electrons are added. The interplanar distance of the lithiated channel becomes even smaller than in the pure graphite case. This could be an indication of two possible regimes for the extra charge distribution. At least for the system size considered in this model, too many extra electrons lead to an excessive increase in the interplanar distances, indicating a possible exfoliation of the material. When more lithium is intercalated, LiC18 and LiC12 (in stage II) are formed. Channels d2 and d4 are now equally lithiated and overlap in Figure 5c and 5d. The behavior is similar to LiC6 stage I. For LiC6 (stage I, see Figure 5d) all the interplanar distances overlap, reflecting the symmetry of the slab. This phase shows an increase in all interplanar distances in a similar way to the pure graphite. Our results show evidence of a possible change in the staging sequence during fast charging, resulting from a different interplanar distance pattern. To gain more insight on the effect of the extra charge on the d-spacing behavior, we calculated the charge density differences between the electron density of LiC36 and its individual components (graphite and Li). Figure 6 shows the electron density difference plots for different number of extra electrons in the simulation box. The blue iso-surfaces represent regions where the electron charge density difference is positive (regions that gained electron density). The yellow iso-surfaces represent regions of negative electron density difference (regions that lost electron density charge). Figure 6a represents the neutral supercell, and Figure 6b focuses around the Li ion. These figures show the electrons being localized around the Li ions and partially shared with the neighboring carbon atoms (C pz orbitals), indicating a partially covalent character. Electrons were withdrawn from the sp2 hybridized orbital within the graphene layer, as can be observed in the yellow iso-surfaces in Figure 6. The charge density redistribution is such that the extra charge is pointing away from the graphene planes, increasing the electrostatic repulsion between the layers. Increasing the extra charge beyond (3e-) triggers a new charge redistribution. For example, the charge density difference for six extra electrons (6e-) is shown in Figure 6e and 6f. The Li ions show a purely ionic character that will strongly attract the negatively charged carbon layers adjacent to them, while the empty layers will experience an increased repulsion given the high density of electrons in the carbon pz orbitals.

23

Figure 5. (a) Schematic representation of the supercell for the simulation of LiCx. Interplanar distance as a function of the extra electrons in the system for (b) LiC36, (c) LiC18, (d) LiC12, and (e) LiC6.

d1

d4 d3

d2

LiC36 d vs charge

LiC18 d vs charge LiC12 d vs charge LiC6 d vs charge

(a) (b)

(c) (d) (e)

24

Figure 6. Electron density differences with different number of electrons: zero extra electrons (a) super cell, (b) around Li-ions; three extra electrons (c) super cell, (d) around Li ions; and six extra electrons (e) super cell, (f) around Li ions.

In order to better understand the kinetics during the fast charging process from an atomistic point of view, the Nudged Elastic Band (NEB) method was used to calculate the energy barrier for Li diffusion at very dilute conditions with the presence of extra electrons in the system. Figure 7a shows the change in energy as a function of a reaction coordinate as calculated by the NEB method. Figure 7b shows the change in the energy barrier for Li diffusion with the number of extra electrons. For up to three extra electrons, the energy barrier decreases when the number of extra electrons increases, mirroring the increase of the interplanar distance in the lithiated channel reported above. Simulations with more extra electrons are in progress. The effect of the concentrated diffusion of Li atoms for specific configurations will be investigated next. Bulk defects and surface and edge effects on Li diffusion will be investigated for select conditions.

Figure 7. (a) Change in energy along the Li diffusion reaction coordinate and (b) energy barrier for lithium diffusion as a

function of extra electrons in the system.

