Calendar life of lithium ion batteries containing silicon and the role of solid electrolyte interphase mechanics

Silicon anodes can theoretically enable lithium-ion batteries with 10x the capacity of current graphite anodes for electric vehicle applications. Silicon anodes traditionally have suffered from over 300% expansion during lithiation, leading to mechanical damage. However, volume expansion problems have largely been overcome through the use of nanomaterials. Conversely, the mechanical and chemical failure of a passivation layer, known as the solid electrolyte interphase (SEI), under cycling and calendar conditions remains poorly understood. Understanding and mitigating these passivation problems is the key to solving calendar life aging problems recently identified for silicon anodes. This work first investigated potentiostatic holds as a qualitative accelerated stage-gate for evaluating calendar aging and identified the required experimental process to do such an experiment reliably. To gain more insight into the mechanical failure of the SEI during cycle and calendar aging, the use of moir´e microscopy was evaluated for making in situ strain measurements of the SEI, but low resolution due to electrode bowing from gas generation limited this technique. Instead, the relationship between chemical and mechanical degradation is explored through scanning electrochemical microscopy imaging on model silicon thin films as a function of potential and time at open circuit potential. Silicon is found to be better passivated in the lithiated state rather than the delithiated state, and within resolution limits, the change in passivation is global rather than the formation of discrete cracks. Passivation is found to decrease with rest time. The role of SEI mechanics in calendar life measurements is further elucidated through specially designed protocols for full cells to understand decay from cycling reference performance tests versus decay from calendar aging. For all rest durations explored, graphite aging was found to be driven by the time since assembly, whereas silicon degradation was dependent on cycling and rest duration where longer rests with the same amount of cycling led to time since assembly dominating degradation.

CALENDAR LIFE OF LITHIUM ION BATTERIES CONTAINING SILICON AND THE ROLE OF SOLID ELECTROLYTE INTERPHASE MECHANICS by Josefine D. McBrayer A dissertation submitted to the faculty of The University of Utah in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Chemical Engineering The University of Utah August 2022 Copyright © Josefine D. McBrayer 2022 All Rights Reserved The University of Utah Graduate School STATEMENT OF DISSERTATION APPROVAL The dissertation of Josefine D. McBrayer has been approved by the following supervisory committee members: Shelley D. Minteer , Chair(s) 06/20/2022 Date Approved Kyle R. Fenton , Member 06/21/2022 Date Approved Swomitra K. Mohanty , Member 06/21/2022 Date Approved Philip J. Smith , Member 06/20/2022 Date Approved Thomas A. Zangle , Member 06/21/2022 Date Approved by Eric G. Eddings , Chair/Dean of the Department/College/School of Chemical Engineering and by David B. Kieda , Dean of The Graduate School. ABSTRACT Silicon anodes can theoretically enable lithium-ion batteries with 10x the capacity of current graphite anodes for electric vehicle applications. Silicon anodes traditionally have suffered from over 300% expansion during lithiation, leading to mechanical damage. However, volume expansion problems have largely been overcome through the use of nanomaterials. Conversely, the mechanical and chemical failure of a passivation layer, known as the solid electrolyte interphase (SEI), under cycling and calendar conditions remains poorly understood. Understanding and mitigating these passivation problems is the key to solving calendar life aging problems recently identified for silicon anodes. This work first investigated potentiostatic holds as a qualitative accelerated stage-gate for evaluating calendar aging and identified the required experimental process to do such an experiment reliably. To gain more insight into the mechanical failure of the SEI during cycle and calendar aging, the use of moiré microscopy was evaluated for making in situ strain measurements of the SEI, but low resolution due to electrode bowing from gas generation limited this technique. Instead, the relationship between chemical and mechanical degradation is explored through scanning electrochemical microscopy imaging on model silicon thin films as a function of potential and time at open circuit potential. Silicon is found to be better passivated in the lithiated state rather than the delithiated state, and within resolution limits, the change in passivation is global rather than the formation of discrete cracks. Passivation is found to decrease with rest time. The role of SEI mechanics in calendar life measurements is further elucidated through specially designed protocols for full cells to understand decay from cycling reference performance tests versus decay from calendar aging. For all rest durations explored, graphite aging was found to be driven by the time since assembly, whereas silicon degradation was dependent on cycling and rest duration where longer rests with the same amount of cycling led to time since assembly dominating degradation. I dedicate this work to my soon-to-be husband, Jorge CONTENTS ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii NOTATION AND SYMBOLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix CHAPTERS 1. 2. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Lithium Ion Batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Silicon Anodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 The History of the SEI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Attempts to Stabilize the SEI on Silicon . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1.1 Electrolyte Additives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1.2 Surface Coatings and Functionalization . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Characterization of the SEI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Aging of Lithium Ion Batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Cycle and Calendar Aging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Degradation Mechanisms Relevant to Calendar Aging . . . . . . . . . . . . . . 1.5 Mechanical Characterization of Silicon and the SEI . . . . . . . . . . . . . . . . . . . . . . 1.5.1 Moiré Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.2 The Relationship Between Stress and Strain . . . . . . . . . . . . . . . . . . . . . . . 1.5.3 Mechanics of Thin Films and Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6 Electrochemical Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.1 Cyclic Voltammetry (CV) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.2 Scanning Electrochemical Microscopy (SECM) . . . . . . . . . . . . . . . . . . . . . 1.6.3 Electrochemical Impedance Spectroscopy (EIS) . . . . . . . . . . . . . . . . . . . . 1.7 Approach to Understanding the Impact of Mechanical and Chemical Instability of the Si Anode SEI on Calendar Life . . . . . . . . . . . . . . . . . . . . . . . . 3 3 6 7 8 10 11 12 12 13 19 23 23 24 27 27 28 31 31 EXPERIMENTAL METHODS AND TECHNIQUES . . . . . . . . . . . . . . . . . . . . . . . 56 2.1 Moiré Interferometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Sample Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.2 Instrument Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.3 In situ Cell Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Coin Cell Fabrication and Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Calendar Life Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Voltage Hold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 56 57 57 57 57 58 58 2.2.4 Coin Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.5 Single-Layer Pouch Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.6 Cylindrical Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.7 Variable OCV-RPT Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Scanning Electrochemical Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Sample Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Redox Probe and Electrolyte Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.4 In situ Cell Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.5 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Calorimetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. 58 59 59 60 60 60 61 61 63 65 66 CRITICAL EVALUATION OF POTENTIOSTATIC HOLDS AS ACCELERATED PREDICTORS OF CAPACITY FADE DURING CALENDAR AGING . . . . . . . . 79 3.1 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.1.1 Description of the Voltage Hold Technique for Studying Calendar Aging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.1.2 Influence of Current Relaxation and Reversible Capacity on the Measured Aging Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 3.1.3 How Hardware Sensitivity and the Prominence of Reduction Reactions Can Affect Voltage Holds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 3.1.4 Evaluating Whether the Anode Overhang Will Affect the Measured Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 3.1.5 Can Voltage Holds Provide Qualitative Information About Calendar Aging? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 3.1.6 A Quick Guide to Using Voltage Holds for Qualitative Screening . . . . . . 101 3.1.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 4. SCANNING ELECTROCHEMICAL MICROSCOPY TO QUANTIFY THE CHANGES IN SILICON THIN FILM PASSIVATION OVER TIME . . . . . . . . . . 121 4.1 Moiré Interferometry as a Possible Method for Tracking in situ SEI Strain . . . 121 4.2 SECM as a Direct Method to Track Changes in Silicon Passivation During Rest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 5. MECHANICAL IMPACTS FROM CYCLING ON CALENDAR AGING MEASUREMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 6. SUMMARY AND CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 APPENDIX: CHAPTER 3 VOLTAGE HOLD SUPPLEMENTAL . . . . . . . . . . . . . . . . . 158 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 vi LIST OF TABLES 1.1 Examples of characterization techniques utilized to study the solid electrolyte interphase on silicon as well as in other electrochemical systems. Each technique’s application is also listed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 1.2 Characterization techniques to measure stress. Desired capabilities include 1) capability to measure stress in amorphous material, 2) capability to distinguish between contributions of stress from the SEI and silicon electrode, 3) in situ operation, 4) to represent a normal battery as closely as possible, and 5) ability for in situ characterization under different conditions. . . . . . . . . . . . . . 36 1.3 Mechanical properties of SEI, Si, and Cu used in initial strain calculations [130] 43 2.1 Electrodes used in this work provided by Argonne National Laboratories. . . . . 67 3.1 Considerations for successfully implementing qualitative voltage hold comparisons to infer relative calendar life performance . . . . . . . . . . . . . . . . . . . 108 NOTATION AND SYMBOLS f1 f2 Fmoire E E′ σ ϵ ν ∆ϵ κcurv Qhold Qirrev Qrev Qloss Rg β i∞ n F D C∗ a itip,N d L κ r glass Frequency of the sample grating Frequency of the reference grating Frequency of the moiré fringe pattern Young’s modulus Biaxial modulus Stress Strain Poisson’s ratio Misfit strain Curvature Capacity passed during a voltage hold Irreversible portion of capacity passed during a voltage hold Reversible portion of capacity passed during a voltage hold Actual capacity lost during a voltage hold Ratio of the radius of the surrounding glass and disc microelectrode R g correction factor Steady state current Number of electrons Faraday constant Diffusion coefficient Bulk concentration Microelectrode radius Normalized microelectrode tip current Distance from SECM substrate surface Normalized distance from SECM substrate surface Dimensionless rate constant Radius of glass surrounding microelectrode cond itip,N Normalized probe tip current, approaching a conductive SECM substrate ins itip,N Normalized probe tip current, approaching an insulating SECM substrate ACKNOWLEDGEMENTS I would like to thank all of my family and friends who have supported me throughout my PhD, especially my parents and fiancé. I could not have done this without them. I also want to give a special thanks to Chris Apblett, Katie Harrison, Kyle Fenton, Darwin Serkland, and Shelley Minteer for being wonderful mentors and advisers. I have learned so much from all of them. In particular, I would like to thank Katie, who made my transition between advisers at Sandia extremely smooth - she truly inspires me because she is such a kind person in addition to being an amazing scientist and mother. I also want to thank Chris for seeing me through my MS and PhD and making sure I was in good hands before he retired, he has made a lasting and positive impact on my career and passion for electrochemistry. I appreciate all of the support I have gotten from the Power Sources Technology group at Sandia, the Minteer group at the University of Utah, and everyone in the Silicon Consortium Project. I appreciate all of the discussions on interesting data (and life) and know I have made life-long friends and colleagues during my PhD. I especially want to thank Max Schulze and Marco Fonseca Rodrigues; working with them has been an absolute pleasure and has allowed me to grow greatly as a researcher. There are so many people that have helped with this work, I want to thank Steve Wolfley for making my thin film samples, J.P. Bullivant for his ALD deposition of alumina, Eric Allcorn for discussions on calorimetry, Darwin Serkland for teaching me how to build microscopes, and Noah Schorr for teaching me SECM. Noah and Darwin were always there to answer my questions regardless of if that meant working through lunch or working late, so I can’t thank them enough. I want to thank my committee for their time and consideration and for meeting with me regularly before the pandemic started to discuss ideas and challenges. Lastly, I want to thank the Vehicle Technologies Office for funding this work. This research was supported by the U.S. Department of Energy’s Vehicle Technologies Office under the Silicon Consortium Project, directed by Brian Cunningham, and managed by Anthony Burrell. Sandia National Laboratories is a multimission Laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525. Lawrence Berkeley National Laboratory is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the US Department of Energy under contract no. DE-AC02-05CH11231. This work was conducted in part by the Alliance for Sustainable Energy, LLC, the manager and operator of the National Renewable Energy Laboratory for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-08GO28308. The submitted manuscript has been created by UChicago Argonne, LLC, Operator of Argonne National Laboratory (“Argonne”). Argonne, a U.S. Department of Energy Office of Science laboratory, is operated under Contract No. DE-AC02-06CH11357. This manuscript has been authored by UT-Battelle, LLC, under Contract DE-AC05-00OR22725 with the U.S. Department of Energy. Lawrence Berkeley National Laboratory is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the US Department of Energy under contract no. DE-AC02-05CH11231. The views expressed in the work do not necessarily represent the views of the DOE or the U.S. Government. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. Government purposes. x CHAPTER 1 INTRODUCTION This chapter is adapted with permission from two published manuscripts: Used with permission of IOP Publishing, Ltd, from Mechanical studies of the solid electrolyte interphase on anodes in lithium and lithium ion batteries, J. D. McBrayer, et al., vol 32, no. 50, copyright 2021. Used with permission of Springer Nature BV, from Calendar aging of silicon-containing batteries, J. D. McBrayer et al., vol. 6, no. 9, copyright 2022; permission conveyed through Copyright Clearance Center, Inc. Lithium ion batteries have become pervasive energy storage devices with diverse applications, from personal electronics to electric vehicles. To enhance performance and expand the use of lithium ion batteries in larger operations, including vehicles and grid storage, new materials for the cathode and anode are required [1–4]. Current lithium ion batteries contain a graphite anode and a lithium metal oxide cathode. For improvement of the anode, materials such as silicon and tin are being investigated for their greater volumetric and specific capacities. In particular, silicon has gained interest, because it has a high theoretical capacity of 3,579 – 4,200 mAh/g, [2, 5] over 10 times that of graphite. Despite its high capacity, silicon suffers from an unstable solid electrolyte interphase (SEI) and over 300% expansion of the silicon when it alloys with lithium during lithiation [6]. The SEI is a complex, heterogenous, passivating layer that forms on the anode and cathode during the initial cycles of a Li ion battery [7, 8]. The anode film forms due to the potential window of the anode being below the stability window of the electrolyte, resulting in reduction of the solvent and salt on the anode. A desirable SEI is ionically conductive and electronically insulating such that lithium ions can diffuse through, allowing battery operation, while the passage of electrons is inhibited, preventing the continuous reduction of electrolyte [9]. To act as an effective stabilizing layer, the SEI must be both 2 (electro)chemically and mechanically stable once formed. The formation of cracks in the SEI are hypothesized to allow direct contact of the electrode and the electrolyte resulting in electrolyte reduction [9–11]. Chemical side reactions that change the character of the SEI or continuous electrochemical reduction reactions that thicken the SEI can also lead to poor cycle performance due to the irreversible loss of the lithium inventory and the increasing impediment to lithium ion diffusion [3]. This can result in the isolation or electrochemical deactivation of Si particles [12] and increased cell resistance [9]. In commercial cells containing graphite anodes, a relatively stable SEI allows a long cycle life, on the order of 1000s of cycles [13]. Conversely, the SEI that forms on silicon provides poor passivation, which reduces cycle life [9]. The silicon anode SEI formation and instability are not well understood [9]. As bulk silicon lithiates and delithiates, a volume change of over 300% causes the silicon to crack [1, 14]. It is commonly assumed that this cracking causes the SEI to crack as well, creating new surfaces for electrolyte to be reduced [9–11]. However, cracking and other damage in the silicon electrode due to expansion, have been minimized through the implementation of nanomaterials [1, 2, 15–17]. Yet the instability of the SEI persists, suggesting there are multiple routes of degradation or that the improved mechanical stability of the nanosilicon materials does not translate to the SEI. For example, a breathing behavior has been observed in the silicon SEI, irrespective of silicon cracking, where the SEI thickness changes as a function of potential, [18, 19] implying that the decomposition products are not fully insoluble in the electrolyte and that undesirable electrochemical processes do not stop after the formation of the SEI. The use of nanomaterials has greatly improved the cycling capability of high-loading silicon anodes, but calendar life, or time dependent fade of a battery is still far from the goals set forth by the Department of Energy Vehicle Technologies Office [20]. This large technical gap is depicted in Figure 1.1 where industrial cells are beginning to reach cycling goals, but the same cells have poor calendar lives. To date research into improving the calendar life of silicon anodes has been limited to a few studies [21–23]. In order to improve the calendar life of silicon anodes for lithium ion batteries, an understanding of the complex mechanical-electrochemical degradation of the SEI is needed. This must include an integrated understanding of the mechanism of mechanical versus chemical 3 failure of the SEI as a function of both potential and time. 1.1 Lithium Ion Batteries Batteries are composed of an anode, cathode, separator, and electrolyte. During discharge, oxidation and reduction occur at the anode and cathode, respectively. Primary batteries are disposable after a single discharge. Secondary reversible batteries undergo multiple charge-discharge cycles and act as both galvanic (providing energy) and electrolytic (requiring energy) systems depending on the direction of current flow. For secondary batteries, the cathode and anode electrodes retain their names despite the switch in oxidation and reduction upon charging [5]. During discharge of a full cell lithium ion battery (delithiation of the anode), the battery operates galvanically and electrons flow from the anode to the cathode through the external circuit. During charge (lithiation of the anode), the battery is an electrolytic system and an external power supply is necessary to force electrons in the opposite direction [5]. The operation of a lithium ion battery is depicted in Figure 1.2. The first commercial lithium ion batteries were sold by Sony in 1991 and have since revolutionized the energy storage field [5]. Lithium ion batteries have structured electrodes that store lithium ions that intercalate in between the layered or tunneled structures. The lithium ions move back and forth between the anode and cathode. Despite good cycle life, current graphite anodes are reaching their theoretical limit in terms of energy density. In order to meet increasing portable energy demands, new materials, such as silicon, must be explored. 1.2 Silicon Anodes Silicon anodes are a promising candidate to either replace graphite or act as an additive in a silicon-graphite anode. Silicon is the anode in a full cell and can be paired with traditional lithium ion cathodes such as nickel-manganese-cobalt oxide (NMC). The electrolyte is typically a mixture of carbonates with lithium hexafluorophosphate (LiPF6 ) as the salt. Silicon is often studied in half cells to simplify the system. In a half cell, silicon acts as the cathode and a lithium metal anode acts as an infinite source of lithium. Half cells are useful, because they focus on one half of the overall reaction. In the silicon system, the half 4 reaction of interest is shown in Equation 1.1. The reduction of the electrolyte solvents and salt are also of interest because the products that form the SEI are generally assumed to be insoluble. Si + xLi + xe− ←→ Li x Si (1.1) Silicon as an anode has been studied extensively using transmission electron microscopy (TEM), [16,24,25] nuclear magnetic resonance (NMR) spectroscopy, [26–28] Raman, [29–34] x-ray diffraction (XRD), [29] scanning electron microscopy (SEM), [25] atomic force microscopy (AFM), [35–37] neutron reflectometry (NR), [18, 38] scanning transmission electron microscopy electron energy loss spectroscopy (STEM-EELS), multibeam optical stress sensor (MOSS), [39–41] and x-ray photoelectron spectroscopy (XPS) [35, 42] among other techniques. Key findings include the level of lithiation possible at room temperature, [1, 43] the transition from crystalline to amorphous silicon during lithiation, [14, 17, 24, 29] and the presence of cracking and pulverization of the silicon beyond a critical size limit of nanomaterial [16]. Theoretically, the full lithiation of silicon is Li22 Si5 ; however, at room temperature, only lithiation to the metastable phase of Li15 Si4 is possible [1, 43]. Pristine crystalline silicon becomes amorphous as it lithiates, until it crystallizes at Li15 Si4 [14, 17, 24, 29]. Lastly, cracking and pulverization can largely be avoided in nanomaterials sized below 100-200 nm thickness for thin films, 150 nm diameter for nanoparticles, and approximately 250 nm diameter for nanowires [1, 2, 15–17]. The mechanism of the lithiation of silicon is complicated and several phase transformations are involved [15, 27]. One commonly proposed mechanism of lithiation of crystalline silicon involves a two-phase region of lithiated and unlithiated silicon leading to a lithiation front observed by TEM [24]. Lithiation is initiated as silicon bonds break at the interface of the silicon and electrolyte, forming silicon-lithium clusters. As this occurs, the silicon becomes amorphous. The remaining silicon clusters are further separated as more lithium enters the electrode. Once the energy barrier of breaking all of the silicon clusters is overcome (at < 50 mV versus Li/Li+ ), the electrode recrystallizes [27]. In crystalline silicon, different planes have different interfacial mobilities. For example, the 110 surface lithiates, and therefore expands, the fastest. This leads to anisotropic expansion and lithiation in crystalline silicon and isotropic lithiation in amorphous silicon [4, 14]. 5 Differences in polycrystalline materials cause changes in local potential gradients and therefore differences in lithiation [44]. At room temperature, the fully lithiated phase is Li15 Si4 . At room temperature, only one plateau in the voltage versus time curve is present, describing predominately two, metastable phases [4]. During the first lithiation, three processes occur corresponding to a-Si to a-Li∼2.0 Si (∼300 mV versus Li/Li+ ), a-Li∼2.0 Si to a-Li∼3.5-3.75 Si (∼100 mV versus Li/Li+ ), and a-Li∼3.75 Si to c-Li3.75 Si (∼50 mV versus Li/Li+ ) [45]. On subsequent cycles, the peaks can shift, because the electrode will not return to its original crystalline form. An example of a cyclic voltammogram (CV) of an amorphous silicon thin film is given in Figure 1.3, where the middle lithiation peak has shifted from 100 mV to 200 mV. Cyclic voltammetry is an electrochemical technique where the voltage is scanned linearly between two voltages and the current response is recorded. It provides information on what potential electrochemical processes occur. For composite electrodes, containing silicon and graphite, the lithiation is further complicated by the superposition of the mechanism of lithium intercalation into graphite and the alloying of silicon and lithium. In either case, to reach the electrode, the lithium ions must first pass through the SEI at the interface between the electrode and the electrolyte. The silicon-electrolyte interface is dependent on the electrode form factor, crystallinity of the silicon, and the initial functionalization of the silicon surface [4]. Three commonly used forms of silicon include silicon nanoparticles, nanowires, and thin films. Maintaining the silicon below a critical size limit helps mitigate cracking and mechanical damage to the electrodes [1,2,15–17]. The use of nanomaterials increases the surface area which increases the reactivity with the electrolyte. Nanoparticles and nanowires are often combined with a binder and conductive additive to create a slurry, increasing the surface area available for lithiation. Silicon thin films are typically used as model electrodes to simplify the system. Silicon without any surface treatment will inherently have a native oxide. This oxide film is believed to be reduced during the first lithiation to form lithium silicates (Lix SiOy ) [46]. Silicon etched with HF to remove the native oxide is hydrogen-terminated and has been found to be more reactive with the electrolyte and form a thicker SEI with a different composition than that formed in the presence of a native oxide [47]. Scanning electrochemical microscopy (SECM) showed that electron transfer rates were low in the presence of the native oxide and higher when hydrogen-terminated. Electron transfer was 6 found to be highly heterogeneous in both cases [48]. The form factor and size of silicon nanomaterial used directly impact the expansion during lithiation which leads to cracking and pulverization of the silicon as well as electronic isolation of the silicon and delamination of the electrode from the current collector [49]. These mechanical degradation pathways cause associated electrochemical degradation including irreversible loss of Li, continuous electrolyte consumption, isolation or electrochemical deactivation of Si particles, [12] and increased cell resistance due to film formation [9]. Furthermore, chemical side reactions with contaminants in the electrolyte as well as the opposing electrode can also decrease the cycle life of the battery. For example, the reaction of LiPF6 and H2 O is believed to form HF which can etch the native SiO2 layer [50]. The combination of these degradation pathways is believed to result in poor cycle life when using silicon anodes. 1.3 The History of the SEI The general structure and concept of the SEI was first published in the 1970s. Dey observed the decomposition of propylene carbonate on graphite [51] and Peled proposed the model of the solid electrolyte interphase in 1979 [52]. Peled theorized that alkali and alkaline earth metals always have a thin, passivating layer, called the SEI. This layer forms immediately upon contact of the electrode with the electrolyte [53] and acts as a solid electrolyte, inhibiting the passage of electrons while allowing ions to conduct through the layer. Peled and coworkers built on this publication with several studies on the SEI on various materials [52, 54–59]. Dahn et al. studied the cycling of lithium/graphite and lithium/carbon coke cells in ethylene carbonate (EC) and propylene carbonate (PC) electrolytes. They observed the formation of the SEI only during the first lithiation of the carbon, concluding that the SEI prevented further reduction of the electrolyte and led to stable cycling. The SEI formed on graphite with PC was found to be a poor conductor of Li ions, indicating that small changes in the electrolyte can cause drastic changes in the SEI performance [60]. Many studies have demonstrated that composition was highly dependent on the electrolyte [7,61,62]. For example, PC solvent has been found to be more reactive with graphite and lithium anodes than ethylene carbonate, and cannot be used alone. With both solvents, 7 Li2 CO3 was observed as a common SEI component and was found to contribute to an effective, compact passivation layer with a lower impedance, [61, 62] protecting the electrolyte from further reduction while allowing for efficient lithium transport to the electrode surface. Polymeric species were observed and thought to be beneficial for added elasticity, particularly for an SEI formed on lithium metal [7]. It became generally accepted that the structure of the SEI involved two layers including a compact, inorganic inner layer and a more porous, organic outer layer although the composition and thickness of these layers differ depending on the electrolyte and electrode. For example, the SEI on silicon is generally thicker than the SEI on graphite [63]. Peled then published an updated model of the heterogeneity of the SEI and proposed the existence of microphases in the SEI as shown in Figure 1.4 [8]. For many cases, Peled et al. found that the SEI could be treated as if it only consisted of one sublayer and that all other resistances and capacitances were much less than that of the SEI. In this case, a simple RC circuit could still often be used to represent the SEI. Aurbach et al [7]. proposed a mechanism for the formation of the SEI by (1) charge going from the electrode into solution, (2) electrolyte getting reduced near the electrode, (3) products precipitating onto the electrode and then partially dissolving, and (4) re-precipitating the least soluble products. Steps 3 and 4 reach a steady state of dissolution and precipitation [7]. On graphite, although there is a microscopic steady state of dissolution and precipitation of SEI components, the overall passivation layer is effective and stable. This allows for the long cycle life of lithium ion batteries [13]. However, the SEI formation on other anode materials, such as lithium and silicon, is not stable and can change dramatically as a function of time and potential [7, 18]. Thus, there is motivation to better characterize SEI properties and determine effective approaches to stabilizing the SEI on a variety of electrode surfaces. 