(a) (e) (c)

(b) (d) (f)

(a) (b)

25

Performance and Cost Modeling BatPaC is a publicly available spreadsheet tool developed at Argonne for the design of automotive lithium ion batteries and the estimate of their cost when manufactured in large volume [3]. It was used in 2017 to estimate the cost of batteries capable of fast charging, defined as being able to recharge 80% of the battery’s capacity (i.e., from 15% to 95% SOC) in a given time [4]. Based on available data, it was estimated that the battery design was constrained by a maximum allowable current density of 4 mA/cm2 to prevent lithium deposition in the graphite anode [5]. A combination of electrolyte transport limitation and interfacial impedance causes lithium plating on the separator side of the negative electrode. At high loadings, electrolyte transport limitations push the electrode current distribution towards the separator and causes lithium plating. At medium and low loadings, the interfacial impedance has a greater impact on lithium plating. In this effort, electrochemical modeling is utilized to examine the appropriate BatPaC fast charge design approach. In the model, initial parameters derived from A12 graphite//NMC532 CAMP electrodes were optimized for the simulations. A high rate diffusion coefficient for lithium through the graphite was used. The initial negative interfacial area specific impedance (ASI) of 4 Ω cm2 was kept constant, while the initial positive interfacial ASI of 15 Ω cm2 was reduced to 5 to be more consistent with optimized industrial electrodes [6]. No changes other than electrode loading were made. While having a reference electrode in an actual cell is a significant challenge and always involves some compromises, including a perfect reference electrode in the model is simple. The simulation negative reference voltage presented here is the graphite electrode potential versus a perfect lithium reference electrode at the center of a single separator. As graphite becomes more lithiated, the open circuit voltage (OCV) approaches lithium potential through a series of single and two-phase regions. The BatPaC charging protocol (see below) takes advantage of this by using higher currents early in the charging process.

1. 15-75% SOC at a current density needed to meet the charging time specification for 15-95% SOC, but not to exceed the maximum allowable.

2. 75-85% SOC at a current density that is 75% of that in step 1. 3. 85-95% SOC at a current density that is 50% of that in step 1.

Figure 8 presents a comparison of 15-minute charging protocols for a 15-95% SOC swing of the baseline cell (56-µm negative electrode, cycling at 2.15 mAh/cm2, and 4.2 V) used in this study. Taking the minimum negative reference electrode voltage as an indicator of the effectiveness of the charging protocol, clearly the BatPaC protocol has advantages over constant current charging. Also shown in Figure 8 is a modified charging protocol (see below) that more closely follows the staged charging process of graphite.

1. 15-50% SOC at a current density needed to meet the charging time specification for 15-95% SOC. 2. 50-75% SOC at a current density that is 75% of that in step 1. 3. 75-95% SOC at a current density that is 50% of that in step 1.

In this example, the modified BatPaC charging protocol has advantages over the other protocols. In fact for this case the negative reference voltage is always positive. For this work, the modified BatPaC charging protocol is used for comparison of loading and charging time.

26

Figure 8. Simulation comparison of 15-minute charging protocols for 15-95% SOC swing on baseline cell.

The minimum negative reference and OCV during the 15- and 10-minute charging protocol over a range of loadings are given in Figure 9. Also shown in Figure 9 is the maximum charging current. It is clear that as long as the negative voltage is above that of a perfect lithium reference electrode, lithium will not plate; however, the voltage at which lithium will start to plate is not well defined and dependent on many factors. Here, the graphite OCV at the separator is an easily calculated and useful quantity in that if the OCV goes to zero there is no place for the lithium to easily intercalate. The minimum negative reference for 15-minute charge goes negative at about 60 microns and 9 mA/cm2, but the 10-minute charge is always negative. The minimum negative OCV for a 15-minute charge goes negative at about 100 microns and 16 mA/cm2 and 10-minute charge goes negative at 73 microns and 18 mA/cm2. While it appears that there is a clear path forward for thick conventional electrodes to be charged in 15 minutes, charging them in 10 minutes remains a challenge. Also, using a single current density limitation for a range of characteristics would not be optimum.

Figure 9. Minimum voltage and maximum current density as function of negative electrode thickness for (left) 15 minute

15-95% SOC swing and (right) 10 minute 15-95% SOC swing.

A better option may be to set a limiting potential difference and calculate a limiting current based on the estimated negative electrode impedance. Figure 9 shows that decreasing the charge time shifts the minimum negative reference and minimum negative OCV curves towards lithium plating. However, further study is needed to establish an appropriate limiting potential difference (i.e., OCV vs. reference). BatPaC has a separate

27

electrolyte transport limitation, so charge limitation can focus on the ASI for medium to thin electrodes. Finally, as the electrode active area goes to zero, the negative interfacial ASI dramatically increases with decreasing electrode thickness, which impacts the maximum charging rate.

Conclusions Our comparison of the electrochemical model to experimentally measured lithium and phase concentration distributions within a graphite electrode during charging generally shows good agreement, considering no negative electrode parameters were adjusted. An atomistic level study examining the impact of uncompensated electrons on lithium diffusion and staging processes indicated a significant increase of the interlayer spacing in graphite for several LiCx phases. Electrochemical modeling is being utilized to examine appropriate fast charge limitations used in the BatPaC model.

Milestones and Deliverables Determine model parameters: parameter set for Superior Graphite based negative electrodes will be developed. – Ongoing Examine rate-dependent diffusion: possible mechanisms for apparent increase of lithium diffusion coefficient in graphite at higher rates will be examined.

References 1. K.G. Gallagher, D.W. Dees, A.N. Jansen, D.P. Abraham, and S.-H. Kang, “A volume averaged approach

to the numerical modeling of phase-transition intercalation electrodes presented for LixC6,” Journal of Electrochemical Society, 159 (12) A2029-A2037 (2012).

2. K. Yao, J. Okaninski, K. Kalaga, I. Shkrob, and D. Abraham, “Quantifying lithium concentration gradients in the graphite electrode of Li-ion cells using operando energy dispersive X-ray diffraction,” Energy and Environmental Science (2019).

3. P. Nelson, K. Gallagher, I. Bloom, D. Dees and S. Ahmed, "BatPaC: A lithium-ion battery performance and cost model for electric-drive vehicles," 2018. [Online]. Available: http://www.cse.anl.gov/batpac/.

4. S. Ahmed, I. Bloom, A. Jansen, T. Tanim, E. Dufek, A. Pesaran, A. Burnham, R. Carlson, F. Dias, K. Hardy, M. Keyser, C. Kreuzer, A. Markel, A. Meintz, C. Michelbacher, M. Mohanpurkar, P. Nelson, D. Robertson, D. Scoffield, M. Shirk, T. Stephens, R. Vijaygopal and J. Zhang, "Enabling fast charging - A battery technology gap assessment," Journal of Power Sources, 367, 250-262 (2017).

5. K. Gallagher, S. Trask, C. Bauer, T. Woehrle, S. Lux, M. Tschech, P. Lamp, B. Polzin, S. Ha, B. Long, Q. Wu, W. Lu, D. Dees and A. Jansen, "Optimizing areal capacities through understanding the limitations of lithium-ion electrodes," Journal of The Electrochemical Society, 163(2), A138-A149 (2016).

6. Ricardo, "2013 MY Ford C-MAX Energi Battery Cell Teardown Analysis" (2016).

28

XCEL R&D: Performance Characterization and Post-Test Analysis (Argonne National Laboratory)

Contributors (Argonne National Laboratory): David Robertson, LeRoy Flores, Zhenzhen Yang, Alison Dunlop, Stephen Trask, Bryant Polzin, Andrew Jansen and Ira Bloom

Background One of the objectives of these projects is to determine how small pouch cells, which contain selected graphite anodes and oxide cathodes, respond to extreme fast charging. One of the key points here is to detect lithium plating as soon as it occurs using electrochemical methods. Another aspect is to provide materials characterization data, such as X-ray diffraction, Raman spectra, and diffusion coefficient measurements, which can be used to understand the observed performance response and for modeling. Another objective is to determine the physical and chemical changes that extreme fast charging caused. These changes will be characterized by post-test analysis of the fast-charged cells. A suite of materials and surface characterization techniques will be used.