1.3.1 Attempts to Stabilize the SEI on Silicon Many different approaches have been investigated to try and stabilize the SEI on silicon. Protection and control of the silicon surface are believed to be paramount to stabilization. Artificial SEIs are one possible solution. An artificial SEI can be formed in situ (from an additive or cosolvent) or ex situ (a coating added prior to battery operation) [1]. The 8 silicon surface can be functionalized with specific chemical groups [17, 64] or coated with materials such as graphite or silicon dioxide. Ideally, the SEI’s structure and composition could be controlled and tuned to improve its stability. 1.3.1.1 Electrolyte Additives Electrolyte additives have dramatically improved the performance of lithium ion batteries. For example, one publication reported the capacity retention with fluoroethylene carbonate (FEC) as 87.8 % after 80 cycles as compared to 67.9% without FEC [65]. FEC and vinylene carbonate (VC) are the two most common, and best performing, examples of additives for silicon lithium ion batteries [66]. Other additives studied include dioxolane, succinic anhydride, lithium bis(oxalato)borate (LiBOB), lithium difluoro(oxalato)borate (LiDFOB), salts to form zintl phases, and tris(pentafluorophenyl) borane (TPFPB) [65]. Pentafluorophenyl isocyanate (PFPI) was found to give similar performance as VC and FEC when 2 % by weight was added to the electrolyte. Despite comparable capacity retention, the cell using PFPI had lower coulombic efficiency and was more likely to undergo oxidative decomposition [67]. A variety of silanes [68, 69] have been studied as additives because the linkage of Si-O-Si bonds on the surface of the silicon is thought to help better passivate the surface leading to a higher capacity and capacity retention [68]. Additives such as LiBOB and succinic anhydride have also been studied and although they improved performance over the base electrolyte, they still don’t provide the cycle life and coulombic efficiency seen with the use of VC or FEC [70]. It should also be noted that a variety of electrode form factors (thin films, nanoparticles, etc.) and compositions (completely silicon, silicon-carbon mixtures, etc.) are studied and that most studies perform internal comparisons between a variety of additives. If this is not done, the comparison between additives becomes difficult. Furthermore, additives that work in one situation may not work in a full cell or with different anode formulations. Although FEC is agreed to improve specific capacity and cycle life, the exact composition, and therefore the proposed cause of the enhancement, has varied across the literature. Despite variations in base electrolytes, additives concentrations, cycling conditions, and characterization techniques, there is a consensus that the SEI, in the presence of FEC, is thinner, more uniform, has a lower propensity to crack, and contains a greater percentage 9 of polymeric species [47, 71–75]. It has also been found that FEC creates a more distinct transition from the silicon to the SEI, whereas without FEC, the SEI and silicon fuse together indicating greater reactivity with the silicon [72, 73]. The fluorine content and its role in a more stable SEI, however, is debated in the literature [71]. Some observed that adding FEC creates a denser, more stable SEI containing more LiF [72–80]. Fluorine ions are thought to be formed from the reduction of FEC, providing a fluorine source in addition to the LiPF6 salt. The FEC also undergoes a ring-opening, leading to the formation of poly(VC) [65, 76] and poly(FEC) species [75]. The fluorine ions can then react with silicon-oxide species forming a SEI containing LiO2 and LiF [65, 79]. LiF provides better lithium transport through the SEI while the higher polymer content provides flexibility [65, 75]. Conversely, others have seen a decrease or little change in the LiF content in SEI layers formed in the presence of FEC as compared to the base electrolyte [71, 81–84]. This brings into question whether the fluoride in LiF comes from LiPF6 or FEC. A lower concentration of LiF means a longer cycle life, because less cyclable lithium is irreversibly lost [82, 84]. Veith et al. found that FEC-free conditions led to a SEI containing much more LiF than when FEC was present [71]. The FEC SEI contained a greater concentration of polymers. The improved flexibility allowed the SEI to readjust and move unlike a LiF-based SEI. Despite the improvement with FEC, the continuous consumption of FEC with cycling leads to an eventual breakdown in performance. Furthermore, the SEI still breathes during each cycle. Veith et al. found that the SEI was more organic at high lithiation states and more inorganic at low lithiation states. This was attributed to the dissolution of the polymer [71]. Haruta et al. [10] created a model film of LiF on a silicon thin film to try and study its efficacy as an SEI. Initial coulombic efficiencies were improved over bare silicon but gave poor capacity retention in comparison to cells with base electrolyte and FEC added to the electrolyte. This indicates that both inorganic and organic components of the SEI are necessary for good performance. VC also improves silicon lithium ion battery performance [75, 84]. The double bond in VC is thought to polymerize during decomposition forming insoluble polymeric materials [85]. The main benefit of VC as an additive, stems from the increase in polymeric species, adding flexibility to the SEI [78, 84]. Too much polymeric character, however, is thought 10 to lead to a thicker SEI [75]. Furthermore, if too much VC is utilized then the resistance of the cell has been found to increase, decreasing efficiency and increasing self-discharge. For an additive to be successful, it must help form a mechanically stable SEI that can expand and contract without cracking while remaining permeable to Li ions. The SEI must be able to prevent the continuous reduction of electrolyte. Other useful additive characteristics include having a higher reduction potential than the base electrolyte, being thermally stable, having minimal reactivity with cell components, and having high anodic stability [66]. To outperform FEC, the additive must create an SEI that doesn’t consume the additive each cycle, or consumes so little that the desired cycle life can be achieved. These criteria can aid in the rational design of new electrolyte additives to build a more stable SEI. The exact composition and structure of the SEI, with and without FEC, are still debated; however, the improved performance is not. Both VC and FEC have a carbonate ring structure and carbonate solvents have been most successful in silicon lithium ion batteries so far. Furthermore, if the polymeric character of the SEI is more important than the presence of LiF, then a fluorinated additive may not be required. A more recent approach is to use electrolyte additives to form a zintl phase to protect the surface of the silicon and allow for the formation of a stable SEI. The silicon surface is normally electron imprecise which leads to high reactivity. The use of additives in the form of MTFSI where M = Mg2+ , Al3+ , results in a reaction with the silicon to form a less reactive surface by creating an electron precise surface containing the M species replacing the dangling bonds present in native silicon [86, 87]. 1.3.1.2 Surface Coatings and Functionalization A thick silicon dioxide layer is thought to prevent the electrolyte from being affected by the potential of the silicon surface and thus inhibit electrolyte decomposition. However, at these thicknesses, lithium also can’t pass through easily, causing the silicon to not fully lithiate [47]. The use of electrochemical quartz crystal microbalance (EQCM) has also shown that the lithiated silicon dioxide layer starts to dissolve and lose Li2 O to the solution resulting in irreversible capacity loss [47]. The native oxide on silicon helps to prevent direct contact between the electrolyte and the silicon surface, but thicker oxides quickly 11 start to inhibit lithium transport. Thus, silicon dioxide coatings alone are unlikely to solve silicon’s poor cycle life. Gao et al. [88] used click chemistry to functionalize silicon nanoparticles with a variety of chemical groups. The term, click chemistry, describes reactions that are easy to perform, high yielding, create inoffensive byproducts (removed by nonchromatographic methods), and are stereospecific [89]. The covalently bound functional groups, along with the normal reduction products of the electrolyte, created a more chemically stable SEI. A dramatic improvement in capacity retention and coulombic efficiency were observed over that of unfunctionalized silicon nanoparticles. Although the functionalization improved chemical stability, the SEI still suffered from mechanical pulverization, but at a slower rate than the untreated silicon. Another method that has been pursued is to create a hollow, core-shell structure surrounding the silicon to protect the silicon from interaction with the electrolyte while simultaneously providing space for the expansion of the silicon [3,90,91]. Hollow nanospheres and nanotubes have been investigated. Typical shell materials include carbon and metal oxides such as TiO2 . The addition of nonelectrochemically active surface materials reduces the energy density of the electrode. Other major issues with this approach include often complex synthesis pathways and that if the shell is mechanically or chemically compromised, there is nothing left to protect the underlying silicon. Therefore, self-healing SEIs are also being investigated [90]. When the SEI is formed, there is some loss of lithium from the system to form the passivation layer on the silicon surface. Prelithiation either through the electrolyte, the cathode, or the anode can aid in preventing capacity fade by providing extra lithium for the lithium lost to the formation of the SEI [1]. Using this in combination with other investigated techniques can help mitigate the negative effects of an unstable silicon SEI. However, to create a truly stable passivation layer, comparable to the graphite SEI, a more thorough understanding of the SEI on silicon is necessary. 1.3.2 Characterization of the SEI The SEI, on both silicon and other anode materials, has been studied to better understand its structure and composition. The SEI is thin and air sensitive, so its characterization is nontrivial. Different characterization techniques used to study the SEI and their appli- 12 cations along with some examples from the literature are shown in Table 1.1. Each study is labeled as either ex situ or in situ. In situ methods are preferable because ex situ processes have been shown to damage the SEI and potentially lose information on the outermost, organic layer [15, 92, 93]. In situ studies are those in which the data is acquired within the cell. The electrode is not removed and is never exposed to ambient conditions. Chemical reactions between silicon and electrolyte begin as soon as the two come in contact. The electrochemical contribution to the SEI through the reduction of the electrolyte begins at potentials below ∼ 1.8 V but is dependent on the electrolyte [44]. The thickness and composition of the SEI on silicon have been measured both ex situ and in situ with conflicting results [13, 18, 92]. It is estimated to be typically around 50 nm or less, but has also been reported as thick as 200 nm. The variation in measured thickness is a function of characterization technique (ex situ versus in situ), electrolyte choice, and cycling conditions [11]. The SEI on silicon has been found to change substantially during cycling in thickness and composition; the silicon SEI layer has been said to breathe since it expands and contracts as it grows, unlike its graphite analog [19, 71]. The SEI is also observed to be rough with nonuniform coverage [42] although the use of ex situ techniques creates uncertainty in whether this is due to nonuniform formation or is observed because of damage during sample preparation. The nonuniformity, reactivity with air, and fragility of the SEI, makes its characterization difficult and necessitates the use of in situ methods. 1.4 Aging of Lithium Ion Batteries There are several performance metrics that are important for batteries including capacity (how much charge can be stored by the battery), rate capability (how quickly the battery can be charged and discharged), cycle life (how many cycles the battery can undergo before reaching 80% of the original capacity), and calendar life (how long the battery can rest in the charged state until it loses 80% of its original capacity). This work will focus on calendar life and cycle life. 1.4.1 Cycle and Calendar Aging Batteries undergo constant current or galvanostatic cycling to quantify cycle life. A constant current is applied and the battery cycles between an upper and lower voltage 13 cutoff. This is demonstrated in Figure 1.5A for a lithium-nickel-manganese-cobalt-oxide (NMC) 622 - 80% Si full cell. The applied current is often described in terms of c-rate. The rate of 1C is the current required for the battery to discharge in 1 hour. Thus, a c-rate of C/10 would take 10 hours to discharge and a rate of 4C would take 15 minutes to discharge. Constant current constant voltage (CCCV) cycling is also common in which constant current is applied until a voltage limit is reached and then the voltage is held at the cutoff value until the current decays below a specified value. This allows for full charge/discharge of the battery since overpotentials due to cell impedance and lithiation/delithiation hysteresis can make the cell potential different than expected once the CC step finishes. The number of cycles achieved before the capacity drops below the 80% of the initial value, is the cycle life of the battery. Another important measure of battery performance is the calendar life or how long the battery can operate (including rest periods in the charged state) before the capacity drops below 80 %. When a battery rests, parasitic processes continue, consuming lithium inventory that could have been useful capacity. Calendar life is quantified by periodic cycling in between rest periods in the charged state. The intermittent cycling is required to check changes in capacity of the cell and is called a reference performance test (RPT). An example of a basic calendar aging test is shown in Figure 1.5B. More rigorous United States Advanced Battery Consortium (USABC) calendar aging also include voltage pulses to keep the battery at a similar potential during aging despite losing charge. 1.4.2 Degradation Mechanisms Relevant to Calendar Aging Calendar aging of LIBs causes capacity fade and impedance rise. Traditionally, most capacity fade due to calendar aging is caused by losses of Li+ to the SEI of graphite anodes, and is largely a function of the electrochemical potential of the anode at the state-of-charge (SOC) of storage [94]. The lower the negative electrode potential (i.e., the higher the cell SOC), the larger the driving force for electrolyte reduction and concomitant capacity loss [95]. In addition to these irreversible self-discharge processes, reversible self-discharge can cause apparent capacity losses, decreasing charge during storage that can be recovered after additional cycling [96]. Impedance rise can occur through more diverse pathways: from surface transformation at cathode particles, [97] pore inaccessibility due to significant 14 SEI growth at the anode, [98] and electrolyte depletion following sustained decomposition [99]. These aging mechanisms depend on storage temperature and chemical composition of the electrolyte, anode and cathode [98]. Evaluating calendar aging is resource- and timeintensive. Given the multiyear lifetimes typical of today’s LIBs, calendar life is generally extrapolated from shorter tests rather than directly measured. Even the shorter experiments used for extrapolation extend through several months or years, often at elevated temperatures to accelerate aging [94, 100]. The universal use of graphite anodes have helped make aging trends and life forecasting more predictable, notwithstanding, many uncertainties arise when moving to different electrode chemistries, such as the addition of silicon. During the past two decades, silicon has been studied as a promising anode material that can increase cell energy while decreasing manufacturing costs. The Li-Si reactions lead to severe dimensional changes that contribute to rapid performance fade in Si-containing electrodes. Mitigating this and other silicon-specific phenomena has been the focus of many battery scientists in recent years. Consequently, Si-containing batteries are now increasingly a reality in the market, with cells used in electric vehicles reportedly containing small amounts of Si as an additive to graphite anodes, [101] and many companies exploring Si-rich electrode compositions. Given that many effects of calendar aging arise at the anode, and the chemistry of Si is markedly different from that of graphite, this shift in anode composition has an impact on how the cells age as a function of time. As Si anodes approach technological maturity, a critical question must then be answered: How do chemical and electrochemical interactions between silicon and the electrolyte affect calendar aging, especially compared to stable electrodes like graphite? Information from the U.S. Department of Energy (DOE) indicates that this remains an open question [20]. Figure 1.1A contains data from several leading manufacturers of Si-containing cells, demonstrating how cycle lives and energy densities have improved over the past decade and are quickly approaching performance targets set by the DOE. Conversely, Figure 1.1B shows that the calendar lives from the latest iteration of these high-performing cells remains unsatisfactory, suggesting that strategies to improve cycling are insufficient to promote long-term stability. Despite this striking calendar life technical gap, very few studies have probed the long-term stability of Si-containing cells in the absence of cycling [21–23]. A rare example is the work by Zilberman et al. which examined 15 cylindrical cells with 3.5 wt% of Si in the anode that were stored for 11 months at 25 ◦ C [96]. The study found that most of the capacity fade observed at all SOCs was due to permanent losses of Si active material capacity. This active material loss is typically attributed to the electrical isolation of silicon either from cracking or the formation of a resistive layer due to reactivity with the electrolyte, blocking Li+ transport. Conversely, traditional graphite anodes should remain stable under these same conditions, with Li+ trapping in the SEI as the main capacity fade mechanism. While the stability of the active material is of concern, there is also evidence suggesting that the Si SEI is less passivating when compared to graphite. Kalaga et al. estimated the currents attributed to these parasitic processes and suggested that the rate of SEI “growth” was higher on silicon-graphite anodes (15 wt% Si) compared to Si-free electrodes [102]. Based on these studies, active material loss, chemical instability, and Li+ inventory consumption appear to be exacerbated with the addition of Si, contributing to increased self-discharge and the inadequate calendar lives observed in Figure 1.1B. Silicon and its SEI are extremely reactive. The addition of Si to the anode promotes a variety of new side-reactions, resulting in gassing, dissolution of the SEI, and electrolyte decomposition, some of which can be traced back to the early periods of cell life. Figure 1.6 displays a summary of the different failure mechanisms, stemming from the reactivity of Si and its SEI, that may contribute to poor calendar life. When Si comes into contact with conventional carbonate electrolytes, surface changes occur as a result of chemical processes, even when a native oxide layer is present [103, 104]. Complementary studies by Pekarek et al. and Seitzinger et al. demonstrated the reactivity of various materials and chemical groups by examining the solid-phase and gaseous components following exposure to electrolyte, [105, 106] and found that different molecular coatings have varied levels of stability. Hydrogen-terminated Si, acidic SiO2 , and silyl ester surface groups were found to be reactive, whereas basic SiO2 and silyl ethers were more chemically robust, demonstrating the importance of material choice and control of interfacial chemistry for silicon anodes. Furthermore, there is growing evidence suggesting that Si increases the anode reactivity even after the formation of the SEI. While the SEI on graphite can be made notably stable by electrolyte additives, [107] the Si SEI appears to be highly dynamic [108] and 16 inherently nonpassivating (Figure 1.6). It has been reported that the passivation layer of model Si wafer electrodes breathes during cycling, becoming thinner and more inorganic during lithiation but thicker and more organic during delithiation with certain electrolytes [18]. These compositional changes are believed to originate from the precipitation and dissolution of SEI components such as lithium ethylene dicarbonate (LiEDC), LiPF6 decomposition products, and polyethylene glycol (PEG) oligomeric species [109]. The addition of fluoroethylene carbonate (FEC) to the electrolyte, known for its beneficial properties to Si, [110] did not preclude this breathing but only reversed the order of observed thickness changes despite improving silicon passivation [71]. Further evidence of the dynamic SEI formed in the presence of FEC and the reactivity of silicon comes from the observed consumption of FEC with cycling [111]. Dynamic SEI thickness and composition are also observed on graphite, especially during formation, [112] but the graphite SEI eventually passivates indicating that the relationship between a dynamic SEI and passivation is not fully understood. Although the reactivity of Si with the electrolyte during cycling has been well documented, fewer studies have directly probed the stability of Si surfaces at open circuit conditions experienced during calendar aging. Yin et al. investigated how the SEI of Si wafers evolved during rests at intermediate SOCs, and found that a decrease in passivation and increase in reactivity could already be observed after brief (1–10 hr) periods [104]. Such behavior persisted after hundreds of lithiation-rest cycles and was associated with the intrinsic reactivity of organic components of the SEI. Stetson et al. also observed the shedding of an organic-rich layer of the SEI after 45 hr at open circuit, leaving a less resistive deposit at the Si surface [113]. These studies suggest that organic SEI components are unstable either because they continue to react or because they are soluble in the electrolyte, or both. A full understanding of the reactivity and reaction pathways of various Si SEI components has yet to be achieved [114]. The examples above show that Si is chemically unstable with salts and organic solvents commonly used in LIB electrolytes and that this instability does not subside upon the formation of the SEI, as the “protective” layer itself is not static (Figure 1.6). Hence, sustained losses of Li+ inventory are expected in Si-containing anodes, which is highly problematic for the calendar life of batteries. While anticipating the effects of silicon surface instability 17 on calendar aging is relatively straightforward, possible consequences of SEI “breathing” are less clear. These cyclic compositional changes could lead to unexpected dependencies of calendar aging on SOC, in which case the knowledge acquired from graphite-based cells would not be directly transferable to Si anodes. Thus, there is a need to perform comprehensive calendar aging studies for Si-rich cells at multiple SOCs and temperatures to gain insight into the mechanism of SEI growth. Furthermore, although consequences of volumetric changes have largely been overcome to improve cycle life using morphology and particle-size controls (Figure 1.1A), the dimensional change of silicon may still interfere with typical calendar life experiments. Self-discharge and concomitant contraction during storage would slowly disrupt the SEI, exposing Si to the electrolyte (Figure 1.6). Because of the combined reactivity and volume change of silicon, such mechanical failures are more detrimental than on graphite. Similarly, perturbation to the silicon SEI during periodic cycling to measure calendar life (reference performance tests, RPTs) is more substantial than on graphite, leading to experimental outcomes that could depend on cycling frequency. Further studies are required to understand the contributions of mechanical degradation, as a function of SOC, to poor silicon calendar life. Measuring stresses in silicon and the SEI in situ over time could help quantify how mechanical failure relates to self-discharge. Silicon promotes a runaway hydrolytic HF cycle. Generation of hydrofluoric acid (HF) is a known consequence of using LiPF6 -based electrolytes, as this salt is hydrolyzed by residual moisture introduced during cell assembly (Equations 1.2 - 1.4) [102, 115]. LiPF6 ⇌ Li+ + PF6− (1.2) LiPF6 ⇌ LiF + PF5 (1.3) PF5 + 2H2 O → 3HF + HPO2 F2 (1.4) Many components in traditional LIBs are comparatively inert to HF, including commercial binders, graphite anodes, and polyolefin separators [116]. Thus, batteries with these tradi- 18 tional components typically tolerate the HF generated by water contamination up to 1000 ppm without significant performance degradation [117]. However, a completely different scenario occurs when Si-based materials are present in the cell. In this case, HF reacts with any oxide-containing Si species to generate more water as well as gaseous and soluble products: [118] 4HF + SiO2 → SiF4 + 2H2 O (1.5) 6HF + SiO2 → 2H2 O + H2 SiF6 (1.6) Generation of water allows further hydrolytic production of HF, which can continue indefinitely if the oxide-containing Si species are accessible to HF (Figure 1.6). The consequences of this chemical attack to surface Si-O groups were highlighted in a recent paper by Ha et al., which showed that as little as 50 ppm of moisture in the electrolyte is sufficient to pit the surface of Si wafers. Such reactions were observed to expose fresh surfaces to the electrolyte, leading to additional parasitic reactions and resulting in thick, nonuniform SEI layers. This runaway cycle of etching and HF regeneration could have severe implications for the calendar life of cells with Si electrodes. PF6 – hydrolysis is a relatively slow process, [115] and thus the effective concentration of HF and its byproducts in the electrolyte is expected to increase over time until achieving equilibrium, while the LiPF6 necessary for cycling the cell is simultaneously depleted. The susceptibility of the oxygenated Si surfaces to acidolysis can eventually generate problems of interfacial incompatibility in the anode. Much of the research on binders for Si electrodes have explored these functional groups as anchoring points for polymer chains, [119] which would slowly vanish due to fluorination. Given the large dimensional changes of Si during cycling, these weakened interfaces could eventually cause electrode delamination and active material loss. Additionally, while HF etching of Si is typically limited to the oxide-containing surface layers in aqueous solutions, the non-aqueous electrolyte and extreme electrochemical potentials that are present in batteries result in a different chemical environment. As such, it may be possible for HF and its byproducts to etch and dissolve away the silicon active material itself, as evidenced by 19 deep pitting of Si wafers extending far below the native surface oxide layers [118]. Finally, it is possible that enriching the electrolyte with gaseous and soluble Si species (such as SiF4 and H2 SiF6 ) can impair cell health due to these soluble species being reprecipitated at the cathode, also known as “crosstalk.” Silica-like deposits have been detected at the cathode surface in Si-based full-cells, [120] indicating that these compounds can travel through the separator to be oxidized in the positive electrode, becoming a possible source of impedance rise. Most of the reactions above are chemical in nature and can occur independent of electrochemical cycling, making them likely causes of both active material loss and impedance rise during calendar aging of Si-containing cells. 