Results One of the objectives of this work is to determine when lithium plating occurs and the factors that affect it. For this work, we used coins cells containing Round 2 electrode materials in full cells. After formation and one, 6-C charge (C/2 discharge) at 30°C, lithium was clearly seen on the surface of the 1506T anode (see Figure 1). The experiment was repeated with a 1-C charge. As expected, no lithium was seen on the surface of the 1506T anode (Figure 2). The amount of lithium in the anode was quantified after twenty 1-C and 6-C charge cycles (at 30°C). A photograph of the 1506T anode after twenty 6-C charges is shown in Figure 3.

Figure 1. Optical photograph of the anode after one 6-C charge

in the glove box, showing a “ring” of lithium metal.

Figure 2. Optical photograph of the anode after one 1-C charge.

Figure 3. Optical photograph of the anode after

twenty 6-C charges.

Here, anodes from 1-C- and 6-C-charged cells were washed two times in excess dimethylcarbonate, swirling for 1 minute each. After cleaning, the anodes were immersed in 1 mL HPLC-grade water. Elemental analysis of the water showed that 185.2 mg and 474.2 mg of lithium were present in the 1- and 6-C anodes, respectively, after subtracting lithium that was present as LiF.

29

Microstructural characterization of the 1- and 20-fast charged anodes showed some differences in the overall morphology of the lithium deposit. Figure 4 shows scanning electron micrographs (SEM) of the anode after 1 fast charge cycle.

Figure 4. SEM images of the anode surface after one 6-C cycle. The images to the right represent the magnified regions indicated on the image on the left. The red line on the bottom-right image represents the interface between regions 3

and 4.

Four distinct areas were seen, as indicated in Figure 4. In region 1 (center of the disk), dendrites were seen. They appear to be in layers and are large. Region 2 has some gouged-out areas and some dendrites on the surface of the graphite. Region 3 (periphery of the deposit) shows compacted dendrites, and region 4 looks like unreacted graphite. Figure 5 shows the SEM results from the anode after twenty 6-C cycles. As before, four regions were seen. Region 1 appears to show the unreacted electrode surface with some dendrites. Region 2 shows compacted dendrites, and regions 3 and 4 appear to be the unreacted electrode surface. It is highly likely that region 1 does not show the true electrode surface. Presumably, the real surface was covered with lithium dendrites that adhered to the separator and were subsequently removed during cell disassembly. Additionally, at the interface between regions 1 and 2, region 2 appeared to be on top of region 1.

159

bottom1

43

2

3

top

1234

30

Figure 5. SEM images of the anode surface after twenty 6-C cycles. The images to the right represent the magnified regions

indicated on the image on the left.

Conclusions Lithium plating was observed after one 6-C cycle. The amount of surface lithium present after twenty 6-C cycles was about 2.6 times that present after twenty 1-C charges. The morphology of the lithium deposit evolves with cycling. Lithium dendrites were seen and tended to become denser and more compact with cycling.

Future Work We will continue the work to characterize the factors that affect lithium deposition, time and temperature. In a separate line of work, we will explore how the charge rate affects the diffusion coefficient. We gratefully acknowledge support from the U. S. Department of Energy (DOE), Office of Energy Efficiency and Renewable Energy, Vehicle Technologies Office. Argonne National Laboratory is operated for DOE Office of Science by UChicago Argonne, LLC, under contract number DE-AC02-06CH11357. The U.S. government retains for itself, and others acting on its behalf, a paid-up nonexclusive, irrevocable worldwide license in said article to reproduce, prepare derivative works, distribute copies to the public, and perform publicly and display publicly, by or on behalf of the government.

11

3 and 4

2

234

31

XCEL R&D: Battery Testing Activities (Idaho National Laboratory)

Eric Dufek (INL), Tanvir Tanim (INL), Kevin Gering (INL), and Daniel Steingart (Princeton)

Background Extreme fast charging (XFC) of Li-ion batteries can create life and safety issues. Among the issues are shortened battery life due to enhanced loss of lithium inventory and electrolyte degradation and enhanced safety concerns due to potential short creation by Li dendrites. The detection and monitoring of Li plating onset and evolution over aging is a significant challenge. In operando detection schemes to understand the dynamics of Li plating and the role that aging has on Li plating are vital to enable fast charging of specific energy cells. Discreetly identifying Li plating is difficult without performing a destructive post-test analysis. The objective of the proposed work at INL and Princeton is to overcome these limitations using electrochemical analysis and non-destructive ultrasonic acoustic methods that can be directly applied in operando to understand onset and growth of Li plating during XFC. The team is closely coordinating with other efforts at the National Renewable Energy Lab (NREL) and Argonne National Lab (ANL) to understand the key limitations that enhance the probability of Li plating. A key to the efforts is understanding the interplay between materials, electrode structure and use conditions. The ability to understand the interplay will be distinctly aided in this project using the joint electrochemical and ultrasonic tools which ultimately will aid in the scientific understanding required to facilitate the XFC of batteries for electric vehicles. Electrochemical methods which can be used to identify Li plating include the use of differential capacity (dQ/dV) and quantitative analysis of the cells charge and discharge profiles. These tools give pertinent information associated with both kinetic and thermodynamic processes which occur in batteries, and as such provide direct ability to better understand how variation in materials and electrodes have an impact across the life of a battery. There are limitations on the type of techniques which can be used to compliment electrochemistry in operando using standard cell formats. One method which is showing promise is the use of ultrasonic measurements. Ultrasonic measurements rely on acoustic waves propagating through a structure, such as an electrode, which are modulated by its properties and encode structure/property relationship data. These relationships are directly tied to the material and mechanical changes which occur in a cell during cycling and can be used to characterize change in a non-destructive, real time manner. Coupling ultrasonic and electrochemical measurements will enable a more complete evaluation of the impacts of XFC to be understood. In particular both methods are expected to produce distinct and complimentary signals which signal the deposition of Li on the negative electrode of a battery during aggressive charging conditions. Tracking changes with the coupled ultrasonic and electrochemical methods over life and as batteries age will provide pertinent information related to the distinct conditions which drive Li plating during XFC.

Results At the end of FY18 Round 1 cells from ANL were near completion of evaluation at INL. The first quarter of FY19 was focused on three distinct efforts: 1) characterization and quantitation of the failure modes for the Round 1 cells 2) Evaluation of Round 2 cells and 3) Initial activities to identify electrolytes with enhanced transport properties. As shown in previous reports the cells from Round 1 which were charged under different protocols had a wide variation in performance and fade. Identification of the fade included multiple analysis methods including understanding the evolution of end-of-charge and end-of-discharge voltages as well as differential capacity (dQ/dV) with fade emulation using ‘Alawa’ software developed by Dubarry et. al.[1] From the combined analysis it was found that loss of lithium inventory and loss of positive electrode material (PE) were the dominate fade modes. The loss of PE material was distinct with loss from the PE ranging from

32

15-38% (Figure 1). Confirmation of fade was accomplished after testing by cycling extracted PE from aged cells in freshly made coin-cells. The percent fade of the PE mirrored the values arrived at through the ‘Alawa’ analysis. The extent of fade was greater than the overall cell fade due the cell design which included initial voltage offset of 7-8% and the initial N:P ratio which was 1.17 at a C/10 rate at the beginning of life. A key component of the PE fade was associated with the emergence of cracks in the secondary cathode particles observed using SEM and under lithiation of the cathode at the end of discharge as indicated using XRD.

Figure 1: Cell balance evolution for a poor performing cell (A) and a high performing cell (B). Note that the offset voltage in

the cell hid the initial fade from the PE. (C) dQ/dV data for PE material extracted from two cells at the end of cycling.

Figure 2: Capacity fade for Round 2 cells over 450 cycles at high rates. At 6C distinct increase in fade and variability is

observed.