1.5 Mechanical Characterization of Silicon and the SEI The techniques in Table 1.1 provide information on the morphology and chemical composition of the SEI, but there have been few studies on the mechanical properties of the SEI. Understanding the mechanical properties and instability of the SEI is important to prevent SEI failure and a subsequent reduction in cycle life. Additionally, stress in the silicon can affect the potential by 60-120 mV/GPa [29]. As with the SEI, the silicon beneath is also dynamic: expanding during lithiation and contracting during delithiation. It is often assumed that the expansion of silicon occurs linearly with the state of charge and that the silicon is confined by the underlying substrate, only allowing expansion out of plane [92]. Some options of possible mechanical failure mechanisms are shown in Figure 1.7. Cracking of the SEI versus cracking of the silicon are related, but it is not clear how. One hypothesis is that as the silicon expands, the SEI cracks, new electrode surfaces are exposed, and further reduction of the electrolyte occurs. In situ AFM has allowed for the study of the formation of cracks in the SEI and silicon. Cracks were observed in the SEI without cracking in silicon islands patterned on a copper current collector [121]. These SEI cracks, did lead to exposed silicon surfaces, however the SEI did not fill the entire crack, but formed at the bottom of the crack [121]. It is possible on longer time scales, that the new SEI would grow to the same thickness as the surrounding SEI, but this was not observed. Zheng et al. [122] used SPM to map the Young’s modulus of the SEI as a function of voltage, however the measurements were performed ex situ, after rinsing the electrodes. The study showed that there was a softer, outer layer of the SEI that was more prone to 20 dissolution, especially at increased temperatures. The effect of vinylene carbonate was also studied and they observed an increase in SEI coverage. Without vinylene carbonate, they found coverage of the SEI to only be 51% of the silicon thin film surface [122]. Chen et al. used digital image correlation (DIC) to measure the strain in graphite electrodes and showed the bi-axial strain state doubled from copper disks to copper foils demonstrating the in-plane constraint the substrate imposes on the silicon, with a thinner substrate allowing for more in-plane strain [76]. In situ MOSS has been used to study the stress in the silicon anode, [37, 39–41] as well as the SEI; [121] however, the contributions from the SEI and the anode were not quantitatively distinguished. For a 50 nm silicon thin film, the studies agreed that the silicon electrode in a half cell undergoes an initially compressive, elastic stress. At approximately -1 GPa and 0.5 V, the silicon begins to flow plastically. From 0.25 – 0.15 V, the stress stabilizes at -1.2 GPa. The compressive stress then decreases for the remainder of the discharge cycle. Upon charging, the compressive stress continues to decrease until the direction of flow reverses and the electrode experiences a tensile stress [37, 39–41]. Only data for the 2nd and 3rd cycles were provided to avoid considering the stress during the initial SEI formation cycle. Even for the 2nd and 3rd cycles, a large compressive stress of -1 GPa was observed prior to reaching the onset of lithiation. Thus, the SEI could possibly play a role in the initial stress since the reduction of electrolyte also occurs at higher voltages. Mukhopadhyay et al. [123] studied the stress on a graphite anode while cycling using MOSS. An irreversible change in stress was observed. This change in stress was attributed to the SEI, because it was greatest in the initial cycles and did not seem to depend on the thickness of the electrode. Tokranov et al. [121] then attempted to distinguish between stress in a graphite anode versus the SEI using MOSS. The voltage was held at potentials above the onset of lithiation and the stress was obtained by measuring the change in curvature of the electrode. It was assumed that any change in stress must be due to the formation of the SEI. A substantial change in the measured stress was not observed until 0.6 V which is different than what was observed under constant current conditions, where -1 GPa was observed prior to 0.5 V. Although useful in determining the average stress in the silicon and SEI at equilibrium, this method is disadvantageous, because it is an indirect measurement of stress. MOSS calculates the curvature of the substrate by looking at the 21 displacement between an array of beams incident on the surface. This curvature must then be converted to stress using some variation of Stoney’s equation. Furthermore, the MOSS measures an average stress based on the change in curvature over both SEI and silicon; it is not spatially resolved and measures an average over the entire film. Tavassol et al. [124] measured the change in curvature of a cantilever with a graphite slurry coating. The stress in the cantilever was compared to the strain in a free-standing graphite coating that was removed from the current collector. A stiffness parameter was defined to relate the strain, stress, and potential of the cell. The stress was found to be a function of the rate of lithiation of the electrode while the strain in the electrode was purely a function of the amount of lithium in the graphite. Only data from the third cycle onward were reported. Kumar et al. published two works studying the strain in the solid electrolyte interphase. In the first, in situ AFM was used to track the displacement of easily identifiable surface features to calculate the strain in the SEI as a function of state of charge (SOC) and estimate the fracture energy of the SEI [125]. This was done for patterned silicon electrodes to enable strain in the lateral as well as the out-of-plane directions. Cracking was observed on the edges of the raised portions of the patterned silicon. The cracks were too small to allow AFM to probe the bottom of the cracks. This led to the second publication, in which Li inventory calculations in combination with AFM and TOF-SIMS were used to relate the capacity loss due to SEI to the observed strain [11]. Liu et al. looked at the macroscale mechanical failure mechanisms of the SEI on anodes [126]. When a thin film on a plastic substrate under tensile stress is above its yield stress, it is likely to crack. Conversely, when a thin film is under sufficient compressive stress, it will wrinkle. This wrinkling can lead to cracking and delamination in layered structures. The cyclic loading can lead to cumulative plastic deformation called ratcheting, where the amplitude of wrinkling is amplified over time, as depicted in Figure 1.7F. Another possible behavior of a perfectly planar thin film with mismatched stress undergoing a stress cycle is shakedown. Shakedown is the restoration of a steady-state elastic behavior in a cyclic process after an initial plastic deformation. This manifests as a shift in the stress-strain curve due to the plastic deformation followed by elastic behavior. This study looked at the relationship between wrinkling and ratcheting by calculating the critical strain required for wrinkling [126]. The results culminated in a phase diagram predicting what behavior (elas- 22 tic without wrinkling, shakedown, shakedown with wrinkling, ratcheting, elastic with wrinkling) will occur at a specific strain and substrate thickness to SEI thickness ratio. A lower volume change may result in stable SEI shakedown, but with large substrate volume changes, the SEI was found to be far more likely to demonstrate ratcheting behavior [126]. Depending on what failure mechanism dominates, the electrodes could be designed to mitigate the failure mode. For example, if wrinkling and delamination are expected, a tensile prestress could be used to offset the expected large compressive stress expected in the SEI during delithiation [126]. Studies on the mechanical degradation of the SEI have provided valuable information on the effect of the SEI deformation on lithium inventory as well as confirmed the structure of a harder, inner layer formed by inorganic compounds and a softer, outer shell composed of organic oligomers. However, there are still many questions left unanswered, including: Where is the observed stress prior to the lithiation potential coming from? What is the relationship between stress in the silicon and stress in the SEI? What is the stress in the SEI and silicon upon initial introduction of electrolyte? To date, there has not been a direct, in situ measurement of the stress exclusively in the SEI on a silicon anode capable of answering these questions. As motivated above, in situ, direct mechanical and electrochemical studies can bolster the understanding of the SEI. It is the first hypothesis of this dissertation that stress evolution in the SEI will affect electrochemical performance of the anode, leading to different electrochemical conditions corresponding to the state of stress in the SEI. There are a multitude of techniques available to obtain the stress or strain of a material [127]. Desired qualities for studying mechanical degradation of the SEI and silicon include 1) capability to measure stress in amorphous material, 2) capability to distinguish between contributions of stress from the SEI and silicon electrode, 3) in situ operation, 4) to represent a normal battery as closely as possible, and 5) ability for in situ characterization under different conditions. Potential stress measurement techniques are listed in Table 1.2 along with their respective advantages and disadvantages and an assessment on how well they fulfill the above six criteria. Based on the above criteria, two of the techniques were explored in this work to try and study mechanical versus chemical failure of the SEI including moiré interferometry and scanning electrochemical microscopy (SECM). 23 1.5.1 Moiré Microscopy Moiré fringes are created when two gratings of similar periodicity are overlaid and offset. Changes in the moiré fringes correspond to displacements in the gratings forming the fringe [128]. The derivative of the displacement field yields the strain in the material. The two gratings are referred to as a sample and a reference grating. The sample grating is patterned directly onto the sample and is expected to change as the sample is strained. The reference grating acts as a static comparison and remains unchanged as the sample deforms. The reference grating can be in physical contact with the sample grating, placed near to the sample, or can be projected onto the sample. The most common approach for gratings are amplitude gratings in which the lines and spaces of the grating have opposite transparency or reflectivity. In a transparent amplitude grating, the lines and spaces alternate between near 100 % transparency and absorption as light passes through the grating. Similarly, reflection amplitude gratings alternate between near 100 % reflection and absorption. The reference grating can also be formed from the interference between two beams of light as in moiré interferometry. In all cases, the shift in the sample grating relative to the reference grating can be calculated from the moiré fringe frequency using Equation 1.7 where Fmoire is the moiré fringe frequency and f1 and f2 are the frequencies of the sample or reference grating with f1 > f2 . Fmoire = f 1 − f 2 (1.7) Moiré can detect on the order of 10 nm changes in displacement and is therefore a useful tool in the measurement of strain [128]. As the sample undergoes small deformations, large changes in the moiré fringe allow for measurements of displacement below the diffraction limit of light. Figure 1.8 shows how the moiré fringe changes as the sample grating deforms. 1.5.2 The Relationship Between Stress and Strain The strain in a material is the ratio of the displacement over the original length. How the material deforms is dependent on its composition and the applied loading. Linear elastic deformation is defined by the linear portion of the stress (σ) – strain (ϵ) curve and 24 strain within this region can be relaxed to regain the original form of the material. The slope of the linear region is called Young’s modulus (E) as shown by Equation 1.8. An example of a stress strain curve and the linear elastic region is given in Figure 1.9. σ = Eϵ (1.8) A smaller value of Young’s modulus indicates a more elastic material. Typical moduli for relevant materials are shown in Table 1.3. For metals, the nonlinear portion of the stress-strain graph represents irreversible, plastic deformation in which the material will not relax to its original shape. Within the linearly elastic region, the relationship between strain and stress is described simply by the proportionality constant, Young’s modulus. However, the relationship becomes more complicated for nonlinear elastic materials or during plastic deformation. During plastic deformation, a single strain value can describe multiple stress states and the relationship between stress and strain becomes more complicated [129]. 1.5.3 Mechanics of Thin Films and Interfaces The mechanics of the SEI, silicon, and current collector are all related and differentiating between electrochemical versus mechanical degradation of the SEI is difficult. The presence of binder and conductive additives in slurries only increases this complexity. Therefore, as a starting point, the use of model silicon thin films can be useful in understanding the mechanical interaction between the silicon electrode and the SEI. Blanket silicon on a thick current collector or substrate can limit the in-plane strain of the SEI which is of interest in understanding crack formation in the SEI. To use a simplified thin film system, several requirements and assumptions are necessary. These include good adhesion between all interfaces (inner SEI to silicon and silicon to current collector) and a thin or compliant substrate to observe strains on the backside of the electrode. AFM studies have shown that the inner SEI and silicon are well adhered, [130] implying that a backside technique may be feasible to observe strain in the inner SEI, but not the poorly adhered outer organic SEI layer. Previous studies have shown that a large change in stress occurs as the SEI forms on a silicon thin film despite being above the lithiation potential [39]. The portion of this stress that is caused by an in-plane strain can be quantified as the misfit strain between 25 the SEI and silicon. The misfit strain is defined as how much the SEI (film) would expand laterally if it were not restricted by the underlying substrate. This misfit strain also likely changes as a function of potential as the composition and thickness of the SEI changes and as lithium passes through the SEI and lithiates the silicon at potentials below ∼ 0.5 V versus Li/Li+ . The misfit strain for a two-layer system is given by Equation 1.9 where f is the force on the membrane as a result of the misfit strain, hf is the film thickness, hs is the substrate thickness, b is the length of the film and E’s and E’f are the biaxial moduli of the substrate and film, respectively. The Biaxial modulus is defined in Equation 1.10 where νs is the Poisson’s ratio. The strain in the SEI will result in a strain and curvature in the silicon and then the substrate as shown in Figure 1.10. The left portion of the figure depicts a simplified version of the force applied to each layer due to the misfit strain if there were no curvature, but in reality each film has curves and has a through-thickness strain distribution. If the substrate is compliant or thin enough and well adhered, the strain visible from the backside of the substrate is greater. ∆ϵ = f Es′ bhs E′ = − −f E′f bh f E (1 − ν ) (1.9) (1.10) A common way of determining the stress in a thin film on a substrate is to measure the curvature of the material and use Stoney’s equation [131] to calculate the corresponding stress. Stoney hypothesized that metals are deposited under varying amounts of tension, straining the material they are deposited on. This strain causes the underlying material to bend and by measuring the amount of bending, the tension in the deposited film can be calculated. No knowledge of the elastic parameters of the film is required. The main assumptions in the original version of Stoney’s equation are that (1) the film and substrate thicknesses are uniform and have the same radius, (2) the substrate thickness is much greater than the film thickness and that both thicknesses are much less than the radius, (3) the film and substrate are homogeneous, isotropic, and linearly elastic, (4) the strains and rotations of the substrate are infinitesimal, (5) the film stress is in-plane isotropic, (6) the curvature is constant in all directions, and (7) all nonzero stress and curvature components 26 are uniform throughout the interface [132]. Stoney’s equation was later improved for biaxial stress which replaced Young’s modulus with the biaxial modulus (Equation 1.10). Several other publications have worked towards relaxing the assumptions of the original equation [132]. Stoney’s equation is shown in Equation 1.11 where κcurv is the curvature, hf is the film thickness, hs is the substrate thickness, σmf is the mean stress in the film, and E’s is the biaxial modulus. κcurv = 6h f σm f Es′ h2s (1.11) A more general expression for the curvature, where the film and substrate can be any thickness, is given by Equation 1.12 [133]. κcurv = 6Es′ E′f (h f + hs )h f hs ∆ϵ E′f 2 h f 4 + 4Es′ E′f h f 3 hs + 6Es′ E′f h f 2 hs + 4Es′ E′f h f hs 3 + Es′ 2 hs 4 (1.12) The through-thickness strain distribution of the film and substrate can be determined from Equations 1.13 and 1.14 [133] for the coordinate system shown in Figure 1.10, with the origin at the interface between the film and substrate. ϵf = −f + E′f κcurv (y − δ) bh f (1.13) ϵs = f + Es′ κcurv (y − δ) bh f (1.14) The neutral plane, where no strain occurs due to curvature, is given by position δ, shown in Equation 1.15. δ= h2f E′f − h2s Es′ 2(h f E′f + hs Es′ ) (1.15) Using Equations 1.12, 1.13, 1.14, and 1.15, it is possible to estimate the strain distribution in a substrate due to misfit stress in the SEI. Sheldon et al. [11,125,130] patterned Si islands on a quartz wafer with a current collector between the Si and the quartz wafer. Using AFM, 27 they observed a strain in the SEI only when the Si islands were small enough that the SEI was not constrained by the underlying substrate [11, 125]. In finite element method (FEM) models, they predicted an SEI strain of ≈ 0.01 strain at the surface of the inner SEI at a state of charge (SOC) of 0.1 at the edge of an island where the SEI was no longer constrained by the substrate. A SOC of 0.1 corresponds to an estimated 10% lithiation of the silicon. At this SOC, the SEI should have partially formed and lithiation is just beginning. Taking 0.01 strain as the misfit strain that the entire SEI would undergo if it were not constrained by the substrate, the strain distribution in the silicon and substrate is estimated and is depicted in Figure 1.11 for a 50 nm Si thin film on a 2 µm Cu substrate. Parameters for the calculations are given in Table 1.3 [130]. These calculations are meant to estimate the required strain resolution to observe changes in strain in the substrate due to the SEI. Figure 1.12 shows how the strain at the backside of the substrate changes as a function of substrate thickness and modulus while keeping silicon and SEI properties constant. To maximize the signal to noise ratio of changes in strain on the backside of the electrode, due to a change in SEI strain, the thickness of the Cu substrate should be minimized or a different substrate with a lower modulus can be utilized to enable the use of moiré microscopy to directly measure in-plane strain. 1.6 Electrochemical Characterization Electrochemical techniques can help study SEI formation. Typical, galvanostatic battery cycling involves a constant current, with high and low voltage cutoff values, which does not give specific information about the surface of the electrode (an example of this is shown in Figure 1.5A). Other useful electrochemical methods relevant to the current work include cyclic voltammetry (CV), differential capacity (dQ/dV) plots, scanning electrochemical microscopy (SECM), and electrochemical impedance spectroscopy (EIS). 1.6.1 Cyclic Voltammetry (CV) Cyclic voltammetry is a technique where the potential is cycled linearly between two potential values while the current response is recorded. The presence of peaks in the cyclic voltammogram (CV) indicate the presence of electrochemically active species that are reduced or oxidized at the potential of the peak giving information on the lithiation 28 and delithiation of silicon as well as the reduction of the electrolyte to form the SEI. An example voltammogram for amorphous silicon is given by Figure 1.3. Differential capacity (dQ/dV) plots, derived from constant current cycling data, can also provide similar information where peaks show an increase in charge transferred as a function of voltage indicating a potential where there is increased reduction or oxidation. Differential capacity plots are generated from galvanostatic cycling data by taking the change in charge over the change in voltage which indicates how much chemistry is possible at each potential step. 1.6.2 Scanning Electrochemical Microscopy (SECM) SECM can show changes in reactivity in the surface, which does not by itself give information on mechanical failure; however, by imaging the surface it is possible to observe SEI cracking and poor SEI passivation [134]. If a silicon thin film with a thickness below 150 nm is used, cracking of the silicon is not typically observed, so changes in reactivity observed by SECM could show mechanical failure of the SEI. SECM is a probe technique that is useful in evaluating the kinetics and reactivity of an electrode surface. The SECM system is a 4 electrode system with two working electrodes, a counter electrode, and a reference electrode. One of the working electrodes is an ultramicro electrode probe (UME), which is typically small, with a radius on the order of 10s of ums to nms, and determines the feature size that can be resolved on a substrate surface. The other working electrode lead can be used to polarize the substrate (the surface under investigation). When a UME is in the bulk solution with a reversible redox probe and polarized to a potential to reduce or oxidize the molecule, a steady state current is achieved which is dictated by radial diffusion of the redox probe to the UME surface. The steady state current is given by Equation 1.18 where i∞ is the steady state current, a is the UME radius, D is the diffusion coefficient of the redox probe in the electrolyte, C ∗ is the bulk concentration of the redox probe, n is the number of electrons transferred in the redox probe reaction, and β is the correction factor for probe geometry (given by Equation 1.17 for a microdisc probe tip) [135, 136]. For microdisc electrodes, R g (Equation 1.16) is the ratio between the radius of the glass surrounding the disc (R glass ) electrode and the radius of the disk electrode itself (a). As R g increases, β goes to 1 and there is no correction needed for geometry. An example of the current in the bulk electrolyte is shown in Figure 1.13 on 29 the left for the Fc/Fc+ redox couple, which is a common reversible redox couple used with SECM. The CV of a UME has a sigmoidal shape unlike the typical “duck” shape of CVs on macroelectrodes due to the lack of diffusional limitations. Figure 1.14 shows the difference between the CV of Fc/Fc+ at a macro Pt electrode and a Pt UME. For the rest of the discussion on SECM, ferrocene will be used as an example redox probe. Rg = β = 1 + 0.639(1 − r glass a 2 2 1 1 ArcCos ) − 0.186[1 − ( ArcCos )2 ] π Rg π Rg i∞ = 4nFDC ∗ aβ (1.16) (1.17) (1.18) One of the most common types of SECM is feedback mode. In this mode, the UME is brought toward a surface that is in a solution with a reversible redox probe (in this case, ferrocene). If the surface is insulating, the presence of the surface starts to inhibit diffusion of Fc to the probe tip and the probe current decreases to less than the steady state current (Figure 1.13 middle). A completely insulating surface will lead to pure negative feedback. Conversely, if the surface is conductive, allows electron transfer, and is at a potential where it is energetically favorable to do so, the Fc+ that was oxidized by the UME can be reduced by the substrate then oxidized again by the UME. This creates a positive feedback loop and the tip current is greater than the steady state current (Figure 1.13 right). When the UME is brought to the surface at a constant speed the resulting curve of the UME current is called a probe scan curve (PSC). When the UME speed decreases as it approaches the surface, the resulting curve is called a probe approach curve (PAC). Both curves can be used to gain kinetic information about the substrate surface by plotting the normalized tip (Equation 1.19) against the normalized distance from the surface, L (Equation 1.20). The curves are a function of the distance from the substrate, the reactivity of the substrate, and the geometry of the probe. In Figure 1.15, examples of a positive feedback curve of a gold substrate is shown as a dashed black line, the negative feedback curve of an insulating alumina substrate is shown as a red dashed line, and a mixed feedback curve of a partially 30 passivated silicon surface is shown as a blue dashed line. Equations 1.22 - 1.25 [135–138] can be fitted to the approach curve to obtain values for κ, which is a nondimensional rate constant given by Equation 1.21 where D is the diffusion coefficient of the redox probe, k f is the forwards rate constant of the redox reaction, and a is the UME radius. The fit for each curve in Figure 1.15 is given by a solid line. itip i∞ itip,N = d a (1.20) ak f D (1.21) L= κ= α = ln2 + ln2(1 − cond itip,N ≈ α( R g ) + ins itip,N ≈ itip,N ( L, R g , κ ) ≈ (1.19) 1 2 1 2 ArcCos ) − ln2[1 − ( ArcCos )2 ] π Rg π Rg π 1 2 + (1 − α ( R g ) − ) ArcTanL β( R g )4ArcTanL 2β( R g ) π 2.08 (L R0g .358 2.08 (L R− g0 .358 cond itip (L − 0.145 R g ) + 1.585 + 0.0023R g ) + 1.57 + lnR g L + πR g 2 πR g ln (1 + 2L ) (1.22) (1.23) (1.24) ins ( L, R ) − 1 itip g 1 + , Rg ) + 0.006R g +0.113 0.0236R g +0.91 κ (1 + 2.47LκR0.31 κ ) g )(1 + L (1.25) Surfaces can be imaged for laterally changing kinetics by keeping the probe a set distance from the surface and scanning across the surface. Because the tip current is a function of both surface kinetics and distance from the surface, the tip current can be convoluted. One solution to this is to probe thin films which are flat relative to the size of the UME tip preventing current changes due to changes in surface height. One major benefit of this technique is that the reactivity of the substrate can be probed without necessarily polarizing to the redox probe potential. In the case of ferrocene, the redox potential is ∼ 3.35 versus Li/Li+ . When using SECM to study the SEI on silicon, 31 polarizing to such high potentials may irreversibly change the SEI and make the kinetics of the surface irrelevant to the actual SEI passivation behavior. However, not polarizing the substrate means that on a conductive surface where ferrocene received an electron from the substrate to create the positive feedback loop, the substrate will be oxidized during imaging and the potential of the substrate will shift throughout the experiment. This effect should be minimal since the current is on the order of nA, which would be equivalent to a c-rate of 5e-5 C and take 20,000 hours to fully charge or discharge the half cell whereas an experiment is on the order of 1-3 hours. 1.6.3 Electrochemical Impedance Spectroscopy (EIS) Potentiostatic EIS typically applies a small AC voltage signal to the cell, on the order of 5 mV, causing the cell to deviate from steady state behavior. An AC current output is measured and the impedance is then determined over a range of frequencies (typically 0.01 Hz to 1 MHz). The resulting response provides information on different contributions of impedance within the cell with different phenomena dominating at different frequencies. For example, charging of the electric double layer has a high characteristic frequency while diffusion dominates at low frequencies. A useful parameter for tracking the SEI is looking at changes of interfacial properties such as the interfacial resistance. Equivalent circuit models can be used to fit the impedance response and give physical meaning to the values obtained. Figure 1.16 shows a simple equivalent circuit for a battery. The ohmic resistance due to the electrolyte is given by Rb . The resistance and constant phase element of the SEI are given by RSEI and CPESEI , respectively. A constant phase element describes the capacitance of a nonplanar surface which is necessary for composite electrodes. The charge transfer resistance and constant phase element of the electrode are given by Rct ) and CPEelec . Wwarburg describes diffusion into the electrode. EIS is useful in understanding the changes in the surface resistance which can give information about SEI thickening and conductivity changes. 1.7 Approach to Understanding the Impact of Mechanical and Chemical Instability of the Si Anode SEI on Calendar Life Without a chemically and mechanically stable SEI, a lithium ion battery will rapidly fail. In order to improve calendar life, in addition to cycle life, the interplay between 32 chemical and mechanical degradation must be understood to inform mitigation strategies. The methods to improve cycle life do not appear to be directly transferrable to calendar life, so calendar aging of silicon must be systematically studied and understood in order to develop batteries that function well under cycle and calendar aging conditions. This dissertation studies the chemical and mechanical failure of the SEI during calendar aging. Calendar aging can be time intensive to study which can be extremely limiting when trying to develop new materials. Chapter 3 of this work focuses on developing a rapid qualitative stage gate to evaluate calendar aging. It was determined that, although difficult on silicon, such a stage gate is possible for qualitative purposes when experiments are carefully planned and executed. A procedure of how to get reliable results is discussed. Next, in Chapter 4, the passivation of model silicon thin films is investigated using SECM. Understanding how and why the SEI fails to passivate silicon during calendar aging despite good cycle life will be paramount to removing the technical gap shown in Figure 1.1. SEI strain was first studied via moir´e interferometry, but it was determined to not have sufficient tolerance to tilt during in situ measurements, and was discounted. Using SECM, the failure of SEI on silicon delithiated to 1.5 V versus Li/Li+ (a common high voltage cutoff for half cell cycling) was observed, providing direct evidence of the effects of over-discharging cells containing silicon. Passivation was more robust at 100 mV versus Li/Li+ (in the lithiated state). In both the lithiated and delithiated states, passivation decreased with rest time. The changes in passivation were global rather than localized cracking and the surface passivation became more heterogenous with time. Chapter 5 looks at the impact of disturbing the SEI with periodic cycling (RPTs). RPTs are necessary as a state of health check when aging cells at open circuit potential (OCV), but the frequency of cycling may impact the fade rate, convoluting calendar aging. The impact of mechanical failure of the SEI during RPTs was quantified using a variable OCV-RPT protocol to deconvolute whether temporal or cycling aging dominates. For the materials studied, it was found that time mattered more than cycling for rest periods of 1, 2, and 3 months whereas cycling started to impact the silicon at shorter rest periods of 1, 2, and 4 weeks. The relevance of these measurements is discussed since the Si electrodes used in this project already had poor cycle life. Lastly, microcalorimetry was used as a bridge between the mechanical and chemical understanding. 