For the Round 2 cells, evaluation was completed during Q1 of FY19. As with Round 1 cells, a suite of different charge protocols were used. As shown in Figure 2 there was a distinct difference between cells cycled at 4 and 6C. Post-test disassembly of the indicated a significant quantity of Li deposited on the surface of the negative electrode for one of the worst performing 6C CCCV cells. Some non-uniform Li plating was seen in the cells cycled at 4C CCCV. Currently similar analysis to quantify fade between the cells is underway. In the third focal area several electrolytes with transport properties which exceeded that of the Gen2 electrolyte were identified using the Advanced Electrolyte Model (AEM). Currently experimental evaluation of the different blends to better understand their ability to improve performance are underway.

33

Conclusions Analysis of Round 1 cells indicated that two dominate modes of failure occurred during fast charging. Both loss of lithium inventory and loss of PE occurred and led to fade for all cells. The extent of fade varied significantly between cells and charging protocols. Similar difference in performance has also been seen for Round 2 cells, though some protocols such as the 4C CCCV protocol have shown less cell-to-cell variability. Refined fade for the Round 2 cells and experimental characterization of high transport electrolytes are also underway.

Milestones and Deliverables Q1 – Quantitate single layer pouch cell failure modes – Complete, manuscript in preparation Q2 – Identification of enhanced transport electrolytes – In progress Q3 – Characterization of electrochemical Li reversibility – In progress Q4- Correlate electrochemical and acoustic signals – In progress

References [1] M. Dubarry, C. Truchot, B.Y. Liaw, Synthesize battery degradation modes via a diagnostic and

prognostic model, J. Power Sources. 219 (2012) 204–216. doi:10.1016/j.jpowsour.2012.07.016.

34

XCEL R&D: SLAC National Accelerator Laboratory

Hans-Georg Steinrück, Chuntian Cao, Michael Toney (SLAC)

Approach Our approach is to use synchrotron X-ray methods to guide cell design and charging protocols of extreme fast-charging Lithium ion battery (LiB) cells, in which a major concern is parasitically plated Li metal. Accordingly, our objective is to directly measure the presence or absence of Li plating in such cell. For this purpose, we combine spatially resolved X-ray diffraction (XRD) and X-ray micro-computed tomography (micro-CT) techniques to characterize the lithium plating behavior in single-layer pouch cells. These experiments will be performed on pristine cell and cells under different numbers of fast charging cycles, and eventually during operation. Spatially resolved XRD (Figure 1 a) is used to characterize the heterogeneity of Li plating in order to understand the battery failure mechanism. XRD experiments with area detector allows for fast data acquisition. By measuring the full XRD pattern, we can quantify the spatial distribution plated Li, and obtain the cathode and the anode degradation information at the same time. In spatially resolved XRD measurements, the X-ray beam is rastered across the surface of the single-layer pouch cell, and the spatial resolution is determined by the X-ray beamsize. We obtain a 2D map of the entire pouch cell by using sub-millimeter resolution SR-XRD, and map the grains of plated Li using μm-sized X-ray beam. X-ray micro-CT (Figure 1b) is an ideal non-destructive imaging technique capable of collecting 3D morphological and elemental information with micrometer resolution with a field of view of millimeters. Specifically, this technique is used to monitor the morphology changes including Li plating and structural degradation inside the battery, thus obtaining a clear and intuitive understanding of Li plating and correlating the macroscopic battery capacity fading behavior to microstructure changes.

Figure 1. (a) Illustrations of the configuration of single layer pouch cell, the setup of the XRD experiment with area detector,

and the 2D rastering map. (b) Recent results from micro-CT.