33 Table 1.1. Examples of characterization techniques utilized to study the solid electrolyte interphase on silicon as well as in other electrochemical systems. Each technique’s application is also listed. Ref. Si Ref. Other Technique SEI Application In situ Ex situ In situ Atomic force microscopy (AFM) Strain, topography [11, 92, 125] [42, 122] [139, 140] Auger electron microscopy (AES) Surface composition, depth profiling, morphology, quantify Li [63, 141] [142] Differential scanning calorimetry (DSC) SEI growth, phase changes [143] Electrochemical impedance spectroscopy (EIS) Impedance of the SEI [125, 145] [35, 146– 148] Electrochemical quartz crystal microbalance (EQCM) Tracking of interfacial changes in mass [47] [149, 150] Electron energy loss spectroscopy (EELS) Composition [44, 80] [151] Focused ion beam (FIB) Imaging, cross sections [82] [6, 152] Fourier transform infrared spectroscopy (FTIR) Composition Inductively coupled plasma optical emission spectroscopy (ICP-OES) Composition, elemental analysis [13] [13, 42, 84, 153] [153] Ex situ [144] [154] [72, 155] [156] 34 Table 1.1 Continued Ref. Si Ref. Other Technique SEI Application Matrix-assisted laser desorption/ionization (MALDI-TOF) Composition Multibeam optical stress sensor (MOSS) Stress Neutron reflectometry (NR) Composition, morphology Nuclear magnetic resonance spectroscopy (NMR) Li identification, composition Raman Composition Scanning electrochemical microscopy (SECM) Imaging, electron transport across the SEI Scanning electron microscopy (SEM) Imaging, morphology [42, 63, 74, 78, 84] (Scanning) Transmission electron microscopy ((S)TEM) Imaging [44, 80] Surface enhanced Raman spectroscopy (SERS) Composition Thermal gravimetric analysis (TGA) Thermal stability, composition [169] [170] Time-of-flight secondary ion mass spectrometry (TOF-SIMS) Depth profiles, composition [11, 79, 153] [121, 171] In situ Ex situ In situ Ex situ [157] [121] [158] [53] [140] [72, 155, 159] [76, 85] [154, 160, 161] [48] [162] [165] [147] [163, 164] [121] [166] [167, 168] 35 Table 1.1 Continued Ref. Si Technique SEI Application X-ray absorption spectroscopy (XAS) Composition X-ray photoelectron spectroscopy (XPS) Surface composition In situ Ref. Other Ex situ In situ Ex situ [172– 174] [42, 53, 63, 74, 80, 84, 141] [72, 121, 155, 175] 36 Table 1.2. Characterization techniques to measure stress. Desired capabilities include 1) capability to measure stress in amorphous material, 2) capability to distinguish between contributions of stress from the SEI and silicon electrode, 3) in situ operation, 4) to represent a normal battery as closely as possible, and 5) ability for in situ characterization under different conditions. Technique AFM Cantilever Pros Cons Desired Capabilities - Angstrom level resolution - Requires physical contact with the electrode surface Amorphous Capable? - Localized lithiation through AFM tip In situ? - Difficult to do long-term cycling Vary conditions? - Capable of measuring Young’s modulus, strain, and stress - Magnifies the displacement of the electrode reducing the necessary resolution - Difficult to de-convolute the displacement of the cantilever due to silicon versus SEI - Average over the entire sample SEI versus Si Stress? Normal geometry? Ref. [11, 122, 125, 139, 176– 178] Amorphous Capable? SEI versus Si Stress? In situ? Normal geometry? Vary conditions? [157, 179– 183] 37 Table 1.2 Continued Technique Pros Cons - Limited by resolution of imaging techniques Digital Image Correlation (DIC) EQCM Fiber Bragg Grating (FBG) - Method to track particles or other features and calculate strain - Measures the change in mass on the electrode surface - Measure stress of electrode in normal battery system - Complex data analysis and may require expensive software - May require speckle pattern on electrode which can possibly affect electrochemistry - - Measuring mechanical properties during the formation of the SEI may be difficult to track since the mass on the electrode will not be established - Potential chemical reaction between electrolyte and optical FBG Desired Capabilities Amorphous Capable? SEI versus Si Stress? In situ? Normal geometry? [25, 76, 184– 187] Vary conditions? Amorphous Capable? SEI versus Si Stress? In situ? Normal geometry? [188] Vary conditions? Amorphous Capable? SEI versus Si Stress? In situ? - Requires modification of surface that may change SEI formation Ref. Normal geometry? Vary conditions? [189] 38 Table 1.2 Continued Technique Pros Cons Desired Capabilities Ref. - Tags are large Fluorescence Microscopy - Can tag and track single molecules, so may be able to separate SEI and silicon - Many tags are not stable in the potential window of interest - Potential quenching of fluorophores upon incorporation into the SEI Amorphous Capable? SEI versus Si Stress? In situ? [190] Normal geometry? Vary conditions? Amorphous Capable? Full field diffraction x-ray microscopy (FFDXM) - Measure stress in crystalline materials - Can provide information on interfacial defects SEI versus Si Stress? - Sample must be at least partially crystalline In situ? Normal geometry? [191] Vary conditions? Amorphous Capable? Laser acoustic wave (LAW) - Measure Young’s modulus - Nondestructive technique - Ex situ technique where the SEI is dried fast SEI versus Si Stress? In situ? Normal geometry? Vary conditions? [192] 39 Table 1.2 Continued Technique Pros - Minimal change to the system Moiré Interferometry and Microscopy - Simple instrumentation - Resolution of 0.5 µm to nm scale in displacement - Only measures in-plane strain MOSS Nanoindentation - Proven technique for the measurement of curvature through liquid and via the backside of the electrode - Determines Young’s modulus Cons - Can’t easily distinguish between the SEI and Si stress directly - Complicated data analysis and sample preparation - Can’t easily distinguish between the SEI and Si stress directly Desired Capabilities Amorphous Capable? SEI versus Si Stress? In situ? Normal geometry? Amorphous Capable? SEI versus Si Stress? In situ? - Destructive method Amorphous Capable? Uncertainty when done under dry conditions for SEI [193– 195] Vary conditions? - Average over entire wafer and must make assumptions to get from curvature to stress - Requires open cell geometry Ref. Normal geometry? Vary conditions? [25, 39, 41, 41, 43, 121, 196– 200] SEI versus Si Stress? In situ? Normal geometry? Vary conditions? [9] 40 Table 1.2 Continued Technique Raman Scanning electrochemical microscopy (SECM) Strain-induced elasticbuckling instability for mechanical measurements (SIEBIMM) Pros - Can potentially measure stress in the SEI and Si separately and simultaneously - Spatial information on the electrochemical activity of the surface Cons Desired Capabilities - Resolution limited by the diffraction limit of light Amorphous Capable? SEI versus Si Stress? - Many components of the SEI are not Raman active or have weak Raman signals In situ? - Requires a redox probe in solution which may not be stable with all parts of the cell Amorphous Capable? - Open cell format Normal geometry? Ref. [29, 30, 32, 33, 201– 204] Vary conditions? SEI versus Si Stress? In situ? [134] Normal geometry? Vary conditions? Amorphous Capable? - Method to determine the modulus of buckling thin films - Requires good adhesion to a soft substrate that will buckle due to thin film stress SEI versus Si Stress? In situ? Normal geometry? Vary conditions? [139] 41 Table 1.2 Continued Technique Pros - Non-contact method Substrate curvature (other) - Can use a variety of measurement techniques to quantify substrate curvature then convert to stress Cons Desired Capabilities - Can’t easily distinguish between SEI and Si stress directly Amorphous Capable? - Indirect measurement of stress Normal geometry? Ref. SEI versus Si Stress? In situ? [205– 208] Vary conditions? - Expensive TEM - View of changes on electrode surface with angstrom level resolution - In situ TEM convoluted by SiN window - Difficult to do multiple test conditions - Can’t easily distinguish between SEI and Si stress directly Amorphous Capable? SEI versus Si Stress? In situ? Normal geometry? Vary conditions? [25] 42 Table 1.2 Continued Technique Pros Cons Desired Capabilities Ref. - Not possible to look through electrolyte without special equipment - Inexpensive White Light Interferometry - Readily available - Changes in indices of refraction convolute data - Can’t easily distinguish between SEI and Si stress directly Amorphous Capable? SEI versus Si Stress? In situ? [209] Normal geometry? Vary conditions? - If look at backside of wafer, obtain an average value Amorphous Capable? SEI versus Si Stress? XRD - Can show changes in stress in a material - Sample must be at least partially crystalline In situ? Normal geometry? Vary conditions? [29, 210– 212] 43 Table 1.3. Mechanical properties of SEI, Si, and Cu used in initial strain calculations [130] . Property Value ESEI 5 GPa νSEI 0.30 hSEI 20 nm ESi , SOC=0.1 90 GPa νSi 0.26 hSi 50 nm ECu 128 GPa νCu 0.36 hCu 2000 nm Figure 1.1. Performance gap in the calendar life of silicon cells. (a) Evolution of the cycle life achieved by leading manufacturers of Si- and SiOx-containing cells. Energy densities and cycle lifetimes have shown substantial progress over the last decade, quickly approaching the performance targets set by the U.S. Department of Energy (DOE). (b) Calendar life of the state-of-the-art Si-containing cells shown in panel a, which fall short from the DOE’s performance targets demonstrating a large technical gap that needs to be addressed. Calendar life data are only shown for cell formulations tested in 2020. Cycle life and calendar life are defined as the number of successive cycles or months of storage, respectively, at which the cells lose 20% of their initial capacity. 44 Figure 1.2. Li ion battery operation during charge and discharge of a full cell. Figure 1.3. Representative cyclic voltammogram of amorphous silicon thin film. Since the film was amorphous to start, it does not undergo a crystalline to amorphous transition. Coin cell half cell configuration with working electrode of 50 nm Si on 2 µm Cu current collector and counter/reference electrode of Li metal. Electrolyte of 1.2 M LiPF6 in 3:7 EC: EMC. Scan rate of 0.1 mV/s at room temperature. 45 Figure 1.4. Schematic of heterogeneous nature of SEI reproduced with permission from [8]. Used with permission of IOP Publishing, Ltd, from Advanced Model of Solid Electrolyte Interphase Electrodes in Liquid and Polymer Electrolytes, E. Peled, vol. 144, no. 8, copyright 2022. 46 A B Figure 1.5. Example of galvanostatic (constant current) cycling at a rate of C/10. 47 Figure 1.6. Mechanisms of long-term performance fade in Si-containing cells due to silicon and SEI reactivity. Schematic representation of the various issues that contribute to the poor calendar life observed in silicon-containing cells. The SEI formed on silicon undergoes continuous changes due to SEI reactivity with HF. This persistent electrolyte reactivity experienced by Si and its SEI can accelerate capacity fade, while also generating solid deposits that can block pores in the anode. After extended storage, both electrolyte consumption and pore obstruction can contribute to power fade. At a more fundamental level (inset), the SEI experiences constant morphological and compositional changes, affecting its ability to protect the Si core. Mechanical disruption of the surface layers is also possible during storage, as the particles will slowly contract due to self-discharge. The resulting exposure of Si to the electrolyte can continue to feed the hydrolytic cycle of PF6, generating additional HF that can be harmful to the cell. Additionally, soluble products of electrolyte decomposition can diffuse and react at the surface of the cathode, causing unknown consequences to cell health. These processes may manifest themselves differently depending on temperature and SOC during storage as well as electrolyte/electrode composition, and they are certainly exacerbated by the high surface areas of Si nanostructures. 48 (a) (b) Acve Material Acve Material (c) (d) Acve Material Acve Material Pore stretching during lithiaon Small pores hinder electrolyte X (e) Acve Material Stretched pores allow electrolyte through (f ) Cycling Cycling Acve Material Acve Material Acve AcveMaterial Material Figure 1.7. Possible mechanical failure modes of the solid electrolyte interphase (SEI) including (a) SEI cracking (inner and outer), (b) cracking of electrode leading to cracking of the SEI, (c) delamination of the SEI (inner and outer), (d) penetration of the SEI by a dendrite, (e) stretching of a porous SEI allowing electrolyte access to the electrode, and (f) wrinkling of the SEI when under compressive stress leading to ratcheting with further cycling. The brown represents phases composed of organic species while blue represents phases composed of inorganic species. The cartoon of the SEI is based on Peled’s mosaic model [52] and the often observed roughly two-layer model of the SEI. 49 Figure 1.8. Changes in the moiré fringe pattern as a result of changes in the sample grating. Gratings 1 and 2 are the reference and sample gratings, respectively. Stress 50 σ1=E1ε1 σ2=E2ε2 Strain Figure 1.9. Basic stress-strain curves for a high (E1 ) and low (E2 ) modulus material. The red portion of the graph indicates the linear elastic region. Figure 1.10. Example of the resulting curvature of the silicon electrode and substrate due to misfit in-plane strain in the SEI. 51 Figure 1.11. Estimate of the through-thickness strain distribution of 50 nm Si thin film and 2 µm Cu substrate due to misfit stress in SEI. 52 Figure 1.12. Strain calculations for different locations within the copper substrate and the change in strain due to substrate thickness. A) Change in strain at the back Cu surface, Si-Cu interface, and neutral plane as a function of copper thickness. B) Change in strain at the back substrate surface as a function of substrate modulus for two different substrate thicknesses. 53 Figure 1.13. Example of feedback mode SECM with steady state (left), negative feedback (middle), and positive feedback (right). The tip current decreases as it approaches an insulating surface and increases as it approaches a conductive surface. 60 10 0 40 0 Current (nA) Current (µA) 20 -20 -40 -60 -80 -10 -20 -30 -40 -50 -60 -70 1.6 1.4 1.2 1 Voltage vs. Ag/Ag+ pseudo reference -80 1.4 1.2 1 Voltage vs. Ag/Ag+ pseudo reference Figure 1.14. CV of a macro (3mm diameter) Pt disk electrode (left) and a micro disk electrode (25 µm in diameter) Pt electrode (right). The scan rate is 50 mV/sec and the reference electrode and counter electrode are a Ag wire and Pt mesh, respectively. The electrolyte is TBAPF6 in acetonitrile with ferrocene and was at room temperature. 54 Figure 1.15. Example of a probe approach curve of the positive feedback for a conductive gold substrate (black), of the negative feedback of an insulating alumina-coated surface (red) and of the mixed feedback of a partially passivated silicon thin film (blue). The dashed lines are experimental data whereas the solid lines are the fit to Equation 1.25 55 Figure 1.16. Example of a simple equivalent circuit for an electrochemical cell with Rb as the ohmic resistance. The resistance and constant phase element of the SEI are given by RSEI and CPESEI , respectively. A constant phase element describes the capacitance of a nonplanar surface. The charge transfer resistance and constant phase element of the electrode are given by Rct ) and CPEelec . Wwarburg describes diffusion into the electrode. CHAPTER 2 EXPERIMENTAL METHODS AND TECHNIQUES Parts of this chapter were adapted with permission from a published manuscript: Used with permission of IOP Publishing, Ltd, from Critical Evaluation of Potentiostatic Holds as Accelerated Predictors of Capacity Fade during Calendar Aging, M. C. Schulze, M.T.F. Rodrigues, J.D. McBrayer, et al., vol. 169, no. 5, copyright 2022. 2.1 Moiré Interferometry 2.1.1 Sample Fabrication Thin film electrodes were fabricated with thin Cu current collectors to minimize the restraint of the silicon and SEI. Photoresist was spun onto a silicon wafer and cured. The electrodes were then deposited via e-beam evaporation of 50 nm of SiO2 , 50 nm of Si, and 500 nm of Cu. The thickness of 500 nm of Cu was chosen based on the calculations from Figure 1.12. The electrodes were made into 5/8 in diameter circles using a specially machined shadow mask. Moiré 4 µm silicon sample gratings with 50% fill factor were then patterned onto the electrode backs at a thickness to minimize reflection at a 633 nm wavelength. The electrode preparation is summarized in Figure 2.1. The electrodes were then soaked in acetone overnight to release the thin electrodes from the wafer. The electrodes were handled with membrane tweezers and rinsed in successive baths of acetone and IPA. The electrodes were then glued, using silver epoxy, to stainless steel washers on the Cu side for mechanical support. The electrode grating was visible through the washer hole. The supported electrode was then cleaned using O2 plasma for 3 minutes to remove residual organics. The SiO2 layer acted as a protective layer to prevent the Si from interacting with the photoresist and to prevent Si oxidation during the O2 plasma clean. A 1% v/v HF solution was dripped onto the silicon side of the electrode and left for 4 minutes to remove the deposited SiO2 layer. The electrodes were then rinsed 57 in successive DI water baths, dried using Ar gas, then transferred into the glovebox within 20 minutes. This cleaning process led to consistent electrochemical performance of the 50 nm Si layer as demonstrated from the CV in Figure 2.2. 2.1.2 Instrument Design First, geometric moiré was explored where the reference grating was created by projecting a grating onto the surface of the sample (Figure 2.3A), but there was too much variation in the measured strain with changing sample height. Next, a moiré interferometer was built. A diagram of the house-built moiré interferometer is shown in Figure 2.3B. A 633 nm collimated laser is used to first align the sample using auto collimation. Then a 20 µm grating with 50% fill factor made from gold coated cover glass is inserted to create the reference gratings from the ±1 diffraction orders. The 0 order beam was blocked since it isn’t necessary to create the reference grating. 2.1.3 In situ Cell Design Gas generation during SEI formation can affect measurements in two ways by 1) rupturing the thin electrodes or 2) skewing strain measurements to include pressure rather than stress generated purely from electrochemical formation of the SEI. Therefore, the in situ cell design contains a gas chamber to minimize bulging of the electrode foil. A small gap was left between the lithium metal counter electrode at the silicon foil working electrode and a 2325 Celgard separator was placed over the Li metal to prevent accidental internal short circuits. The cell is shown in Figure 2.4. 2.2 Coin Cell Fabrication and Testing 2.2.1 Calendar Life Testing Two types of calendar testing were investigated in this work. The first used a voltage hold while monitoring the current decay to try and evaluate calendar aging more rapidly. The second was a specially designed protocol to try and quantify the impact of mechanical disturbance of the SEI during periodic reference performance test (RPT) cycling on how calendar life is calculated. Table 2.1 shows the electrodes used in this work. All electrodes were dried at 120 °C under dynamic vacuum overnight prior to use. For variable RPT experiments, 2032 stainless steel coin cells were fabricated as shown in Figure 2.5. The 58 cathodes are all 14 mm in diameter and the anodes are 15 mm in diameter. The separator was 2325 Celgard with a diameter of 19 mm. Two to three spacers (12.7 mm in diameter and 0.5 mm thick each) were used depending on the test. The spacers were placed on either side of the cell stack to ensure consistent contact on both sides. For EIS, three 0.5 mm thick spacers were found to be imperative for consistent electrical contact within the cell. The galvanostatic cycling was comparable with two or three spacers. 2.2.2 Voltage Hold This section describes the experimental setup for Chapter 3. 2.2.3 Materials The composition of all electrodes used in this work is detailed in Table 2.1. Electrodes were dried overnight under dynamic vacuum at 120 °C (for PVDF binder) or 150 °C (for LiPAA binder) prior to use. Most tests with lab-scale cells used Gen2 electrolyte (1.2 M LiPF6 in a 3:7 wt:wt mixture of ethylene carbonate and ethyl methyl carbonate) with 10 wt% fluoroethylene carbonate (FEC). Certain tests used Gen2 containing 2 wt% vinylene carbonate (VC), 2 wt% ethylene sulfite (ES) and 2 wt% tris(trimethylsilyl) phosphite (TMSPi); usage of Gen2 + VC/ES/TMSPi is explicitly acknowledged. Gen2 and FEC were procured from Tomiyama and Solvay, respectively. TMSPi and ES were purchased from Sigma Aldrich, while VC was acquired from TCI America. All electrolyte components were used as-received. 2.2.4 Coin Cells The 2032-format cells used 14 mm and 15 mm electrodes, a 19 mm Celgard 2325 separator, two 0.5-mm thick stainless steel spacers, and a stainless steel wave spring. In full-cells, the anode was the larger electrode; in half-cells, the Li electrode was larger. 40 µL of electrolyte was added to each cell, corresponding to at least 4x the total pore volume of electrodes and separator for all systems. The electrolyte was either Gen2 + FEC or Gen2 + VC/ES/TMSPi. The identity of electrodes used in each case is explicitly acknowledged in the text, and the detailed composition can be found in Table 2.1. LFP-based cells were formed at a rate of C/10 between 2.7 - 3.42 V for Gr cells and 2.7 - 3.35 V for siliconcontaining cells. Voltage holds were performed at either 3.35 or 3.335 V. NMC-based cells 59 were cycled at C/10 between 3.0 and 4.1 V, and were aged at open circuit after a full charge. Cycling tests and voltage holds were performed on Maccor 4100 cyclers at 30 °C or at the indicated temperature. 2.2.5 Single-Layer Pouch Cells The xx3450-format cells were prepared using 14.1 and 14.9 cm2 electrodes. An “overhang” is typically reserved for the anode, to reduce the likelihood of local anode overcharge due to electrode edge effects such as potential gradients, concentration gradients, or assembly imperfections such as electrode misalignment. Here, cells with both cathode and anode overhangs are explored to probe the effect of these excess areas on the outcome of voltage hold experiments. A total of four types of cells were prepared: i) LFP versus Gr-1 (anode overhang); ii) LFP versus Gr-1 (cathode overhang); iii) LFP versus 15% Si-Gr (anode overhang); and iv) LFP versus 15% Si-Gr (cathode overhang). Four cells of each type were assembled. Note that these cells had a N/P ratio ¡ 1, as the excess Li+ inventory allows long-term voltage holds at high SOCs to be performed (see text below for details). Celgard 2325 was used as the separator, and the electrolyte was Gen2 + 10 wt% FEC. The electrolyte volume added to each cell was equivalent to 4x the total pore volume of electrodes and separator. The formation protocol included three C/10 cycles between 2.7 – 3.42 V for Gr cells, and 2.7 – 3.35 V for Si-Gr cells; at 3.35 V the anode experiences 100 mV versus Li/Li+ , thus limiting the extent of Si expansion. Cells were degassed and resealed after the formation cycles. Voltage holds were performed at 3.335 V and 30 °C for various periods on either a Maccor 4100 or a custom-made high-precision cycler. 2.2.6 Cylindrical Cells The 18650-format cylindrical cells were acquired from LithiumWerks (formerly A123). The cell model used was Nanophosphate® APR18650M1-B, which has a graphite anode and LFP cathode matched for the cell to nominally deliver 1.1 Ah of capacity when cycled between 2.0-3.6 V. Cells were tested immediately when received or refrigerated for future use. Voltage holds were performed at a variety of potentials as specified in the text. 60 2.2.7 Variable OCV-RPT Protocol Four different electrodes (80% Si, 15% Si-Gr, 10% Si-Gr, and Gr-2) were assembled into 2032 format coin cells against NMC622. To deconvolute the effect of RPTs, all cells were cycled the same amount and rested for the same total amount of time over ∼9 months as shown in Figure 2.6. In sets of 3, each cell underwent a set of 3 cycles every 1, 2, or 4 weeks or 1, 2, or 3 months after formation cycles. The cells were cycled at a rate of C/10 between 3 and 4.1 V. All rest periods started at 4.1 V and ended with discharge to 3 V to quantify the capacity lost during rest. At the end of 8 months, the cells were cycled continuously to result in all cells ending with the same number of cycles and rest time. The cathode, NMC622, was punched into 14 mm diameter circles and the anodes were punched into 15 mm diameter circles. The separator was 2325 Celgard and the electrolyte was 40 µL gen2 (1.2 M LiPF6 in 3:7 ethylene carbonate: ethyl methyl carbonate (EC:EMC)) + 10% fluoroethylene carbonate (FEC) unless otherwise specified. The cells were cycled and stored at 30 °C. 2.3 Scanning Electrochemical Microscopy 2.3.1 Sample Fabrication Flat surfaces, relative to the tip size of the UME makes SECM more straightforward because there is less convolution of the tip current from the distance between the tip and the surface and the surface kinetics. Therefore, this work used thin films on planar wafers. There were some defects on the sample due to handling, but they were less than 0.7 µm deep, so would have minimal impact on changes in current due to changes in height, although could act as spots of high reactivity if copper was exposed. The flat surface remained so after cycling to the delithiated or lithiated state as well, so changes in height were not responsible for trends observed with aging. An example of one of the worse defects observed on silicon in the delithiated state is shown in Figure 2.7 to demonstrate the maximum change in height observed, which was always less than 0.06 L. Degenerately doped silicon wafers with the low resistivity of 0.004 to 0.04 Ohm-cm were prepared by electron beam (e-beam) evaporation of 50 nm of gold on the backside for improved contact. The front side had three layers, also e-beam evaporated, including 500 nm of Cu, 50 nm Si, and 50 nm SiO2 as shown in Figure 2.8. The samples were diced 61 into 1.75 x 1.75 cm squares and then were masked with Kapton tape on one half of the front side and the entire backside of the wafer to maintain electrical contact. Alumina was deposited using a Picosun Sunale R150 atomic layer deposition instrument at the Center for Integrated Nanotechnology (CINT). The reactor temperature was 200 °C. The first precursor was trimethylaluminum (TMA) with a pulse time of 0.1 seconds and a purge time of 6 seconds. The carrier gas and flow was N2 and 150 SCFM, respectively. The TMA temperature was 20 °C. The second precursor was water with a pulse time of 0.1 seconds and a purge time of 6 seconds. The carrier gas was also N2 , but with a flow rate of 200 SCFM. Both the TMA and water temperatures were 20 °C. There were 500 cycles resulting in an alumina thickness of ≈ 50 nm. The alumina acted as a pure negative feedback for SECM which allowed for leveling and determination of the exact location of the surface to determine L when doing PAC and SECM for the silicon half of the sample. The alumina acted as a pure negative feedback regardless of cycling as shown in Figure 2.9. The Kapton tape was then removed from the silicon side and placed on the alumina side to protect it during further processing steps. The samples were cleaned using O2 plasma for 3 minutes. The SiO2 layer acted as a protective layer to prevent oxidation of the silicon thin film electrode. The SiO2 layer was then etched using 1% v/v aqueous HF for 4 minutes. The Kapton on the alumina prevented the HF from etching the alumina. The samples were brought into the glovebox within 20 minutes of being etched to minimize regrowth of a native oxide. 2.3.2 Instrumentation A CH Instruments (CHI) 920D SECM was used for all experiments. Pt disk UMEs were also purchased from CHI and had a diameter of 25 µm. The probes were polished consecutively using 1 µm and 0.3 µm alumina paste. The probe tip diameter and approximate R g were confirmed by optical microscopy (Figure 2.10). The R g was determined to be around 2.14 from fitting probe approach curves and 2.30 from imaging. 2.3.3 Redox Probe and Electrolyte Selection Finding a suitable redox molecule for use in lithium ion batteries can be difficult. A redox probe must at the minimum be stable against lithium, stable with cell components (electrolyte, silicon, etc.), and exhibit chemical and electrochemical reversibility. Ferrocene 62 is a common choice for a facile, reversible redox probe. Despite these helpful properties, ferrocene can often be problematic for battery systems because the ferrocene redox potential ( ∼ 3.35VversusLi/Li+ ) is high relative to typical anode operating potentials. Bringing the anode up to 3.35 V versus Li/Li+ likely disrupts the SEI and may convolute any information gleaned about the passivation of the surface. SECM helps with this problem by allowing the probe tip, rather than the substrate, to polarize to ferrocene redox potentials. However, it would be ideal to be able to polarize both the UME and the substrate (the silicon thin film) without worrying about damage to the SEI. In order to do this a redox probe with a lower potential relative to lithium is required. Two other redox probes were considered including TEMPO (redox potential of 3.5 V versus Li/Li+ ) and cobaltocene (redox potential of ∼ 0.22 versus Li/Li+ ). TEMPO was not used because of the inability to get any redox peaks on a pristine, oxide-free, silicon thin film. Cobaltocene was promising because of its lower redox potential, good CVs on silicon thin films, and stability in the gen2 electrolyte. However, the solution was a mixture of reduced and oxidized cobaltocene species which convoluted SECM feedback. This may still be a good candidate for future battery research since bulk electrolysis may be able to help get all the colbaltocene in the same redox state. Ferrocene was used as the redox probe in this work and the substrate was not polarized during characterization. The probe was stable in propylene carbonate with LiClO4 as well as with silicon and cell parts as long as the silicon potential was not raised to Fc potentials. Initially SECM was attempted in gen2 electrolyte with added ferrocene to avoid introducing the SEI to a new electrolyte. However, the Pt UME surface area was found to change throughout an experiment, convoluting the steady state current. Since gen2 electrolyte is pretty volatile, a change in steady state current (Equation 1.18) is expected due to solvent evaporation and a corresponding increase in Fc concentration. According to Equation 1.18, an increase in bulk concentration of the redox probe should cause an increase in current; however, a decrease in current with time was observed. By switching to propylene carbonate with LiClO4 for SECM, the steady state current was much more consistent. When in gen2 electrolyte, the steady state current decreased monotonically by 13 % over a 4 hour experiment while in propylene carbonate, the steady state current varied randomly by about 5 % over a similar time frame. From looking at Equation 1.18, 63 there are two possible variables that could be changing leading to a decrease in steady state current: 1.) C ∗ could decrease if the Fc reacts in the electrolyte and 2.) the geometry (β and a) could change from a film depositing on the UME. The diffusion coefficient is unlikely to change significantly. Soaking experiments of Fc in gen2 with intermittent CVs did not show a significant change in the Fc concentration, but a change in the CV was observed when a Pt macro electrode was polarized to Fc redox potentials in the presence of Fc/gen2. The peak current decreased suggesting a blocking film was deposited on the surface. There was also a shift in the redox potential indicating possible reactivity with the lithium reference. A probe approach curve involves polarizing the UME and moving the probe towards the surface, so the voltage holds during probe approach curves appeared to result in a blocking film on the platinum that led to the decrease in steady state current. Switching to propylene carbonate resulted in stable CVs before and after voltage holds. An added benefit is the low vapor pressure of propylene carbonate, which decreased evaporation during experiments. 2.3.4 In situ Cell Design The cell design is shown in Figure 2.11 and has two configurations. The first is for cycling the battery in the electrolyte of interest (40µL electrolyte volume) and the second is for doing SECM (1 mL electrolyte volume). Electrolytes investigated included gen2, gen2 + 3% FEC, and 0.7 M lithium bis(oxalato)borate (LiBOB) in 3:7 EC:EMC. All chemicals were purchased from Sigma Aldrich except the gen2 electrolyte which was purchased already prepared from Tomiyama. In the cycling configuration, the electrode stack is composed of the substrate (sealed by a Kalrez o-ring), 2325 Celgard separator, 0.75 mm thick and 8mm in diameter Li metal, and a Cu rod that makes contact to the lithium. The lithium is gently scraped prior to use. Contact is made to the Cu rod by two stainless steel rods that push, what will be the reference and counter electrode lithium during imaging, into the sides of the Cu rod. The Cu rod is also secured by the top sealing plate which creates a reproducible pressure to the cell stack. Contact is made to the substrate by pressing indium wire into the back of the conductive substrate and using conductive Cu tape (both the Cu and adhesive are conductive) to connect from the indium wire out to an alligator clip. The entire cell body is made out of PEEK and the o-rings are Kalrez to prevent reactivity between the 64 o-ring and substrate. After cycling and calendar aging tests, the top sealing plate is removed along with the Cu rod, lithium, and separator. The two stainless steel contacts become the lithium counter and reference electrodes for a 3-electrode cell. Propylene carbonate (PC) with 100 mM LiClO4 and ∼ 2 mM Fc is added to the side of the electrolyte reservoir to prevent spraying directly on the substrate and disturbing the surface. Between experiments, the cell bodies were cleaned by rinsing and sonicating sequentially in DI water, acetone, and then IPA for 15 minutes each. The parts were then dried at 60 C overnight under dynamic vacuum and hot loaded into the glovebox. The metal contact parts were cleaned the same except for a shorter sonication time of 5 minutes to prevent overheating. The thickness of the o-ring was found to be important for allowing even diffusion to the electrode surface, so a #112 (thickness of 0.103 in) rather than a #012 (thickness of 0.07 in) Kalrez o-ring was used to seal against the silicon. The effect of the o-ring thickness can be observed in a CV of a model system (TBAPF6 and Fc in acetonitrile with a platinum coated wafer as the working electrode in the SECM setup) in Figure 2.12. The flattening of the redox peaks is observed when the gap between the PEEK cell body and substrate is too small (i.e., the o-ring is too thin), suggesting a difference in reactivity across the surface. The peak splitting is also greater which means lower electrochemical reversibility. Another important variable for good cell cycling was sufficiently low electrolyte volume. When the imaging configuration (1 mL of electrolyte) setup was used with a seal on top and cycled, high voltage plateaus were observed at C-rates of C/3 or slower. An example of this plateau behavior is shown in Figure 2.13. A large current of 1C was required to over come the high voltage plateau. This is not optimal for SEI formation, most batteries are formed at C/10 or C/20 rates. By moving to the cycling versus imaging configuration setup with low and high electrolyte volumes, respectively, lower currents could be used for forming the SEI. However, the downside to this approach is that although the cycling situation is more comparable to a normal coin cell battery, it does require pulling off the separator before imaging which could possibly damage the SEI, which could make the surface look more reactive when measured by SECM than it actually would be normally. Unless specified otherwise, SECM cells were cycled at C/5 in this work as a compromise 65 between a rate that is sufficiently slow for good SEI formation and fast enough for high throughput. The cells were cycled within an argon glovebox with less than 1 ppm water and oxygen and were sealed under argon. The data from the SECM is from multiple cells aging together in tandem, rather than a single cell that was disassembled and investigated at different time points. The reason for this is that the addition of the ferrocene to perform SECM studies contaminated the cell, resulting in anomalous cycling behavior after an SECM measurement was taken (Figure 2.14). To avoid this, several cells were used, with each cell only being used once per SECM experiment. Even after rinsing with electrolyte several times, the cell undergoes more rapid self discharge during rest when trace Fc is present. Both curves in Figure 2.14 show the change in voltage during rest. The blue curve is a cell that cycled three times, lithiated down to 100 mV before starting the rest. The orange curve is a cell that cycled three times ending in the delithiated state at 1.5 V. Propylene carbonate with ferrocene and LiClO4 was added and the surface was characterized. The cell was then rinsed several times with gen2 electrolyte, then 40 µL of gen2 electrolyte was added and the cell stack was replaced, the cell was resealed and then lithiated down to 100 mV where it then entered the rest shown. The rapid change in voltage could be due to the remaining trace presence of Fc or because of damage to the SEI when removing the cell stack. Adding a voltage hold at the end of delithiation and lithiation helped prevent rapid voltage change due to hysteresis in adding/removing lithium for the silicon, but this didn’t fully solve the rapid change in potential when trying to characterize the same cell at different aging states. 2.3.5 Data Analysis All SECM images in this work are given in terms of κ (Equation 1.21). The diffusion coefficient of Fc in 100 mM LiClO4 in PC was found to be 3.58 e-10 m2 /s. This was determined by the slope of a steady state current versus Fc concentration calibration curve (Figure 2.15). From Equation 1.18, the slope is given by 4nFDaβ. All values are known, so the diffusion coefficient, D, can be determined. The diffusion coefficient can then be used to determine k f from κ using Equation 1.21. The location of the surface was determined from probe approach curves on the alumina half of the sample which acts as a pure negative feedback reference. The probe approach 66 curves were fit assuming pure negative feedback (κ assigned a value of 10e-6) to determine L and define a coordinate system for each experiment. The sample was leveled using three points on the alumina. The surface position was then calculated for the entire substrate which allowed for the calculation of κ for each point on an SECM scan because L was known at every point. Because of error in the fit of the negative feedback curve, a systematic offset of 2 µm was required to define the position of the actual surface. If L and itip,N are known, then κ can be calculated from Equations 1.16 - 1.25 which removes the effect of differing L values between cells. 2.4 Calorimetry All microcalorimetry experiments were carried out isothermally at 30 °C in a TAM IV thermostat 2 channel calorimeter from TA Instruments. Before each measurement, a baseline was created by adding 4 dummy cells (containing only the stainless steel components: case, cap, spacers, and wave spring) to a 20 mL stainless steel ampoule. Once the signal stabilized over a few days, the dummy cells were removed and real cells were added. The system was allowed to equilibrate again before taking the new heat generation value. Once stable, the signal was averaged and the baseline average was subtracted to give the total heat generation rate. Assuming a constant rate, the heat loss could then be correlated to the capacity loss. To study calendar aging, 4 LFP - 80% Si cells were fabricated with 40 µL gen2 electrolyte and underwent three formation cycles between 3.35 V and 2.7 V. Four cells were used to increase the signal measured by the calorimeter while the cells were at rest. The cells were built with 3, 0.5 mm thick stainless steel spacers to improve internal contact for EIS measurements. The cells were taped to prevent shorting and then placed in the 20 mL stainless steel ampoule and the “post formation” data point was acquired. The cells were then stored for approximately 3 weeks, in the charged state at 3.35 V. After 3 weeks, the “pre-RPT” data point was obtained. The cells were cycled three times and put back into the calorimeter for the “post-RPT” data point. EIS measurements were taken prior to each calorimetry measurement. 67 Table 2.1. Electrodes used in this work provided by Argonne National Laboratories. Graphite 1 (Gr-1) Graphite 2 (Gr-2) 91.83 wt% Superior Graphite SLC1506T 91.83 wt% Superior Graphite SLC1520P 2 wt% C45 carbon additive (Timcal) 2 wt% C45 carbon additive (Timcal) 0.17 %wt oxalic acid 0.17 %wt oxalic acid 6 wt% PVDF binder (KF-9300, Kureha) 6 wt% PVDF binder (KF-9300, Kureha) 37.4% electrode porosity 35.6% electrode porosity 47-µm-thick composite coating 6.49 mg/cm2 and 2.1 mAh/cm2 45-µm-thick composite coating 15% Si-Graphite (15% Si-Gr) 73 wt% Hitachi MagE3 graphite 15 wt% 200 nm Si (Paraclete Energy) 2 wt% C45 carbon additive (Timcal) 10 wt% LiPAA binder (Sigma-Aldrich) 48% electrode porosity 28-µm-thick composite coating 2.97 mg/cm2 and 2.45 mAh/cm2 (50 mV cutoff) 80% Silicon 80 wt% 200 nm Si (Paraclete Energy) 10 wt% C45 carbon additive (Timcal) 10 wt% LiPAA binder (Sigma-Aldrich) 47.3% electrode porosity 10-µm-thick composite coating 1.10 mg/cm2 and 1.5 mAh/cm2 (100 mV cutoff) 6.28 mg/cm2 and 2.0 mAh/cm2 10% Si-Graphite (10% Si-Gr) 78 wt% Hitachi MagE3 graphite 10 wt% 50-70 nm Si (Nano&Amor) 2 wt % C45 carbon additive (Timcal) 10 wt % LiPAA binder (Sigma-Aldrich) 46% electrode porosity 33-µm-thick composite coating 3.63 mg/cm2 and 1.56 mAh/cm2 LFP 90 wt% Johnson Matthey LiFePO4 5 wt% C45 carbon additive (Timcal) 5 wt% PVDF binder (5130, Solvay) 38.8% electrode porosity 98-µm-thick composite coating 19.70 mg/cm2 and 2.66 mAh/cm2 68 Table 2.1 Continued NMC622 90 wt% LiNi0 .6Mn0 .2Co0 .2 (Targray) 5 wt% C45 carbon additive (Timcal) 5 wt% PVDF binder (5130, Solvay) 37.1% electrode porosity 58-µm-thick composite coating 9.78 mg/cm2 and 1.58 mAh/cm2 (3 – 4.3 V) 69 A B C Figure 2.1. Moiré sample preparation. A) Electrode preparation for in situ moiré interferometry experiments. First AZ 2035 photoresist was spun onto a wafer. Then a shadow mask (shown in B) was placed on the wafer to create 0.625” circles while 50 nm SiO2 , 50 nm Si, and 500 nm Cu were e-beam evaporated. Then, 500 by 500 µm arrays of Si gratings with a thickness of 62 nm and a periodicity of either 4 and 5 µm were patterned by photolithography on the Cu side of the electrodes (C). The wafers were soaked in acetone overnight to release the foils. They were then attached to washers with silver epoxy for mechanical support. The silicon side was cleaned with O2 plasma then the SiO2 was removed with 1% v/v HF. 70 Figure 2.2. Example of consistency of CVs in a stainless steel coin cell. The working electrode is 50 nm Si on 500 nm Cu foils after removing the protective 50 nm SiO2 layer. The graph shows three replicates. The scan rate is 0.1 mV/s, the electrolyte is gen2. The counter and reference electrode is lithium metal. 71 A B Figure 2.3. Configuration of house-made A) moiré microscope and B) moiré interferometer. In both cases a white light source is used for plain imaging and can be replaced with a camera for auto collimation for sample alignment. When aligning, the grating is removed. For A) the grating is created by projecting a grating onto the sample surface. The 20 µm reference grating was illuminated with light collimated from a fiber-coupled high-power blue LED at 460 nm wavelength. For B) the reference grating is formed from the interference of the ±1 diffraction orders in the plane of the sample. The ±1 diffraction orders were formed from the collimated 633nm laser passing through a 20 µm grating. The reference grating period of 20 µm was demagnified by the ratio of the focal lengths of the objective lens (35 mm) and reference grating lens (80 mm). The illumination period was further multiplied by 0.5 due to rejecting the zero-order diffraction beam, resulting in a reference periodicity of 4.38 µm in the plane of the sample. 72 Figure 2.4. Cell design for in situ moiré. RE = reference electrode, WE = working electrode, and CE = counter electrode. Figure 2.5. Example assembly of a 2032 stainless steel coin cell. From top to bottom: stainless steel cathode cap, stainless steel wave spring, 1-2 stainless steel disc spacers, cathode, separator, anode, 1 stainless steel disc spacer, and stainless steel anode case. 73 Figure 2.6. Protocol for variable OCV-RPT tests to determine the impact of disturbing the SEI with cycling intermittently during rest. 10 µm Figure 2.7. Example of worst case height change over cycled and delithiated silicon thin film. 74 Figure 2.8. Sample preparation for SECM. A degenerately doped silicon wafer is coated with 50 nm of gold on the back and then 500 nm Cu, 50 nm Si, and 50 nm SiO2 . The wafer was then diced into 1.75 cm squares. Kapton tape was used to mask half the sample. Alumina (50 nm) was then deposited using ALD. The Kapton tape preventing alumina deposition on the SiO2 surface was removed. A new piece of Kapton was placed on the alumina half of the sample to prevent HF etching. The SiO2 was then removed using a 1% v/v aqueous HF solution. 75 Figure 2.9. Example of consistent alumina negative feedback probe approach curve before and after cycling. Figure 2.10. Optical image of UME R g . 76 Top-sealing plate Ultra micro disk electrode Stainless Steel PEEK cell body Figure 2.11. Cell design for SECM. The left shows the cycling configuration where 40 µL of the electrolyte is used during cycling. The right shows the conversion to an imaging format where 1 mL of 100mM LiClO4 propylene carbonate with ∼ 2 mM ferrocene. Figure 2.12. CV showing the importance of having even diffusion to the entire substrate surface. By thickening the o-ring thickness from 0.07 in (#012) to 0.103 in (#112), the cell design was improved and behaved as expected in the model system of TBAPF6 with Fc in ACN. The scan rate is 50 mV/s and the working electrode is a Pt coated wafer, the pseudo reference is Ag wire, and the counter electrode is a Pt mesh. 77 Figure 2.13. Example of high-voltage plateau that occurs in cell setup with electrolyte volumes greater than ∼100 µL of gen2 electrolyte. The only way to prevent this plateau is to use a small volume of electrolyte or to cycle the cell at high current densities. The former was selected for this work to try and form a more relevant SEI. Figure 2.14. Example of self discharge with rapid voltage change during rest when trace Fc remains in the cell. This motivates the use of individual cells for each potential and rest condition. The blue curve is a cell that cycled three times, lithiated down to 100 mV before starting the rest. The orange curve is a cell that cycled three times ending in the delithiated state at 1.5 V. Propylene carbonate with ferrocene and LiClO4 was added, and the surface was characterized. The cell was then rinsed several times with gen2 electrolyte, then 40 µL of gen2 electrolyte was added and the cell stack was replaced. The cell was resealed and then lithiated down to 100 mV where it then entered the rest shown. 78 Figure 2.15. Determination of the diffusion coefficient of ferrocene in 100 mM LiClO4 in PC by plotting steady state current for a serial dilution of ferrocene. The diffusion coefficient can be calculated from the slope using Equation 1.18. CHAPTER 3 CRITICAL EVALUATION OF POTENTIOSTATIC HOLDS AS ACCELERATED PREDICTORS OF CAPACITY FADE DURING CALENDAR AGING This chapter was adapted with permission from a published manuscript: Used with permission of IOP Publishing, Ltd, from Critical Evaluation of Potentiostatic Holds as Accelerated Predictors of Capacity Fade during Calendar Aging, M. C. Schulze, M.T.F. Rodrigues, J.D. McBrayer, et al., vol. 169, no. 5, copyright 2022. Li-ion batteries (LIBs) are among the most impactful inventions of the 20th century [213]. After revolutionizing the portable electronics industry, LIBs are now key in enabling a wider adoption of electric vehicles (EVs) and the transition towards a greener electric grid. The economic feasibility of both these applications relies heavily on the durability of the battery pack, as the upfront costs of EV ownership and infrastructure investments are only slowly recovered over time [214, 215]. The longevity of LIBs is constantly challenged by the metastable nature of the electrochemical processes in the cells. Graphite electrodes operate at potentials beyond the limits of thermodynamic stability of the electrolyte, with the solid electrolyte interphase (SEI) serving as a kinetic barrier to uncontrolled reduction reactions [216]. As effective as the SEI can be in enabling LIB operation, it is not perfect, allowing side reactions to continue at slow but persistent rates. Hence, Li-ion batteries will always lose some amount of charge when at rest, even if they are never actively charged or discharged. The consequences of these time-dependent degradation processes are known as calendar aging, and the time needed for such processes to consume 20% of the initial cell capacity is called calendar life [100]. Losses of Li+ inventory due to calendar aging are typically greater during conditions that accelerate the kinetics of parasitic reactions at the SEI, such as elevated tempera- 80 tures (which increase reaction rate constants) and at lower effective anode potentials (that increase overpotentials for electrolyte reduction) [98, 100, 217], Constraining batteries to operate at low temperatures and limited full-cell states-of-charge (SOCs) would naturally delay the course of time-dependent aging, but these are not always practical options, nor are they permanent solutions to the calendar aging problem. As calendar aging effects on Li+ inventory are intimately linked with the SEI, they can be mitigated by employing electrode and electrolyte chemistries that yield more robust SEI layers [218]. Although materials discovery is always a complex and lengthy endeavor, new battery chemistries can be especially difficult to develop when the mere evaluation of any formulation change requires long-term experimentation. Since calendar aging is intrinsically slow, studies typically alternate between long-term storage (∼1 month) and brief reference performance tests (RPTs), which can involve cycling at slow rates and impedance measurements that are used to quantify the time-dependent performance degradation [219]. Thus, capturing meaningful trends from these tests requires resource-intensive studies that can extend for many months or years. A common approach to accelerate these studies is to maintain cells at higher temperatures where the effects of aging are accelerated. However, it is always a challenge to ensure that trends observed at elevated temperatures remain valid at the milder operating conditions that batteries typically experience, [220] and even these accelerated tests still require many months to be completed [100, 217]. The lengthy iterative process needed to improve calendar life is especially problematic for new technologies, as is the case for silicon anodes. Silicon is much more reactive towards the electrolyte and other cell components than traditional graphite electrodes, [105, 106, 221] which can exacerbate problems with calendar aging. Indeed, recent data from companies working at developing Si-rich anodes suggests that high-energy Si cells (> 300 Wh/kg) tend to present particularly low calendar lifetimes, suggesting that this is the main technical barrier preventing near-term commercialization of LIBs with high Si content [222]. The U.S. Department of Energy (DOE) has recognized this problem as a strategic challenge and has allocated resources to unravel the underlying mechanisms and identify strategies for their mitigation. A core point that could benefit such explorations, both by academic groups and by manufacturers, is the development of experimental approaches 81 that can expedite calendar aging experiments to accelerate the innovation process. In the present work, we discuss the merits and limitations of potentiostatic holds as one such expedient method. One characteristic of traditional calendar aging studies is that aging and the measurement of its consequences exist as separate steps. Considering that a typical aging experiment involves 1 month of inactive storage and less than 3 days of active RPT measurement, less than 10% of the total time of these tests are invested in acquiring information that will be used to evaluate cell behavior. In other words, a full month of testing will yield a single data point. One fundamental way to decrease the total test time would be to simultaneously age the cell and measure its capacity loss, converting the entire experiment into an active process. Moreover, assuming that such measurements could be made with sufficient accuracy, their higher time resolution could make general aging trends apparent much faster than in traditional months-long experiments (Figure 3.1). Potentiostatic holds [102,223–225] (also known as voltage holds or float tests) [226–228] could, in principle, provide this type of information. In this technique, full-cell voltage is forced to remain constant using an external source while the current needed to maintain the voltage is recorded. It is generally hypothesized that, when certain conditions are met, the measured current can be correlated with the instantaneous rate of parasitic electron exchanges occurring inside the cell, thus providing a picture of the time-dependent trends of these reactions. The present study compiles extensive analyses on the use of the voltage hold method as an expedient alternative to RPT-based aging studies. We investigate the merits and limitations of the technique, and its applicability to the quantitative, semiquantitative and qualitative description of calendar aging, using commercial 18650-format cells, singlelayer pouch cells, and coin cells. For the reader’s benefit, best practices for applying this technique to accelerate the identification of electrodes and electrolytes with enhanced long-term stability are summarized in the final section of this article. We expect this initial exploration to provide the battery community with useful tools to expedite the development of battery technology. For the interested reader, additional detailed discussion is provided in the Appendix. 82 3.1 3.1.1 Results and Discussion Description of the Voltage Hold Technique for Studying Calendar Aging There are numerous reports of voltage hold methods in the literature, [102,224,225,227– 230] and while they vary in the exact electrochemical protocol used, they have common elements. Figure 3.2A shows an example potential profile of a voltage hold protocol that can be broken down into the three main parts shown in Figure 3.1B. The formation is typically a series of slow cycles used to form the SEI (formation) and to measure the beginning-of-life electrochemical behavior and performance of the cell. The voltage hold cycle that follows typically charges the cell up to a hold potential (Vhold ), holds at that potential for some time, then discharges the cell. In some cases, an RPT comprising a few cycles is run after the hold cycle to help gather more diagnostic electrochemical information. The basis of using such a voltage hold experiment to learn about calendar aging of the cell centers on analyzing the current response measured during the voltage hold, an example of which is shown in Figure 3.2B. The current response comprises contributions from both reversible (Irev , charging) and irreversible (Iirrev , parasitic reactions) processes. The reversible processes arise from continuous charging of the cell as the current decreases, as depolarization makes additional capacity available at Vhold . Over time, these reversible processes should relax to negligible levels as the cell equilibrates to a constant SOC, and the irreversible processes should start to dominate the current response [225]. Thus, the current response of a sufficiently long voltage hold could ideally be used to measure the rate of irreversible parasitic reactions in a cell at a constant SOC. Integration of the current response yields the capacity exchanged during the voltage hold (Qhold ) as shown in Figure 3.2C. Here, the early rise in Qhold is primarily from continued reversible charging (Qrev ) of the cell, with the irreversible capacity (Qirrev ) represented by the smaller slope later in the hold. Qhold can thus be represented by the following time dependent function: Qhold (t) = Qirrev (t) + Qrev (t) (3.1) In principle, data curves like that shown in Figure 3.2C could describe the time depen- 83 dence of side reactions in the cell. If such measurements are sufficiently accurate, and if losses of Li+ inventory occur at much higher rates than other aging modes, these tests could hypothetically be used to infer the functional form of Qirrev (t) and extrapolate the future capacity loss of the cell (Figure 3.2D). While this type of extrapolation oversimplifies the complex processes involved in calendar aging, it could potentially serve as an accelerated method for predicting calendar aging behavior of a cell without using the lengthy RPT-based experiments depicted in Figure 3.1A. Additionally, the time resolution with which the rate of parasitic processes could be described could be useful in supporting the development of electrochemical models able to capture the interplay of the slow processes responsible for calendar aging. These potential benefits have motivated our detailed exploration of this technique. As we demonstrate below, while these optimistic expectations were not fulfilled, the technique still proved useful for the study of calendar aging. The calendar aging of batteries occurs by any number of mechanisms, including the irreversible loss of Li+ inventory (Qloss ) to the anode SEI, impedance rise due to buildup of SEI at the anode, oxygen loss and rock-salt formation in the cathode, [231, 232] side reactions at the cathode surface, or excessive electrolyte consumption (also referred to as cell dry-out) [98, 222]. All these mechanisms are fundamentally caused by parasitic processes between the electrolyte and electrodes, so measuring the time dependency of those parasitic reactions provides information about their individual contributions to calendar aging [17, 102]. Because the loss of Li+ to the SEI is the main mechanism of capacity fade during calendar aging of commercial cells, [94] our work emphasizes this particular aging mode. Our examples also focus specifically on voltage holds of graphite and silicon anodes, due to their ubiquity and high chemical reactivity, [105, 106] respectively. Both graphite and silicon form an SEI and operate at low potentials outside of the electrolyte stability window which makes them susceptible to calendar aging. However, properly passivated graphite typically has good calendar life while the innate reactivity of silicon leads to rapid degradation over time. For this type of methodology to be valid, the voltage hold must be run under conditions where Qrev (t) saturates into a constant value (∂Qrev (t)/∂t = Irev = 0), so the measured Iirrev (∂Qirrev (t)/∂t) corresponds to the rate of parasitic reactions at the anode; that is: 84 ∂Qirrev (t) ∂Qhold (t) = ∂t ∂t (3.2) An essential requirement is that the SOC of the anode must remain fixed so that the measured current corresponds only to parasitic current, rather than also comprising reversible contributions from a changing anode SOC. Continuous lithiation of the anode would cause the measured current to be larger than Iirrev , while a decrease in anode SOC over time would cause the current to be smaller than Iirrev (see discussion in the Appendix). A constant SOC can be maintained in a 3-electrode cell, where the voltage of the anode can be independently held constant versus a stable reference electrode. However, during a voltage hold of a 2-electrode cell, the counter electrode only serves as a stable reference electrode if its potential is relatively constant (insensitive) during the entire voltage hold. While a lithium metal foil used in the standard “half-cell” configuration could possibly serve as such a stable counter and reference electrode for a voltage hold, continuous chemical reactions between the Li metal and the electrolyte can impart complicating factors to the results of the experiment [233]. Instead, a cathode such as lithium iron phosphate (LiFePO4 , LFP) is an ideal counter/reference electrode for a voltage hold because it delivers most of its capacity at a potential near ∼3.45 V versus Li/Li+ . Figure 3.3 shows why a counter electrode with a “flat” voltage profile allows the SOC of a hypothetical anode (Si, in this example) to remain constant during a voltage hold, assuming the cell is fully relaxed; the effect of relaxation on voltage holds is discussed in detail below. In these examples, it is important to understand that irreversible capacity loss due to Li+ inventory consumption by side reactions at the anode causes its voltage profile to capacity-shift relative to the cathode’s voltage profile [234]. This capacity-shifting is a consequence of a changing cathode SOC (as a Li+ and corresponding electron are extracted from the cathode and transferred to the anode) while the anode SOC remains the same (due to the Li+ and electron being consumed by the parasitic side reaction). Using an LFP cathode as an example of an ideal cell setup for voltage holds, Figure 3.3A shows the simulated voltage profiles of an Si-LFP full-cell and how they shift due to irreversible Li+ inventory loss during a simulated voltage hold. In this ideal cell setup, the Si anode capacity is just 60% of the available Li-inventory supplied by the LFP cathode. Thus, a hypothetical Qirrev = 20% during a simulated voltage hold at 3.35 V is fully accommodated 85 by the excess Li+ inventory supplied by the LFP. Such configurations with N/P ratio < 1 have been used in the past for diagnostic purposes [235, 236]. When the cell potential is held at a constant 3.35 V (Figure 3.3B), the Si anode potentials remain nearly unchanged before and after the voltage hold (Figure 3.3C), despite 20% of the LFP’s Li+ inventory being consumed in parasitic side reactions at the anode. This is a consequence of the fact that additional LFP capacity can be accessed with negligible variation in cathode potential. Thus, as the cell voltage is held constant, the anode potential will also remain constant as electrons flow from the cathode to the anode to compensate for the side reactions. This implies that any capacity exchanged through the external circuit during the voltage hold is due exclusively to those irreversible reactions, and thus the technique could potentially be used as a predictor of calendar aging behavior. To highlight the importance of the voltage profile of the counter electrode, Figure 3.3D shows the simulated voltage profiles of a Si-NMC811 full-cell and how they shift due to a 20% irreversible Li+ inventory loss at the Si anode during a simulated voltage hold. In contrast to the Si-LFP full-cell, a Si-NMC full-cell is a nonideal setup for voltage holds due to the “sloped” voltage profile of the NMC counter electrode, even if the Si anode capacity is just 60% of the available Li+ inventory supplied by the NMC cathode. To better understand this, Figure 3.3E shows how holding the cell potential at a constant 3.6 V corresponds to the Si anode potentials in Figure 3.3F, which indicates a rise in the Si anode’s potentials (a decrease in the Si SOC, QSi,∆SOC ) during the voltage hold. In contrast to LFP, accessing additional NMC capacity requires an increase in cathode potential, and it becomes impossible to maintain both the cell voltage and the anode potential invariant. This implies that capacity exchanged during a voltage hold in a Si-NMC cell is a convolution of the irreversible reactions responsible for calendar aging and the change in SOC of the Si anode: Qirrev = Qhold –QSi,∆SOC (3.3) Thus, unless the value of QSi,∆SOC can be determined by some independent methods, a voltage hold in a Si-NMC cell will underestimate its instantaneous rate of side reactions. This is a consequence of the “sloped” voltage profile of NMC cathodes and applies to 86 any counter electrode with a “sloped” profile. This also explains why an ideal Si-LFP cell undergoing a voltage hold can underestimate calendar aging rates if Qirrev is large enough that Li+ inventory supplied by the LFP is fully exhausted and its voltage profile becomes steeply sloped as it polarizes to high potentials (see profile in Figure 3.3A, for example). A detailed analysis of how test conditions affect the correspondence between measured and actual parasitic currents is provided in the Appendix. Even if a voltage hold is run on a cell that meets the requirements for keeping the anode at a fixed SOC, there are additional assumptions that must be verified to ensure that the measured current accurately corresponds to the instantaneous rate of parasitic reactions. First, the reversible reactions must relax quickly during the voltage hold, so that the measured currents are dominated by the parasitic reactions. Importantly, this means that the timeframe of SOC-changing reactions must be known so that a voltage hold can be run long enough that those reactions become negligible. Second, the parasitic reactions at the anode responsible for calendar aging must be much more prominent than other electrochemical processes that can generate current during a voltage hold. Examples of other processes that could contribute to the measured current include reversible selfdischarge (via an electrical leakage current or a redox shuttle contaminant) or lithiation of an overhanging portion of an anode that is typical in full-cell configurations (not strictly reversible or irreversible). Similarly, the rate of reactions at the anode SEI must be much larger than that of electrolyte oxidation at the cathode. Finally, the instrumentation used to measure the lower currents during a voltage hold must be sufficiently sensitive to accurately measure the parasitic reactions, which can be diminishingly small. We explore the validity of these assumptions in the following sections. 