Results We first conducted a preliminary 2D XRD experiment with a spatial resolution of 0.5 mm at SSRL on a single layer pouch cell cycled at 2C for 25 cycles. The results are shown in Figure 2a. By mapping a ~10 mm × 10 mm region on the pouch cell, we have found that the Li plating is inhomogeneous inside the cell, and the amount of plated Li at each position can be quantitatively compared by comparing the peak intensity. Subsequently, we performed X-ray micro-diffraction at APS on single layer pouch cells cycled at 6C charging rate for 100 cycles. We succeeded in obtaining sub-μm resolution by using 0.3 μm × 0.3 μm beamsize, as shown in Figure 2b. A “coarse” scan across a 20 μm × 20 μm region with 1 μm spacing between adjacent positions shows a gradient in Li peak intensity, and a “fine” scan over a 7.8 μm × 7.8 μm area with 0.3 μm

35

spacing shows four Li “clusters” and demonstrates heterogeneity on the sub-μm scale. By mapping the Li plating behavior with different resolution, we were able to resolve the heterogeneity of Li plating at different length-scale. In addition, we conducted micro-CT experiment on NMC cathode on Al foil, as shown in Figure 2c. The image in Figure 2c is a projection of the reconstructed 3D structure of the electrode, from which the NMC coating (~70 μm) and the Al current collector (~20 μm) can be clearly seen, and some microstructures of the NMC coating can also be resolved. We are currently performing further analysis of the micro-CT data, and are measuring the porosity/tortuosity of graphite electrodes. In the near future, we will to perform 2D XRD mapping on full pouch cells with sub-mm resolution, and focus on specific Li plating regions with micro-diffraction (beamtimes are already scheduled). In addition, we will identify hotspots of Li plating in pouch cells; these will then be measured with cmicro-CT, thus connecting the results from micro-CT and XRD techniques.

Figure 2. (a) Results from preliminary 2D XRD experiment with sub-millimeter resolution. (b) Coarse and fine scan results

from X-ray micro-diffraction experiment with sub-micrometer resolution. (c) Illustration of micro-CT experiment on NMC cathode (left) and one projection of the 3D-reconstructed result (right).

36

XCEL R&D: Lawrence Berkeley National Laboratory

Nitash Balsara, Vince Battaglia, March Doeff, Bryan McCloskey, Guoying Chen, Robert Kostecki, Wei Tong, Ravi Prasher (LBNL)

Background The goal of the LBNL team is to detect the initial nucleation event using data available in typical battery management systems. Identifying the initial onset of Li plating during fast charging is a multi-faceted, challenging “needle-in-a-haystack” problem. First, there may be a barely perceptible chemical signature of this event. For example, the presence of lithium may lead to the sudden formation of a small amount of gases due to irreversible reactions between the plated lithium and the electrolyte-soaked solid-electrolyte interphase (SEI). Second, identifying the particular particle where the over-lithiation and Li plating occurs can be challenging at the immediate onset of Li plating. Clearly, the particular particle that is over-lithiated first will depend on the state-of-charge of the electrode, as it will be governed by the amount of lithium intercalated at that time and the lithium concentration in the surrounding liquid electrolyte. The third and final challenge is identifying electrical signatures of the over-lithiation event so that our work can be used in practical battery management systems. We note here that the signatures may be different at different states-of-charge, and that the electrical signature may be inherently non-linear, and thus not clearly evident in standard electrochemical measurements. Our objective is to address all three challenges. Our objective is to deliver a strategy to detect lithium plating during fast charging in 1 Ah cells. We will use different approaches to study electrodes and cells built within this program. We will begin our work with half cells for the first year. In the first year, plating will be studied by a wide variety of approaches: optical microscopy, X-ray scattering, impedance analysis, X-ray microtomography, optical and X-ray absorption spectroscopy, mass spectroscopy, and thermal analysis.

Results We have just begun our investigations. Shown in Figure 1 are results of X-ray tomography of graphite anodes fabricated by XFC team members at Argonne. Three anodes were separated by either a traditional separator or a polystyrene film. The tomogram was measured at the Advanced Light Source at LBNL. The main conclusion from this preliminary scan is that the tomograms are dominated by the Cu current collector. We are evaluating other imaging X-ray tomography modes that will enable filtering the current collector contribution to the imaging.

Figure 1. X-ray tomography of graphite anodes.