3.1.2 Influence of Current Relaxation and Reversible Capacity on the Measured Aging Rate To fully understand how the relaxation of reversible currents (Irev ) during a voltage hold and the corresponding reversible capacity (Qrev ) impact calendar aging studies, we ran a series of experiments using commercial 18650 cells with graphite (Gr) anodes and LFP cathodes (details in Chapter 2). The use of commercially available cells affords us high confidence in reproducible behavior from cell to cell, while the LFP cathodes ensure that voltage holds in these 2-electrode cells keeps the Gr SOC as invariant as possible. 87 As an initial assessment of the relaxation of Irev during a voltage hold, we performed the following experiments. Two sets of 3 cells each underwent the same initial set of three cycles (cycles 1, 2, and 3) at a C/10 rate between 2.0-3.6 V, followed by charging each cell up to a hold potential (Vhold = 3.35 V). One set was then allowed to age at open circuit voltage (OCV) for 30-, 60-, or 90-days (one cell for each aging time), while the cells in the other set were held at Vhold for 30-, 60-, or 90-days. Each cell then completed a RPT of two full cycles (cycles 5 and 6) at a C/10 rate after the aging period. The discharge capacities before and after the aging period can then be used to calculate Qloss and Qrev for each cell: ( Q(cycle3discharge) –Q(cycle6discharge) ) Q(cycle3discharge) (3.4) ( Qdischargeimmediatelya f terhold –Qchargeimmediatelybe f orehold ) Qcycle3discharge (3.5) Qloss = Qrev = The Qcycle3discharge represents the nominal cell capacity at its beginning-of-life before the aging period, while the Qcycle6discharge represents the cell capacity after the aging period. The difference between them is normalized relative to the cell capacity at its defined beginningof-life (Qcycle3discharge ), so Qloss (and related capacities) can be expressed as a percentage of nominal cell capacity, as they are throughout the remainder of the text. Cycle 6, the final cycle of the reference performance test, was selected to give the most accurate estimate of true loss to minimize any residual rebound of the capacity after the voltage hold or OCV rest. The values for Qloss (calculated using Equation 3.4) are plotted in Figure 3.4A and show that the cells that underwent the voltage holds exhibited larger Qloss at all aging times than those aged at OCV. We attribute this accelerated aging rate of the voltage hold cells primarily to their higher SOC during aging (if Qrev values during the voltage hold are not negligible then there will be an increase in cell SOC) compared to the OCV aged cells, a phenomenon commonly observed in many battery systems [94,95]. The differential capacity curves in Figure 3.4B demonstrate that the voltage hold cells have lithiated to a higher SOC phase in the Gr (dQ/dV peak indicated by *), reaching anode potentials that were ∼ 30 mV lower than experienced by the OCV aged cells despite beginning their aging step at the same cell voltage. The aging potential of 3.35 V was chosen because the cell 88 exhibits a low dQ/dV value at that potential, which was thought to minimize Qrev during the voltage hold. However, depolarization of the cell proved significant enough that the Qrev was far from negligible (Qrev = ∼ 21%). This suggests that choosing an appropriate Vhold depending on the anode chemistry is important to minimizing Qrev . Considering that Qrev is not negligible during a voltage hold, it is important to understand if a voltage hold is long enough for Qrev to reach a maximum, such that Irev becomes negligible compared to the Iirrev that is responsible for calendar aging. One simple way to do this is to track values of Qrev for voltage holds of different lengths. Figure 3.5 shows Qrev values for the commercial 18650 Gr-LFP cells and several other types of cells that underwent voltage holds for varying lengths of time. Figure 3.5A shows Qrev values of graphite-containing cells that underwent voltage holds between 7.5 and 90 days. Each graphite-containing cell shows little dependency or increase between the shortest and longest hold times, indicating that Qrev has quickly reached a maximum and Irev already becomes negligible for hold times of 7.5 days for the 2032 half-cells and 30 days for the same 18650 LFP cells tested in Figure 3.4. Note the graphite-Li half-cells are 2032 coin cells and exhibit some variation in Qrev (on the order of 1-5%) when tested for different durations. In contrast, the commercial graphite-LFP 18650 cells have rapid and stable relaxation of reversible processes and exhibit a consistent Qrev = ∼21%, with less than 0.1% variation with test time, showing that the format and quality of assembly of the cell used can have an impact on the consistency of results from voltage hold experiments. All in all, both datasets indicate that Qrev saturates rapidly in graphite cells. Figure 3.5B shows Qrev values of silicon-containing 2032 coin cells that underwent voltage holds between 7.5 and 90 days. In contrast to the graphite-only cells, the siliconcontaining cells show a general trend of increasing Qrev values (rather than oscillating around a single value) with longer voltage hold times, indicating that Irev is not negligible even after 90 days. For the Si-graphite electrode (15% Si), the anode gains ∼ 35% of its capacity before the hold after 1 week of testing and gains another ∼ 25% in the following months. This is likely due to the slow kinetics and large voltage hysteresis of the lithiation and delithiation of silicon, which allows continued reversible lithiation of the silicon for long time periods during a voltage hold. This idea is supported by the larger Qrev values recorded for the silicon-only containing cell in Figure 3.5B compared to the mixed silicon- 89 graphite containing cell. These time-dependent voltage hold experiments highlight how the relaxation of the reversible processes (Irev) is dependent on the identity of the electrode being tested. Specifically, active materials exhibiting a large voltage hysteresis (such as silicon) have slow relaxation of reversible lithiation processes, making it challenging to deconvolute the Qirrev component from Qrev during the voltage hold process to forecast calendar aging trends. However, the consistent Qrev values of the graphite cells suggest that they can be promptly investigated using potentiostatic holds. Another experiment that can be used to assess the timescale of the relaxation of the reversible processes during a voltage hold more easily than the multiple experiments described above is an “inverse polarization” test, which is described in detail in the Appendix. The inverse polarization experiment simply runs a voltage hold at a potential where delithiation of the anode (rather than lithiation) is the dominant reversible process. This is achieved by first lithiating the anode, resting the cell at open circuit to allow the relaxation of any voltage hysteresis, then delithiating the anode either by a predetermined capacity or until a target potential is reached, at which point a voltage hold is started. The irreversible processes responsible for calendar aging are electrochemical reduction reactions (Iirrev < 0 in the sign convention of potentiostats) while delithiation is an electrochemical oxidation reaction (Idelit > 0). The current response during an inverse polarization voltage hold is the sum of these currents: Iinversepolarization (t) = Iirrev (t) + Idelit (t) (3.6) Thus, during an inverse polarization experiment, Iinverse polarization will initially be positive as Idelit dominates the current response. As the reversible delithiation processes relax and the irreversible processes start to dominate, Iinverse polarization will switch signs to become negative (see Appendix). When the current response is zero, the reversible and irreversible processes are equal in magnitude, and continued holding of the voltage will drive Iinverse polarization to a minimum negative value before it starts to increase back towards zero, as the irreversible processes slowly self-passivate. The time at which Iinverse polarization reaches a minimum is when Idelit ∼ 0, and the measured current is approximately equal to Iirrev ; this time serves as a quantitative guide for how long voltage 90 holds must be before they start to yield the desired information on irreversible processes. Examples of typical current responses of inverse polarization experiments are shown in the Appendix. As expected, based on the discussion above, graphite electrodes quickly relax (see Appendix), while silicon requires hundreds of hours for the residual lithiation to subside. Thus, voltage hold experiments using silicon-containing electrodes will require much longer times than ones using graphite. It is important to note that the inverse polarization experiment gives information about the timescale of relaxation of reversible delithiation processes and may not necessarily be representative of the relaxation of reversible lithiation processes during a normal voltage-hold experiment. 3.1.3 How Hardware Sensitivity and the Prominence of Reduction Reactions Can Affect Voltage Holds The currents measured during a potentiostatic hold experiment can carry contributions from many processes within the cell that involve an electron exchange. Thus, the technique will only be informative of capacity fade during calendar aging if the rate of parasitic reactions at the SEI is much larger than that of other processes. As discussed in the previous section, continued charging of the cell as it depolarizes is an obvious source of interference, but one that can be overcome by sufficiently long e xperiments. Another example is oxidation side reactions, which can transfer electrons to the cathode, [237] a phenomenon that is typically expected during cathode break-in, though comes at the expense of cell health if it continues beyond the initial cycles. This electron transfer effectively increases the Li+ inventory of the cathode, causing the perceived capacity fade due to reactions at the SEI to be underestimated. Li-ion batteries stored at open circuit will lose capacity over time, [94, 98] indicating that the average rate of reduction reactions at the anode surpasses that of oxidation at the cathode. This is partly due to the sloped voltage profile of layered oxide c athodes: electrons gained through oxidation cause a decrease in cathode potential, lowering the driving force for additional oxidation [238]. This is not the case during a voltage hold, as the cathode can deliberately be maintained at high potentials and thus the oxidation reactions can be amplified. This approach has been used in the past to investigate the effect of electrolyte composition on the surface reactivity of cathodes at high voltages, [224, 224, 239] and was observed to lead to a net gain of capacity by the cell, [97] indicating that the rate of oxidation surpassed that of reduction side reactions. 91 The contributions from oxidation processes to cell capacity and the measured currents can be minimized by selecting a cathode with a flat voltage profile (see Figure 3.3 and the Appendix) that remains at low potentials during the voltage hold. A cathode like LFP satisfies both conditions, increasing the ability of the voltage hold technique to track electron exchanges at the anode SEI. But are these parasitic currents large enough to be measurable in real time? Figure 3.6A √ presents a simulation of the capacity lost by a cell due to calendar aging, assuming a t dependency; such dependency has been empirically observed in many studies [94, 95, 100, 240–242]. During the first month of aging (720 h), a cell with 15 years of calendar life would lose < 1.5% of its nominal capacity. The derivative of this capacity loss with respect to time is shown in Figure 3.6B, indicating the net currents associated with aging processes. After the first month of aging, a cell with 15 years of life would present an instantaneous rate of parasitic reactions of ∼10 µA per Ah of nominal capacity. Translating these values to the currents measured during a voltage hold, the expected magnitudes would be ∼50 nA for a 5 mAh coin cell, 500 nA for a small single-layer pouch cell and > 10 µA for commercial batteries. All these values are, in principle, measurable with modern electrochemical instrumentation. However, accurately sensing the currents generated by smaller cells could be challenging for conventional battery cyclers. For illustration, a lab scale Maccor Series 4100 cycler has an accuracy of 75 nA at its lowest current range (< 150 µA), which could measure the currents generated by a 1.2 Ah cell within < 0.63%; deviations for coin cells would be significantly larger. Here, we use 1.2 Ah cylindrical cells to evaluate quantitative aspects of the voltage hold technique, but also discuss how qualitative cell behavior can still be captured in coin cells if the rate of aging of samples is sufficiently distinct. Hardware specifications for all instruments used in this work are provided in the Appendix. We note that the sensing accuracy could be problematic even for large-format cells depending on the test settings. Using a Maccor cycler as an example again, the current range needed for C/10 cycling of a 1.2 Ah cell (< 150 mA) would have a current accuracy of only 75 µA. To circumvent this limitation, the tests discussed here were performed using automatic current range selection, which switches the sensing circuitry as the current decays during the voltage hold. An alternative would be to modify the experiment to 92 minimize Qrev and prevent the current from varying across multiple orders of magnitude. This can be achieved by allowing the cell to relax at open circuit prior to the hold and then performing the voltage hold at the post relaxation OCV (as shown in the Appendix) [104]. In this case, the test could be performed at a static current range with optimal sensitivity. Naturally, this option requires prior knowledge of the expected current values. 3.1.4 Evaluating Whether the Anode Overhang Will Affect the Measured Currents The electrochemical activity in a Li-ion cell occurs primarily within the geometric region of the incoming Li+ flux in the electrolyte. Consequently, when the battery is charged, lithiation of the anode occurs mainly in areas that are directly aligned with the cathode coating. If the two electrodes have identical dimensions, a slight misalignment during cell assembly could decrease their overlapping area, decreasing cell capacity [21]. To minimize the consequences of assembly imperfections, commercial cells typically contain excess anode area, known as “overhang”, which improves assembly consistency and decreases the likelihood of Li metal plating [21]. The exact size of this excess area depends on cell format. As an example, Lewerenz et al. recently estimated that the anode was 5.7% larger than the cathode in commercial 8 Ah LFP-Gr cylindrical cells [230]. For research-grade coin cells, geometrically oversizing the anode by ∼ 15% has been recommended to improve reproducibility [243]. Although this excess anode area is not the primary destination for Li+ ions, it is also not completely inactive. The overhang is much less prone to undergoing reversible electrochemistry so its SOC will be different from that of the overlapping area. Over time, this gradient will promote the transference of Li+ between central areas of the anode and the overhang [244, 245]. Hence, the SOC of the overhang will slowly approach the average cell SOC after extended storage, which can interfere with observations during calendar aging studies [21, 230, 245]. These interferences could appear as either added charge loss or apparent capacity gain, depending on whether the overhang is initially at a higher or lower SOC than the active anode area. During a voltage hold in LFP-based cells, electrons are transferred from the cathode to the anode to compensate for the charge lost in the latter. It is possible that the existence of an excess anode area could interfere with the correlation between measured current 93 and rate of parasitic reactions, as the overhang could drain charge from the active area or even be directly lithiated at minute currents. To evaluate the impact of the overhang on potentiostatic hold experiments, we performed tests using identical electrode pairs (LFP versus Gr-1 or LFP versus 15% Si-Gr) but varied which electrode contained the excess area: the anode or LFP. In the single-layer pouch cells used in these studies, the cathode had a larger areal capacity than the anode (N/P ratio< 1), to guarantee that the cathode would remain in the flat portion of the LFP voltage profile during the test, even at high anode SOCs and after accounting for the irreversible losses during the initial formation cycles. The near-constant LFP potentials and the slow cycling rates also helped eliminate the risk of Li plating along the anode edges in the presence of a cathode overhang. All cells had a nominal active area of 14.1 cm2 , which is the area of the smaller electrode that is directly underneath the larger 14.9 cm2 counter electrode, with a total of 5.4% of overhang excess. All cells were exposed to three formation cycles at C/10 (2.7 – 3.42 V for Gr-1 and 2.7 – 3.35 V for Si-Gr), and then charged to 3.335 V and held at this potential for 720 hours (one month). During the formation cycles, we observed that LFP-Gr-1 cells with anode overhang had 1-3% higher capacity than the ones with cathode overhang (Figure 3.7A), indicating that a fraction of the excess area appears to be readily accessible during cycling. These general trends were also observed with 15% Si-Gr anodes, although the cell-to-cell variability presented similar magnitude (Figure 3.7B). During the voltage hold, cells with anode overhang tended to require longer times to achieve a steady behavior (Figure 3.7C), which could also be indicative of lithiation of excess anode areas. However, once this initial behavior subsided, all cells presented identical rates of capacity exchange (Figure 3.7C,D), suggesting that the existence of an anode overhang did not appear to significantly affect the trends measured within the duration of these voltage hold experiments. If these results can be generalized, they indicate that the observations obtained with this technique may not be adversely impacted by the existence of excess anode area, making voltage holds compatible with state-of-art cell assembly methods. Can voltage holds provide accurate quantitative predictions of calendar aging? The previous sections indicated that, under the correct conditions, potentiostatic holds could provide information about the rate of parasitic reactions at the SEI. If the technique accurately measures this rate in real time, then it would potentially be able to quantita- 94 tively predict the calendar life of cells as limited strictly by Li+ inventory loss at the anode, as suggested in Figure 3.2D. One direct approach to assess the accuracy of the method is to directly compare the irreversible capacity measured during a voltage hold (Qirrev ) with the actual loss calculated from the cycles before and after the hold (Qloss ). An agreement between these two values would indicate that parasitic processes are correctly measured, and that reversible and irreversible currents can be properly distinguished. Figure 3.8 walks through the comparison between Qhold (t), Qirrev (t), and Qloss for commercial Gr-LFP 18650 cells, with Figure 3.8A showing an example voltage profile of the experiments. These cells are the same ones used in Figure 3.4 and provide a stable system for proof of concept of quantitative analysis capability. Importantly, the irreversible capacity passed during the voltage hold can be calculated by rearranging 3.1: Qirrev (t) = Qhold (t)–Qrev (t) (3.7) where Qrev (t) is the increase in SOC during the voltage hold as discussed in Figure 3.4B. At sufficiently long times, Qrev (t) ideally becomes a constant rather than a function of time, and can thus be calculated as a total value passed by the end of the voltage hold (Qrev (t), 3.5). Figure 3.8B shows the raw data of the Qhold (t) and the true capacity loss (Qloss , 3.4) calculated from the surrounding cycles. Figure 3.8C shows Qirrev (t) with the reversible capacity (Qrev (t) = ∼21% and is constant for the times considered in the commercial 18650 cells, Figure 3.5A) subtracted from Qhold (t) measured during the hold. It is clear that Qirrev overestimates the capacity loss at all cases, and that this error grows with test time. Consequently, Qirrev grows with time much faster than Qloss (Figure 3.8D), indicating that any extrapolation of the measured voltage hold data would lead to a large overestimate of the actual calendar aging rate of the cell. Indeed, attempts to fit and extrapolate the voltage hold data assuming common functional forms used in the literature to describe time-dependent degradation consistently resulted in unexpectedly low calendar lives. These observations strongly suggest that, unfortunately, the voltage hold test may lack quantitative utility. Such a discrepancy between Qirrev and Qloss could be explained by the following: i) parasitic currents are consistently overestimated by the sensing hardware; ii) there could 95 be additional processes contributing to the measured current. The Appendix shows that the expected error based on typical instrument specifications would be less than 0.2 % of the cell capacity for a month-long hold of these commercial cells, which cannot explain the more than 3 % disparity between Qirrev and Qloss in Figure 3.8C. Therefore, we hypothesize that additional processes contributing to the measured current must be responsible for the observed discrepancies. We next consider the nature of these processes. One possibility is that the reversible (anode lithiation) contributions to the measured capacity could remain significant over the duration of the voltage hold, causing the irreversible (parasitic) contributions responsible for calendar aging to be overestimated. However, the inset in Figure 3.8B shows that Qrev as a function of hold time is relatively constant (< 0.5 % variation with no clear trends, which could be caused by small cell-to-cell differences) and therefore, reversible charging is unlikely to be the source of extra capacity. Furthermore, as seen in the Appendix, when an OCV relaxation is performed prior to the voltage hold, reversible lithiation is minimized but the long-term trends are still like a voltage hold without an OCV step before it. This further proves that reversible lithiation cannot be the only source of excess capacity. Yet another possible process is that the large overestimate of irreversible reactions during the voltage hold may also be due to reversible capacity loss. One example of such processes is charge exchange with the anode overhang, which we ruled out based on the data shown in Figure 3.7. Another possibility is the contribution of reversible self-discharge processes. During traditional calendar aging experiments in which cells are allowed to age at open circuit (as in Figure 3.4), it is commonly observed that the discharge immediately after the storage period overestimates the true capacity fade. Processes such as electrolyte oxidation and redox shuttles during long-term storage can temporarily decrease cell SOC without causing permanent loss of Li+ inventory, [21,237,246,247] causing cell capacity to rebound in subsequent cycles. It is conceivable that these temporary losses are included in the charge measured during the voltage hold, leading to artificially high fade rates. From the cells charged to 3.35 V and stored at open circuit (Figure 3.4), we can estimate the reversible self-discharge by taking the difference between the perceived loss (the difference between the discharge and charge immediately after and before the aging step, respectively) and Qloss (calculated using Equation 3.4). For the open circuit cells, 96 the reversible self-discharge is less than 0.6% for all test durations (Figure 3.9) indicating that reversible self-discharge does not completely explain the greater than 3 % difference between Qhold and Qloss in Figure 3.8C. Similar calculations for cells tested using voltage holds (the difference between Qloss , 3.4 and Qirrev , 3.7) resulted in even higher capacity values (1-4 % depending on time). Unlike cells stored at OCV, voltage hold cells are pinned at the setpoint voltage of 3.35 V. Therefore, reversible self-discharge processes should manifest as excess capacity rather than a decrease in voltage as with OCV reversible self-discharge. Comparing the magnitude of this excess capacity during the voltage hold in Figure 3.9 with the difference between Qhold and Qloss in Figure 3.8C, reversible processes leading to excess capacity could be a significant, but perhaps not the only, contributor to the overestimated capacity loss. The cause of this reversible excess capacity during the voltage hold is not fully understood, though is clearly greater than the analogous reversible self-discharge capacity during OCV aging. Due to the presence of additional processes besides reversible charging (Qrev , anode lithiation) and parasitic processes, a better description of the total capacity exchanged during the potentiostatic hold may be: Qhold (t) = Qirrev (t) + Qrev,charging (t) + Qotherrev.processes (t) (3.8) Without a better understanding of this third term, quantitative prediction of calendar aging is difficult. The lack of quantitative behavior in the voltage hold of well-behaved commercial Gr-LFP cells will translate to research-grade systems where the same issues will be present, and possibly exacerbated due to complexities in different potential profiles and increased variability. Can voltage holds resolve semiquantitative calendar aging trends? In the previous section, we demonstrated that potentiostatic holds fall short of perfectly quantifying the electron exchanges that lead to Li+ inventory loss. Even though nonidealities in electrode profiles could theoretically cause underestimation of instantaneous parasitic currents (Appendix), the previous section showed that the opposite behavior occurs in reality: the measured rate of side reactions was consistently larger than the actual rate of parasitic processes. Although a quantitative analysis seems out of reach of this technique, semiquantitative information could still be available if the errors are reasonably systematic (i.e., if deviation is nearly constant across various testing conditions). Semiquantitative 97 analyses could still be very useful for accelerating calendar aging studies, as they provide estimates of the degrees of improvement among a set of variables, such as voltage, SOC and electrolyte composition. To evaluate the merits of potentiostatic holds for semiquantitative analysis, we once more resorted to commercial 18650 format LFP-Gr cells to serve as reproducible model systems. The cells were received with an OCV of ∼3.292 V and presented initial C/10 discharge capacities of 1182 ± 9 mAh. Cells were refrigerated shortly after being received to prevent calendar aging, with select cells being warmed up for testing (at 25 °C). A total of 8 cells were charged to and held at 3.24, 3.26, 3.292, 3.32, 3.335 and 3.35 V for 720 hours (30 days), with duplicates at 3.292 and 3.335 V; the test protocol was like the protocol illustrated in Figure 3.8A. The equilibrium SOCs achieved after relaxation during the hold at these voltages are shown as “x” in Figure 3.10A. These SOCs could differ significantly from that exhibited at the same voltage during constant current cycling (CC, in Figure 3.10A), with the departure depending on the proximity to plateaus in the full-cell profile. With depolarization, cells can gain a significant amount of capacity as plateaus become accessible, making the magnitude of Qrev strongly dependent on the hold voltage. The final SOC after the initial relaxation during the hold is equal to the capacity during the CC charge plus Qrev . As clear from Figure 3.10A, the cell tends to move away from plateaus after the initial relaxation. The Qloss observed for all cells and the measured Qirrev are shown in Figure 3.10B. Just as discussed in the previous section, Qirrev was observed to be consistently larger than the actual amount of capacity that is lost by the cells, and this gap appears to become larger at higher voltages. Interestingly, cells tested at 3.24 and 3.26 V actually gained capacity after the voltage hold (i.e., they displayed a negative loss in Figure 3.10B), while cells at 2.292 V were at the threshold of gaining/losing charge. This type of behavior (capacity gain during calendar aging at low voltages) has been reported previously and has been ascribed to charge being driven out of the overhang areas and distributing across the bulk of the anode, which is at a lower average SOC than the overhang [21, 230, 244, 245]. Since the cells were received at ∼3.292 V, the overhang is presumably at an SOC corresponding to that potential, and the near zero losses observed at 3.292 V supports the hypothesis of the overhang being the source of this behavior. The fact that the measured exchanged 98 capacity is larger than the actual loss irrespective of the aging potential relative to the initial cell OCV again negates major contributions of the overhang to the trends measured during voltage hold within this timeframe of one month. The capacities exchanged during the hold at all voltages, each normalized by the respective cell capacity, are shown in Figure 3.10C. The capacity curves were vertically shifted to have identical values at t = 300 h, to ignore the contribution of Qrev and allow better visual comparison of the trends. The tests presented remarkable reproducibility, as exemplified by the identical behavior of the sets tested at 3.292 and 3.335 V. In most cases, aging is faster (that is, the capacity curves visibly display a higher rate of capacity loss) at higher cell potentials, in agreement with the known trends of SEI growing faster at higher SOCs. A clear exception is the cell held at 3.35 V, which is discussed in detail below. To attempt to quantify these aging trends, we used the terminal slope of capacity √ √ versus t( h) plots as proxy for the rate of calendar aging, as satisfactory linear fits were obtained for most datasets. The fitted slopes are shown in Figure 3.10D (filled circles). The symbols are clearly arranged in the traditional S-shape of the plateaus of graphite; SOCs within a plateau exhibit a similar anode potential, and thus a similar driving force for reduction reactions, leading to similar aging rates [94, 95]. If the voltage hold can accurately resolve the semiquantitative relationship between aging rates at various SOCs, the filled circles in Figure 3.10 should coincide with the gray symbols. Figure 3.10D provides two main observations: i) the slopes extracted from potentiostatic holds generally follow the expected qualitative progression, but not necessarily with quantitative exactitude; ii) the test at 3.35 V largely underestimates the rate of aging. In the Appendix, a detailed discussion about the origin of this behavior is provided. Briefly, it can be traced back to the fact that, at 3.35 V, the cell is no longer at the plateau portion of the LFP voltage profile (see Figure 3.10A), causing aging to induce constant shifts in the electrode potentials over time that consequently lead to the measured current underestimating the rate of parasitic processes. Once that is accounted for, the correct qualitative trends can be obtained (see Appendix), though they remain inaccurate for quantitative purposes. Overall, our studies indicate that comparative aging trends can be directly assessed when experiments are performed within a flatter portion of the cathode profile (such as all voltages except for 3.35 V in Figure 3.10). More generally, it appears that the correct 99 qualitative trends can be obtained when the relative slopes of the voltage profiles of the cathode and anode are accounted for (as shown in the Appendix). Nevertheless, even after correction of the aging rates, the quantitative relationship between rates observed at all SOCs does not follow the expected trends; that is, the measured aging rates do not rigorously follow the path traced by the gray symbols in Figure 3.10D. These observations suggest that potentiostatic holds may be limited to providing qualitative information about parasitic processes. Further exploration of the qualitative aspect of the technique is provided in the next section. 3.1.5 Can Voltage Holds Provide Qualitative Information About Calendar Aging? In the previous section, we showed that the qualitative aging rates recorded during potentiostatic holds at different voltages followed the expected trends of higher parasitic currents at higher SOCs. Despite the quantitative limitations of the technique, this observation suggests that voltage holds could be useful for the fast screening of materials and compositions. This could accelerate the discovery of new methods to improve calendar aging by quickly identifying promising systems, that can then be validated in long-term RPT-based aging experiments or future rapid quantitative methods described in Figure 3.1B. This section will show additional qualitative validation of the voltage hold technique. A second example of potentiostatic holds correctly capturing the qualitative aging behavior of cells is shown in Figure 3.11. The relative trends observed in the voltage hold should correspond to the same trends in more traditional calendar aging experiments when new materials are compared to a baseline. The best performing material would therefore be the one with the shallowest slope for capacity lost during both the voltage hold and during more traditional calendar aging at open circuit. An example of this for the comparison between electrolytes for a graphite electrode is shown in Figure 3.11A-B. The Gen2+10% FEC electrolyte shows a greater rate of capacity fade in both the voltage hold at 3.335 V and traditional OCV calendar aging compared to the electrolyte containing Gen2 + 2% VC + 2% ES + 2% TMSPi. This demonstrates the qualitative use of the voltage hold since the voltage hold and OCV trends both indicate the same electrolyte formulation decreases the calendar aging rate of the same electrode type. Importantly, note how this 100 conclusion is already evident < 200 h into the voltage hold, which is significantly faster than the time interval covered by the first few RPTs in Figure 3.11. As discussed above, this time advantage is easily accessible for systems that exhibit fast current relaxation, such as graphite. A third example of correct qualitative behavior is shown in Figure 3.11C-D. It is widely reported in the literature that higher temperatures will exacerbate parasitic processes and thus accelerate the calendar aging of LIBs [94, 98, 248–251]. Figure 3.11C shows the exchanged capacity measured during the voltage hold of LFP versus Gr-2 coin cells at 3.335 V at 10, 30 and 50 °C. Curves were normalized and vertically shifted, and only a portion of the data is shown for clarity. Tests at higher temperatures yielded capacity curves with higher slopes, consistent with a thermal acceleration of aging processes. Furthermore, direct extraction of aging rates (from capacity versus sqrtt plots, as in Figure 3.10D) resulted in reasonable linearity of an Arrhenius plot (Figure 3.11D), in agreement with previous studies on the effect of temperature on calendar aging [252–254]. This finding agrees with the work of Lewerenz et al., that also observed an Arrhenius dependency in currents measured during voltage holds at various temperatures [230]. These examples further demonstrate the validity of using voltage holds for qualitative comparisons of the rate of reactions at the SEI across different systems (electrolyte additives) and experimental conditions (temperature and voltage). However, caution must be taken during experiment design to ensure reliable results. As we noted above, differences between material performances during the voltage hold may be subtle, especially for smaller cell formats in which parasitic currents approach the hardware detection limits. Due to the higher variability inherent with research-grade cells, replicates are highly recommended to correctly identify relative trends; additionally, pouch cells tend to present superior reproducibility to more common coin cells. We emphasize that voltage holds should be used to guide the initial down selection of promising systems based on the stability of the SEI layer, and that the technique does not replace traditional long-term experiments, such as the one used by the United States Advanced Battery Consortium (USABC) that utilizes an OCV-RPT style test [219]. Certain aging mechanisms, such as the ones involving loss of accessible active material, will only become evident after such extended experiments. 101 3.1.6 A Quick Guide to Using Voltage Holds for Qualitative Screening Important considerations for using voltage holds for qualitative comparisons to predict calendar life are summarized in Table 3.1. Three of the most important variables for the correct qualitative use of voltage holds include: i) sufficient lithium inventory; ii) sufficient time to allow reversible processes to relax; and iii) a carefully chosen voltage. The protocol used in this paper was designed to only evaluate lithium inventory loss to the SEI. For obtaining reliable results, the cathode must present a large enough lithium supply that is accessible with negligible potential variation. This way, reactions at the SEI and reversible lithiation processes can draw from this supply during the voltage hold while maintaining the electrodes at constant potentials. If the counter electrode does not contain sufficient lithium, a drop in current during the voltage hold may occur, leading to an erroneous conclusion about the quality of the SEI (Figure 3.12A-B). This artifact is caused by the cathode shifting away from a SOC range with a flat voltage profile, causing the anode SOC to vary during the test and the parasitic processes to be underestimated (see Appendix). More specifically, we recommend that tests should be done with LFP counter electrodes in such a way that the entire voltage hold is still within the LFP plateau. Therefore, the total capacity required (from both reversible and irreversible processes, along with initial SEI formation) must be understood. In commercial LFP-Gr cells, the rate of capacity loss is slow enough that voltage holds with cells with N/P ratio > 1 are still possible. In research-grade coin cells, however, excess cathode capacity and electrolyte may be necessary to prevent inventory exhaustion. Qualitative voltage holds will only be accurate if the recorded electron exchanges are primarily due to parasitic processes rather than reversible lithiation. The longer the voltage hold, the more likely this is to be true since reversible relaxation should have asymptotic behavior and subside over time. During experiment design, an inverse polarization test can be used to estimate the relaxation time of reversible processes (see Appendix). Aging trends should therefore be considered only at times much greater than the relaxation time indicated by inverse polarization experiments. The slow relaxation of reversible processes is demonstrated in Figure 3.12C for an 80% Si electrode, where the reversible capacity as a function of voltage hold length increases after 720 hours, indicating the electrode has still not reached equilibrium. Graphite, on the other hand, remains relatively stable at ∼ 1.5% 102 regardless of voltage hold length, suggesting much faster relaxation kinetics. Hence, the data shown in Figure 3.12A is insufficient to assess the relative rates of capacity loss across both anodes, as much of the current for Si arises from reversible lithiation rather than side reactions. In other words, an electrode with slow lithiation kinetics will also lead to a more persistent reversible current contribution, increasing the observed current during the hold. The importance of this point is further highlighted in Figure 3.12D-E, using LFP versus Gr-1 and LFP versus 15% Si-Gr cells. When comparing the capacity trends after only 175 hours into the hold, the Gr electrode looks much superior to the Si-Gr anode, as evident from the smaller slope of the capacity curve. However, if the same cells are compared after 1400 hours, the difference between the two data sets becomes much more subtle. At the beginning, the current is dominated by reversible current, so the silicon electrode, with slower lithiation kinetics, will appear to perform worse. This demonstrates the importance of having a sufficiently long voltage hold even for qualitative comparisons. We note that the determination of the time necessary for an electrode to relax and be dominated by parasitic current is difficult. Although the inverse polarization experiments can suggest approximate times, we performed these experiments in half-cells and under delithiation conditions, so it must be assumed that the times determined under these conditions are representative of and are transferrable to full-cells under anode lithiation conditions. An option to help decrease hold time would be to have an OCV step prior to the voltage hold, with the voltage after relaxation being used as the hold voltage; this minimizes reversible lithiation during the hold so qualitative comparisons can be made more quickly (Appendix). Finally, it is important to consider the choice of test voltage. Since reversible lithiation will occur early during the hold, the effective cell SOC during the test will differ from the one immediately prior to the hold. This is exemplified in Figure 3.12F by the dQ/dV curves of the charge and discharge before and after the potentiostatic hold, respectively, for a commercial LFP versus Gr 18650 cell. When the cell is initially charged to either 3.335 V or 3.35 V, the anode potential is approximately ∼120 mV versus Li/Li+ , as can be inferred from the sharp peak in the differential capacity profile at those potentials. However, the SOC of each cell will progress differently during the voltage hold. When held at 3.35 V, the anode will be lithiated down to the ∼89 mV versus Li/Li+ plateau of graphite (see 103 the new delithiation peak at 3.31 V, labeled *), while the anode potential is insufficient to lithiate graphite to its second plateau potential when held at a cell voltage of 3.335 V. In addition to extending the time needed for reversible processes to relax, the 3.35 V scenario will also expose the cell to a flatter region of the anode profile, which can affect the ability of the test to accurately measure reduction reaction rates (see Appendix). Also, reaching an anode plateau increases the likelihood of exhausting the inventory of the cathode, causing artifacts such as the ones discussed in Figure 3.10D and Figure 3.12B. Although it is difficult to determine the effective SOC during the hold without prior knowledge, we found that using pseudo-OCV profiles provided reasonable success; see Figure 3.10A for details. Considering test voltage can be particularly important when comparing anodes containing different active materials (Si versus graphite, for example). Since the driving force for reduction reactions is proportional to the anode potential, the chosen cell voltage must be as equivalent as possible between different anodes. This common anode potential should ideally be in regions in which the slopes of both voltage profiles is much higher than that of the cathode, to improve the sensitivity of the technique to measuring SEI processes. Although the recommendations above improve the reliability of experiments, care must also be taken when analyzing the data. In the discussion above, we either compared systems through inspecting the visual trends of exchanged capacities over time (Figure 3.11C), or directly extracted slopes from capacity curves after trends remained linear for hundreds of hours. Both procedures ensure that aging can be judged based on general future trends, rather than on an instantaneous behavior. Shifting curves vertically to eliminate the contribution of Qrev is also useful to facilitate visualization. The progression of measured currents is often less clear and less robust (see Appendix), and thus is not directly used in our analyses. Despite the utility of the voltage hold for the fast qualitative screening of materials, it does have limitations. The technique provides information on lithium inventory loss but does not explicitly capture other degradation mechanisms such as the loss of active material due to electrical isolation which has been reported in commercial Si-Gr electrodes [21, 255]. The conditions of the voltage hold are also important, and adequate thought must be given to the potentials of the voltage hold experiments to make fair comparisons 104 between different electrodes. Additionally, voltage hold duration must be sufficiently long to resolve parasitic current trends without being overshadowed by reversible current, and materials with poor kinetics may require very long experiments. Lastly, the approach taken in this paper focuses on understanding calendar life changes due to either electrolytes or electrode materials that affect the stability of the SEI on the anode. Hence it cannot be directly extended to studying the effect of cathodes on the system due to the difference in voltage profiles of different cathodes. For example, the use of an NMC cathode for a voltage hold will lead to a shift in the anode profile with lithium loss and also introduce oxidative instability of the electrolyte as a variable, further convoluting the meaning of the measured current. Such data would thus not be directly comparable to the data acquired from tests using LFP counter electrode, even if the same anode was held at the same effective potential. Finally, results from voltage hold tests were observed to be much more consistent when using formats that are larger than conventional coin cells. The use of pouch cells is encouraged when possible, and replicates are necessary to ensure that real trends are observed. Occasionally, differences in performance between two systems may be too subtle to be reliably measured using short voltage holds, requiring conventional long-term testing for a complete assessment. 3.1.7 Conclusions Decreasing the rate of calendar aging has important practical relevance for the mass deployment of Li-ion batteries, as capacity fade due to parasitic reactions at the SEI can affect the economic feasibility of many applications. Since this time-dependent aging is the cumulative effect of very slow processes, studying and mitigating these phenomena can be extremely resource-intensive. Hence, creating methods that can quickly assess future aging trends is needed to support the development and validation of new active materials and electrolyte additives. This chapter discussed whether potentiostatic holds can be used for such purposes and the results are summarized in Figure 3.13. When cells are maintained at a certain voltage for an extended time, the residual charging eventually subsides and the measured current then becomes indicative of the rate of parasitic electron exchanges occurring at the electrodes. We show that, if experiments are performed with the cathode maintained at 105 relatively low and invariant potentials (such as in the case of LiFePO4 ), then the measured current becomes extremely sensitive to reduction reactions at the anode. Thus, in principle, potentiostatic holds could allow the direct measurement of the time dependency of electron and Li+ losses to the SEI and could quickly provide a description of calendar aging trends. Our work explored experiments in various cell formats to evaluate the quantitative power of this technique. Our studies indicated that the exchanged capacity recorded during the voltage hold consistently overestimated the real capacity loss experienced by the cells during the experiment, exhibiting faster aging trends than expected for the cell chemistry. Moreover, this deviation did not seem to arise from a systematic error, as the relative aging rates exhibited by cells tested at different SOCs did not reproduce the well-established quantitative behavior observed for graphite anodes. The deviation was also observed to increase with increasing hold times, suggesting that errors were both SOC- and time-dependent. Nevertheless, the qualitative trends observed in these tests were generally correct: cells tested at higher SOCs (and thus held at lower effective anode potentials) exhibited faster rates of aging. We further offered two additional examples of success of the voltage hold technique in identifying qualitative trends, by showing the effect of temperature and electrolyte composition on calendar aging. These observations suggest that this method can be used as a screening tool, highlighting formulations and conditions that could deliver improved calendar life. This initial screening would help decrease the parameter space of longer aging studies, optimizing the allocation of resources and improving the likelihood of a successful outcome. We also explored whether the fundamental assumption that charging currents will eventually vanish during the hold was correct. We described how inverse polarization assays could be used to estimate the relaxation times for anode materials and demonstrated that silicon could have significantly slower kinetics than traditional graphite electrodes, even in thin films. Consequently, Si would require long voltage hold experiments for the results to be truly descriptive of aging behavior. Alternatively, we note that analysis of the early current decay of Si-based cells could be an effective tool to evaluate how improvements in electrode and electrolyte composition improve the kinetics of Si. While this study focused on the use of LFP as a counter electrode, we recognize that this 106 is not the case with typical high-energy cells. The favored layered oxide cathodes typically present sloped voltage profiles, which generally makes voltage tests more sensitive to oxidation than to reduction side reactions. While correlating the measurements using these cathodes with capacity fade is difficult, these tests still represent the rates of some parasitic process and can still provide useful insights if carefully designed. We expect that the present work will provide battery scientists and developers with a solid base upon which to build and that future work with improved instrumentation and varying test conditions could expand the capabilities of this method. More importantly, we hope this work helps emphasize the need for designing tools that can expedite the cycle of innovation in the battery field. 107 a RPT b time aging RPT aging … time aging + measuring Figure 3.1. Conceptualizing an ideal accelerated calendar aging experiment. A) Traditional experiments alternate periods of aging at open circuit voltage and reference performance tests (RPTs) that quantify the performance loss. Typically, month-long storage periods are repeated for a year or more to sufficiently resolve the time dependency of calendar aging. The vast majority of the time is spent aging the cells, rather than measuring its consequences. B) Test duration could be expedited if aging and measurement of the resulting capacity loss were performed simultaneously. The enhanced time resolution for information acquisition could drastically shorten calendar aging experiments. 108 Table 3.1. Considerations for successfully implementing qualitative voltage hold comparisons to infer relative calendar life performance Steps Preparation Considerations • Inverse polarization experiments can be performed to determine necessary voltage hold time to allow reversible lithiation relaxation • Consider choice of test voltage to ensure different anodes are at similar potentials, enabling comparison between different systems •Cathode must have flat voltage profile Performing the experiment • Cathode must supply sufficient Li+ inventory to remain at SOCs where its voltage profile is flat during the entire voltage hold experiment • Perform voltage hold sufficiently long to relax reversible lithiation as dictated by inverse polarization results • Plot both hold current and capacity to make trends more clear • Compare trends of capacity plots after removing reversible lithiation contributions Data Analysis • Make replicates to ensure observed trends are valid • Greater consistency may be achieved with pouch cells as compared with coin cells Limitations • The method only probes lithium inventory loss, not other degradation mechanisms 109 Figure 3.2. Depiction of the voltage hold to evaluate calendar liife. A) Typical voltage versus time profile of a voltage hold experiment. B) The current response measured during the voltage hold period comprises contributions from reversible processes that dominate early in the hold and parasitic process that irreversibly consume Li+-inventory but slowly decrease as the SEI matures and becomes more passivating. C) Integration of the current response during the voltage hold yields the exchanged capacity. The significant rise early in the testing period is due to residual low-rate charging of the cell (reversible processes) while the shallower profile later in the hold is more representative of the irreversible processes that affect the cells calendar lifetime. D) The time dependent behavior of irreversible capacity measured during the voltage hold could potentially be extrapolated into the future to predict the calendar aging behavior of a cell, assuming Li+ inventory loss at the anode as the sole calendar aging mechanism. The blue and orange curves represent examples of faster and slower degradation, respectively. 110 Figure 3.3. Depiction of voltage shifts of the cathode for LFP and NMC cathodes. A) Potential profiles of a Si-LFP full-cell before and after a voltage hold where Qirrev = 20%. The correspondence between the Si-LFP full-cell potential profiles B) and the Si anode potential profiles C) before and after the voltage hold at 3.35 V are also shown. D) Potential profiles of a Si-NMC full-cell before and after a voltage hold where Qirrev = 20%. The correspondence between the Si-NMC full-cell potential profiles E) and the Si anode potential profiles F) before and after the voltage hold at 3.6 V are also shown. 111 Figure 3.4. Comparison between voltage hold and OCV aging. A) Qloss values versus aging time for Gr-LFP 18650 cells that aged at OCV or during a voltage-hold. B) Differential capacity traces of a Gr-LFP cells aged at OCV or during a voltage-hold. The differential capacity traces are calculated from the lithiation immediately prior to, and the delithiation immediately after the aging period to demonstrate the effect of the aging method on the cell’s SOC. The * marks the delithiation peak indicating the cell aging via a voltage-hold reached a higher SOC than the cell aged at OCV, due to the reversible lithiation required to maintain the cell voltage at the hold potential. 112 Figure 3.5. Reversible capacity (Qrev ) of different anodes measured during voltage holds, with each data point representing a cell having underwent a voltage hold for 7.5-, 15-, 30-, 60-, or 90-days. A) The Qrev values of several different graphite-containing cells. Significant differences in Qrev values of the graphite-containing cells between the Li half-cells and the LFP full-cell stem from the graphite electrodes experiencing slight differences in absolute electrochemical potentials during the hold; these effects are detailed in Figure 3.10 and related discussion. B) The Qrev values of various silicon-containing cells. 113 √ Figure 3.6. Simulated calendar aging of a hypothetical Li-ion battery assuming a t time dependency for capacity fade and 5, 10 or 15 years of calendar life (time in which charge loss equals 20% of nominal cell capacity). A) Capacity loss relative to initial cell capacity. B) Parasitic current associated with the rate of capacity loss shown in panel a, per Ah of cell capacity. The magnitude of parasitic currents can be sizable for large-format cells. The legend in panel b applies to both panels. 114 Figure 3.7. Evaluating the effect of overhang on the trends measured during a voltage hold. A) Capacities during formation cycles for LFP versus Gr-1 cells with cathode or anode overhang. Capacities are normalized by the weight of active material in the anode. B) Like panel A, but for LFP versus 15% Si-graphite cells. C) Shifted normalized capacities exchanged during a voltage hold at 3.35 V of LFP versus Gr-1 cells. D) Like panel C, but for LFP versus 15% Si-graphite cells. Capacities were arbitrarily shifted to match at 300 h to erase small variations in Qrev and allow a more direct visual comparison of trends. The legend in panel C applies to all panels. 115 Figure 3.8. Comparison between the loss predicted by the voltage hold and actual capacity loss in the Gr-LFP 18650 cells. A) Potential profile of voltage hold experiment run on the Gr-LFP 18650 commercial cells. The voltage holds were done at 3.35 V. B) Capacity loss measured during 90-day V-hold (Qhold ) and capacity loss measured using the RPT and forming cycles of 3 separate cells aged for 30-, 60-, and 90-days, respectively (Qloss ). Capacity loss values are shown as a percentage of the last forming cycle discharge capacity of each cell (Qcell discharge 3). The reversible capacities (Qrev ) shown in the inset plot are determined by subtracting the charge capacity immediately prior to the voltage hold from the discharge capacity immediately after the voltage hold. C) Capacity measured during 90-day V-hold with the 90-day Qrev value subtracted (Qirrev ). D) Shows Qhold arbitrarily shifted to directly compare trends with Qloss . Qhold exhibits a greater slope further indicating the overestimate of loss compared to the actual loss, Qloss . 116 Figure 3.9. Comparison of the reversible self-discharge measured during 30, 60, and 90-day OCV rests and the excess capacity measured during 30, 60, and 90-day 3.35 V voltage holds of commercial LFP-Gr 18650 cells. The values for OCV rest cells were calculated by taking the difference between Qloss (calculated using 3.4) and the perceived loss (calculated using the charge/discharge before/after the OCV aging step). The values for the V-hold cells were calculated by taking the difference between Qloss (calculated using 3.4) and Qirrev (calculated from 3.7). In the case of OCV rest cells, the values shown represent reversible self-discharge while in voltage hold cells, which are pinned at a set voltage (and SOC), the values shown are excess capacity from reversible processes that are not fully understood. The capacities were normalized to the delithiation before the hold. 117 Figure 3.10. Potentiostatic holds of LFP-graphite cylindrical cells at various voltages. A) SOC of the cell immediately before the voltage hold (▲) and at the end of the voltage hold (x). During the hold, the cell SOC grows by a factor equal to Qrev , which can be particularly large at the vicinity of plateaus as charge can be exchanged at small voltage increments. The black voltage profile was obtained at C/10 during charge, while the gray curve is a pseudo-OCV obtained by averaging C/10 charge and discharge profiles. Although the pseudo-OCV does not provide an exact picture of the cell SOC after the hold, it is a good approximation, especially at high SOCs. B) Qirrev (•) and actual capacity loss (x) as a function of hold voltage. Capacity loss was calculated using 3.4. All values are normalized by the initial full capacity of the respective cells. C) Integrated capacity exchanged during the potentiostatic hold. Values were normalized by the initial capacity of each cell, and then shifted vertically to present identical values at t = 300 h. D) Rates of aging extracted from the voltage holds (colored circles, right axis) and normalized capacity fade. The legend in panel c applies to all panels. Data shown as gray symbols indicate the normalized total capacity loss measured at each SOC, as reported in the following works: • ( [252]), x ( [100]), + ( [94]), ⋆ ( [256]), ♦ ( [257]). 118 Figure 3.11. Example of qualitative behavior of the voltage hold. A) Shows the voltage hold of graphite-1 anodes and LFP cathodes with electrolytes of Gen2+10% FEC and Gen2 + 2% VC + 2% ES + 2%TMSPi at 3.335 V in 2032 coin cells recorded on Maccor battery cyclers. The curves are shifted to account for differences in reversible lithiation and their local slopes are compared at 700 hours and were normalized to the charge capacity directly before the hold. A greater slope indicates a larger parasitic current suggesting worse calendar life in the Gen2+10% FEC electrolyte. B) Depicts the calendar aging comparison of the two electrolytes with the same graphite electrode against NMC622 in 2032 coin cells. Three reference performance cycles were performed every ∼720 hours of OCV rest starting at 4.1 V. Capacity retention is calculated from the last delithiation of the RPTs relative to the final delithiation of formation cycles prior to calendar aging. Consistent with the voltage hold, the larger negative slope of the Gen2+10%FEC electrolyte indicates worse performance. Rapid fade for both systems may have been caused by unreasonably large anode overhang and/or reactivity of coin cell parts. C) Shows the voltage hold of Gr-2 versus LFP with Gen2+10% FEC electrolyte at 3.335 V at 10, 30, and 50 °C. The expected trends of increasing slope, indicating increasing parasitic processes, is observed going from 10 to 50 °C. D) Shows the approximate correspondence of √ the temperature data with the Arrhenius relationship. The slopes of the capacity versus t were taken from 150 to 360 hours for the y axis values. The agreement of the OCV measurements and voltage hold experiments along with the correct qualitative behavior of the temperature data provides a proof of concept of the qualitative ability of the voltage hold. 119 Figure 3.12. Examples of important considerations when performing voltage hold experiments. A) Shows an example of a 180 hr voltage hold at 100 mV versus Li/Li+ of a Si and Gr-2 electrode half-cell with sufficient Li inventory while B) shows a 180 hr voltage hold of the same Si electrode with (Li counter electrode) and without (LFP counter electrode) sufficient Li inventory. The potential at the anode was ∼ 100 mV versus Li/Li+ . The drop in current in the limited Li case represents the exhaustion of the Li inventory not a decrease in parasitic processes and results in a lower current than the excess Li case. C) Shows the reversible capacity passed as a function of voltage hold length at 3.35 V for a Si-Gr and Gr-2 electrode (Li counter). The reversible capacity increases with time for the Si indicating the reversible processes still have not relaxed after a 720 hr hold. Conversely, the reversible capacity of the Gr remains ∼ stable with voltage hold length due to more facile relaxation. F) Shows the effect of voltage choice on a Gr-1 electrode (versus LFP). The lithiation is prior to the voltage hold and the delithiation is after a 720 hr voltage hold at either 3.35 or 3.335 V. At 3.35 V, the final plateau of Gr lithiates during relaxation (delithiation peak at ∼ 3.31 V (*)). If the voltage is adjusted to 3.335 V, the last plateau is not lithiated. 120 Quantave Qirrev during hold is inconsistent with actual loss, Qloss X Semi-Quantave Correct general trends with SOC, but incorrect values X Qualitave Capable of evaluang relave calendar life performance with proper experimental setup ✓ Figure 3.13. Summary of results for quantitative, semiquantitative, and qualitative application of the voltage hold to evaluate calendar life performance. CHAPTER 4 SCANNING ELECTROCHEMICAL MICROSCOPY TO QUANTIFY THE CHANGES IN SILICON THIN FILM PASSIVATION OVER TIME Silicon has been found to be inherently nonpassivating leading to rapid capacity fade due to consumption of the lithium inventory from continuous reduction of the electrolyte. Although this has been heavily studied during cycling conditions, passivation during OCV calendar aging conditions has been less of a focus. The mechanism of mechanical degradation of the SEI during both cycle and calendar aging is also not well understood. 4.1 Moiré Interferometry as a Possible Method for Tracking in situ SEI Strain Moiré interferometry was investigated as a technique to measure SEI strain in situ. This technique was promising because it is nondestructive, can be used with a reasonably normal cell setup (i.e., closed cell format), and allows for strain measurements below the diffraction limit of light. Moiré interferometry is useful for a variety of applications in fields including microelectronics, thermal processing, fracture mechanics, and residual stress [258]. Chapter 1 discussed the required resolution to measure SEI strain and how the use of either a more flexible or thinner substrate can improve resolution when looking from the backside of the electrode. Based on these calculations, a thin Cu foil of 500 nm was used as a substrate with a 50 nm Si thin film as the electrode. Initially, a geometric moiré setup was used (Figure 2.3A), but it was found to be too sensitive to changes in height relative to the objective which was common during cycling when using the thin electrode architecture. The setup was then changed to a moiré interferometer to achieve a 122 more uniform fringe pattern with changes in height. Despite achieving more consistent measurements with changes in height and a decrease in static noise (Figure 4.1A and Figure 4.2B, respectively), in situ variability continued. Through Gaussian beam simulations, it was determined that this was due to sensitivity to tip and tilt of the sample. The sample would be aligned at the start of an experiment, but the change in tilt during the experiment caused a large change in erroneous perceived strain behavior making a systematic tilt correct difficult (Figure 4.1C). An auto collimator was built and used to make the sample perpendicular to the objective. However, this became infeasible with the foils in situ. Although the SEI strain can be quite large, on the order of 0.01, because the Si anode and Cu current collector are so much thicker and stiffer, the resultant strain observed on the back side of the Cu is much less: 10 ppm for 2,000 nm of Cu, or 35 ppm for 500 nm of Cu. While the 35 ppm strain is well within the ability of moiré to measure under normal circumstances, the fragility of the thin 500 nm Cu and the susceptibility of thin Cu to bending and bowing in an electrochemical cell made such measurements even more challenging. The thin foils, used to enhance strain resolution, were required because of the confinement of the Si from the Cu, but they also made it nearly impossible to maintain a sufficiently flat surface. Because of these limitations, it was determined that moiré interferometry would be difficult to use as an in situ method for measuring strain at the resolution required. To improve resolution, the foil could have been made even thinner or the grating could have been made even finer with e-beam lithography rather than photolithography. But, this would have likely decreased the signal to noise, particularly with changes in tip/tilt duringin situ measurements. Furthermore, although eventual measurement of in situ strain would likely be achievable, the deconvolution to understand the SEI strain would be difficult and even then, the strain of the SEI does not necessarily indicate mechanical failure of the SEI. For example, delamination of the SEI would not be captured through SEI strain alone. Instead of pursuing the balance between resolution and signal to noise with moiré interferometry, SECM was used as a direct measurement of SEI passivation failure. 123 4.2 SECM as a Direct Method to Track Changes in Silicon Passivation During Rest SECM can provide spatial information on the SEI passivation indicating local versus global failure. To prevent convolution between changes in current due to reactivity and height changes, silicon thin films were used. The silicon thin films were found to calendar age much faster than composite electrodes which was beneficial for studying aging over a few days which was on a time frame that the specially designed SECM cells could remain sealed. Figure 4.2 shows the rapid loss of charge (demonstrated by the increasing voltage) while at rest in a silicon thin film half cell after formation cycles. Cells were either analyzed after three C/5 formation cycles in the delithiated state at 1.5 V versus Li/Li+ (unless otherwise specified)or in the lithiated state at 100 mV versus Li/Li+ . Ferrocene in 100 mM propylene carbonate is used for all SECM experiments. It is important to consider how representative the Fc redox probe activity at a surface is of more relevant electrolyte components (EC, EMC, LiPF6 , etc.). SECM in feedback mode is measuring the ability of the surface to enable the redox reaction of Fc, this could be due to the SEI allowing electrons through to undergo charge transfer with Fc at the electrolyte-SEI interface. Another possibility would be for Fc to diffuse through the SEI and undergo charge transfer at the electrode-SEI interface. Either scenario is relevant to passivation failure of the SEI since the size of Fc is comparable or larger, than common electrolyte components such as EC, EMC, and LiPF6 . So, it is assumed that if the larger ferrocene molecule can undergo charge transfer (whether at the SEI or electrode interface) then this should be a good estimate of the limiting (worst case) passivation behavior towards actual electrolyte species. For both the delithiated and lithiated states in gen2 electrolyte, the passivation of the SEI decreased with rest time as shown in Figure 4.3 A-C and D-F, respectively. The scale bar shows dimensionless rate coefficient, κ (given by Equation 1 .21). The larger the value of κ, the larger the forwards reaction coefficient of the ferrocene redox reaction and the less passivated the surface is. The large images compare A-C and D-F on the same scale bars to track changes over time. The insets show a smaller scale to show detail for the earlier times with lower κ. Figure 4.3 A-C shows increasing rest time from 3 to 26 to 87 hours with corresponding average κ values of 0.039, 0.098, and 2.65, respectively. For A and B, 124 the surface led to negative feedback with itip,N less than 1. At 87 hours, the lithiated surface (now fully delithiated) started to exhibit positive feedback behavior with itip,N greater than 1, indicating an increase in reactivity. These trends were confirmed by doing substrate CVs at the end of an experiment and the general trend of passivated or nonpassivated behavior were consistent with the Fc redox peaks appearing on a nonpassivated surface, but not on a passivated surface (an example of this is shown for the lithiated case as a function of time in Figure 4.4). The surface of the thin film delithiated to 1.5 V was much more reactive than the lithiated surface at all times. This can easily be observed by looking at the difference in the κ scale bar for A-C and D-F. For the delithiated state (D-F) the data was collected at 7, 28, and 58 hours with corresponding average κ values of 1.53, 2.33, and 4.56. The κ values at short rest times are an order of magnitude smaller in the lithiated case (A-C) in comparison to the delithiated case (D-F). For all times, the delithiated surface exhibited enhanced feedback where itip,N was greater than 1. This would imply that the calendar aging of silicon-containing batteries would be worse in the discharged state of a full cell, but it is commonly accepted that lithium ion batteries age faster in the charged state. The top voltage cutoff for the SECM half cells was chosen because half cells are commonly cycled between 50 or 100 mV to 1.5 V in the literature. However, silicon anodes in full cells would realistically not go beyond about 600-750 mV. To investigate if the high potential of 1.5 V was causing damage to the SEI, the delithiation experiment was repeated with formation cycles performed between 750 and 100 mV, ending with delithiation to 750 mV (Figure 4.5). The delithiated silicon still appeared to be less passivated than in the lithiated state, but was much more protective than the SEI that went up to 1.5 V proving that 1.5 V is too high of a potential to maintain a passivating SEI. The silicon delthiated to 750 mV resulted in a more passivated surface with decreased reactivity when compared to the surface at 1.5 V. Hasa et al. [109] has observed the dissolution of LiEDC at 1.5 V which may relate to the poor passivation at this higher potential. This has implications for half cell cycling for proof of performance for new electrolytes and electrode materials and architectures. Because half cells have excess lithium inventory provided by a lithium metal counter electrode, an SEI that is destroyed each delithiation to 1.5 V, would not necessarily manifest as capacity fade because there is enough lithium to rebuild the SEI 125 each cycle, but the continual destruction of the SEI could lead to a different SEI than what would form under full cell conditions. This also has implications for prelithiation of silicon anodes because delithiating to a high potential may be detrimental to the SEI and negate the purpose of prelithiating to overcome initial irreversible losses due to SEI formation and lithiation of high-impedance sites within silicon anodes. Destruction of the SEI in this case just means the lost of its ability to passivate the surface, it does not necessarily mean that the SEI has been removed. To help better visualize the change in the surface over time, the images in Figure 4.3 were averaged and Equations 1.22 - 1.25 were used to calculate the probe approach curves for the average kappa. The resulting probe approach curves are shown in Figure 4.6. Figure 4.6A shows the change in the lithiated surface with rest time. The light to dark blue represents increasing rest time. The red dashed line shows the pristine silicon and the black lines show limiting behavior of highly conducting and insulating films for comparison. Although passivation decreases with rest time, the surface remains more passivated than the initial Si surface until 87 hours where the surface actually becomes more reactive than the initial surface. The potential has risen to 1.03 V at 87 hours and is fully delithiated. Figure 4.6B shows the same information for the change in the surface delithiated to 1.5 V. Light to dark orange represents increasing rest time. At the beginning of rest, the 1.5 V surface is already as reactive as the original silicon surface without any SEI present and then becomes more reactive with increasing time. Delithiating to only 750 mV greatly improves the average passivation of the silicon surface and shows decreased reactivity from that of the pristine silicon surface. The potential after 12 hours of rest of the electrode delithiated to 750 mV is 820 mV. The trend of decreasing passivation with increasing time and potential may not be as detrimental as poor passivation at low potentials because there is less of a driving force for reduction of the electrolyte. However, 600-750 mV (typical potentials at the silicon anode in a discharged full cell) [259] may still be sufficiently low to reduce common electrolyte components. EC and LiPF6 have calculated reduction potentials of 210 mV to 900 mV and 440mV - 590 mV versus Li/Li+ depending on electrolyte coordination and the electrode type [114, 260]. FEC, a common electrolyte additive for silicon anodes, has a calculated reduction potential of 800 mV - 900 mV depending on coordination [114]. Reduction will be energetically favorable when the electrode is at any 126 potential at or below the reduction potential. When aging at OCV, there are two variables that can affect the passivation of the electrodes. This includes time which can lead to degradation of the SEI by a variety of mechanisms; for example, dissolution/thickening or cracking of the SEI as the electrode loses lithium. The second variable is the change in potential as the electrode loses lithium which can also change the character of the SEI. Figure 4.7 shows the average κ and cell potential as a function of rest time. It is clear that the reactivity of the surface is a function of both rest time and potential since the trends in potential with time are similar to that of κ with time. The green data points show the surface delithiated to 750 mV and the κ value is lower than what would be predicted based on potential alone, so potential cannot be the only determining factor of the reactivity of the surface. The error bars are the standard deviation of the average kappa taken over ∼ 500 µm x µm area. This indicates that in both the lithiated and delithiated cases, the heterogeneity of passivation increases with rest time. Although there are “hot spots” of reactivity, the reactivity of the surface appears to be fairly global in that when looking at multiple spots across the electrode, they are pretty comparable. The increase in passivation heterogeneity does not manifest as distinct cracks, but rather fairly broad areas of faster kinetics. The change in passivation with the change in potential may be correlated to the “breathing” mechanism of the silicon SEI where the SEI becomes thicker as the silicon is delithiated and thinner as it lithiates. Furthermore, Stetson et al. observed a decrease in SEI resistivity with rest which they hypothesized was the dissolution of the outer organic layer of the SEI, exposing the more inorganic inner layer of the SEI [251]. This result is consistent with this work, although it is counter-intuitive that the SEI wouldn’t eventually stabilize and equilibrate during rest. The rapid change in potential during rest of the thin films means fairly rapid delithiation which correlates with volume changes which likely disturbs the SEI during OCV aging. To study the effect of fluorinated electrolyte components on aging, two other electrolytes were investigated including 0.7 M LiBOB in 3:7 EC: EMC (fluorine free, little to no HF formation) and gen2 (some HF formation) + 3% FEC (more fluorine, and possibly more HF formation). Figure 4.8 shows the comparison between the three after approximately one day of rest. According to these kappa values, the order of reactivity is LiBOB, average κ = 0.09 < gen2, average κ = 0.10 < gen2 + 3% FEC, average κ = 0.30 indicating that the 127 FEC electrolyte SEI is less passivating and would result in a worse calendar life. FEC is typically used with silicon because it enhances cycle life performance, but there has been evidence that the calendar life of electrolytes containing FEC can be worse than gen2 [102]. Kalaga et al. found that the capacity lost during a 600 hour voltage hold for gen2 versus gen2 + 10% FEC was 6 and 11 %, respectively. Despite demonstrating a greater capacity loss after the voltage hold, less charge was transferred during the voltage hold which does not agree with the greater capacity loss. The study also looked at the post aging electrolyte to quantify PO2 F2− anion using NMR. This byproduct can be used to deduce the amount of HF by Equations 1.2 - 1.4. It was found that the post aging electrolyte for the FEC electrolyte had 25% mole fraction of this degradation product while the similar cells that were cycled rather than calendar aged only had 6% mole fraction. The work proposed that the reason for the apparent contradiction in charge passed during the voltage hold and capacity lost during the hold was because a more stable SEI formed from the FEC electrolyte causing HF to build up in the electrolyte so when the cell was discharged, the silicon expanded and new surfaces became available leading to consumption of the HF all at once after the voltage hold resulting in a greater capacity loss. It was claimed that the rapid attack of a greater concentration of HF all at once caused more damage. However, it seems like the HF production should be self-limiting based on the water content of the electrolyte if the HF cannot etch SiO2 due to a protective SEI. If the SEI formed from a FEC electrolyte instead is not stable with time (but is stable with cycling) and more HF is allowed to interact with Si, etching any present SiO2 , more water could be created and lead to higher levels of hydrolysis products in the electrolyte. Since this is a chemical process rather than electrochemical, there would not be an increase in the charge passed during the voltage hold, but instead a decrease because of the loss of usable silicon. This would also correspond to a greater capacity loss after the voltage hold due to the loss of active material. Furthermore, one possible decomposition pathway of FEC in the electrolyte is hypothesized to form HF after de-fluorination [261–263]. FEC helps enable good cycle life, but since it may result in additional HF which will lead to predominately chemical (rather than electrochemical) attack of the electrode, it is reasonable that the addition of FEC to the electrolyte would decrease calendar life. FEC is thought to help form stable SEI layers; however, performance rapidly decays once the FEC is consumed, [111] meaning that the 128 SEI must be repaired by the FEC during cycling. Without this reparation to balance the detrimental effects of additional HF, calendar life may worsen. This may also be affected by the difference between the chemical decomposition of FEC in the electrolyte versus the electrochemical reduction to form the SEI. Another possible explanation for why the SEI formed from the FEC electrolyte appears to passivate poorly with time as compared to gen2 could be differences in adhesion between the silicon and the SEI formed from gen2 and gen2 + 3% FEC. Since the cycling configuration of the custom cell is a compressed cell stack similar to a coin cell, when the stack is removed for SECM characterization, the SEI could be mechanically removed. There are large portions of the FEC SEI that indicate highly reactive areas that are either partially removed (causing the tip to come in contact with the SEI leading to infinitely high κ values) or completely removed (Figure 4.9). Even in the case of Figure 4.8C, the increase in κ could be due to loosening of the SEI when the separator is removed, allowing more ferrocene to interact with the silicon surface. The large patches of increased reactivity are only consistently observed in the gen2 + 3 % FEC SEI and not in the gen2 or LiBOB formed SEIs. This patchy behavior could also be due to dissolution and damage to the SEI during calendar aging irrespective of the removal of the separator. It is not uncommon for electrodes to be cycled in a coin cell and disassembled for post-mortem analysis, poor adhesion of the SEI could cause incorrect conclusions on SEI composition and characteristics. Unfortunately, due to the high voltage plateau observed with a greater amount of electrolyte (see experimental section, Figure 2.13), the cell could not cycle galvanostatically without being in the cell stack configuration with a relatively small volume of electrolyte. 4.3 Conclusions Calendar aging of silicon is a function of both time and potential. The passivation of silicon decreases with increasing time and potential relative to Li/Li+ . Along with the decrease in passivation with time, the homogeneity of passivation decreases. Despite some local “hot spots” of reactivity, within resolution limits (tip radius of 12.5 µm), the changes in passivation seemed to be global SEI failure rather than local (for example, cracking). The one exception was the SEI formed with gen2+ 3% FEC where large areas of increased reactivity were observed among areas of relatively passivating SEI. 129 In all cases, the delithiated silicon was more reactive than the lithiated silicon. When delithiated up to 1.5 V versus Li/Li+ the surface was found to be more reactive than the pristine silicon surface. This could potentially be the result of destabilization of the surface by SEI oxidation. When the cell was only delithiated up to 0.75 V versus Li/Li+ , the surface was still passivating, but still less so than the lithiated surface. This indicates that the potential of the anode should be kept at or below ∼ 0.75 V for half cell and full cell cycling to prevent decreasing SEI passivation. Lastly, the SEI passivation decreased in order from LiBOB, gen2, and gen2 + 3 % FEC electrolytes. The poor SEI passivation with the FEC electrolyte may be explained by the domination of chemical degradation of both FEC (HF formation) and that the SEI cannot be repaired as it would be with cycling in the presence of FEC. Or the decrease in passivation could simply indicate a decrease in adhesion between the SEI formed with FEC and the silicon resulting in partial delamination of the SEI when the separator and cell stack are removed for SECM characterization. However, the fact that LiBOB and gen2 electrolytes did not do this, indicates that the FEC electrolyte may lead to poor SEI passivation and mechanical integrity when under calendar aging conditions. 130 A B C Figure 4.1. Summary of moiré results.(A) Improvement on z-insensitivity using interferometry, (B) improvements on noise using interferometry by increasing the number of averages per image analyzed and changing to a monochromatic CCD camera, and (C) Gaussian beam simulations of the effect of sample tilt. Similar trends were observed experimentally, but with a larger magnitude. (A) and (B) are compared against 35 ppm of strain previously estimated. 131 Figure 4.2. Example of rapid calendar aging of silicon thin films after three formation cycles. Aging demonstrated by the change in voltage during rest. Blue shows the cycling done before SECM for “delithiated cells” where the rest starts at 1.5 V. Orange shows the cycling done before SECM for “lithiated cells” where the rest starts at 100 mV. 132 300 1.0 200 7.2 0.0 150 4.8 100 2.4 50 3 hrs 0.23 V vs. Li/Li+ 9.6 7 hrs 0.98 V vs. Li/Li+ 0 0 0.0 Distance (μm) κ 0.20 415 2.0 9.6 1.0 332 7.2 0.0 249 4.8 166 0.0 0 0 2.4 28 hrs 1.52 V vs. Li/Li+ Distance (μm) C 415 83 26 hrs 0.34 V vs. Li/Li+ 332 0.00 F κ 12.0 9.6 332 7.2 249 4.8 166 Distance (μm) 415 332 249 0.0 0 0 2.4 58 hrs 2.61 V vs. Li/Li+ 166 83 83 Distance (μm) 415 87 hrs 1.03 V vs. Li/Li+ 12.0 3.0 249 0.07 Distance (μm) 0.13 166 κ κ E 83 B 415 Distance (μm) 0.00 2.0 250 12.0 3.0 332 0.13 0.07 κ 249 0.20 κ D 166 κ 83 A Figure 4.3. SECM results for silicon thin film with SEI formed with gen2 electrolyte. A-C shows a representative example of the change in κ for progressive rest times for the 100 mV lithiated state from 3 hrs to 26 hrs to 87 hrs. D-F shows similar SECM images for the 1.5 V delithiated state from 7 hrs to 28 hrs to 58 hrs. All large images have the same scale bar for comparison purposes. The insets show the same images at a smaller κ scale. 133 Figure 4.4. Example of how surfaces found to be passivated by SECM are confirmed to be passivating or nonpassivating when doing silicon substrate Cversus For passivating surfaces, the ferrocene redox peaks are not observed. The ferrocene redox peaks on a nonpassivating film are only evident in the first cycle indicating some kind of irreversible reactivity when at ferrocene redox potentials. This shows why it is useful to use SECM where only the probe needs to be polarized to oxidize ferrocene. κ 3.0 κ 0.6 50 0 Distance (μm) 415 332 249 166 83 0 0.0 249 1.2 166 0.6 83 0 0.0 Distance (μm) 415 1.2 100 1.8 0.0 332 150 2.4 0.3 332 166 1.8 0.6 83 200 Distance (μm) Distance (μm) 415 2.4 250 3.0 0.9 249 300 0 κ Figure 4.5. Comparison between delithiation to 1.5 V and 750 mV. The left image shows the same silicon surface as in Figure 4.3D where the thin film was cycled between 100 mV and 1.5 three times and then left to rest at 1.5 V for 3 hours. The image on the right shows a silicon thein film that cycled between 100 mV and 750 mV three times and then was left to rest at 750 mV for 12 hours. The large images have the same scale bar for comparison. The inset shows a smaller κ scale. Passivation is greatly improved by only cycling to 750 mV. 134 A B Figure 4.6. Calculated probe approch curves for the average κ values in Figure 4.3 using Equations 1.22 - 1.25. A) shows the lithiated case over time and B) shows the delithiated case over time. The black lines show the theoretical probe approach curve for a highly conductive or insulating film as labeled. 135 Figure 4.7. Average κ and potential values as a function of rest time for the lithiated and delithiated silicon cases. 136 0.54 390 0.43 312 0.22 100 0.52 0.39 234 0.26 156 0.11 78 0.00 0 0.13 0.00 0 415 332 249 166 83 0 Distance (μm) Distance (μm) C κ Distance (μm) 390 0.65 415 200 39 hrs 0.305 V vs. Li/Li+ 332 0.33 0 κ 249 300 B 166 Distance (μm) 400 26 hrs 0.34 V vs. Li/Li+ 0.65 83 κ Distance (μm) A 0.65 36 hrs 0.32 V vs. Li/Li+ 0.52 312 0.39 234 0.26 156 0.13 78 0 Distance (μm) 415 332 249 166 83 0 0.00 Figure 4.8. SECM images for a lithiated silicon thin film after being A) formed in gen2 electrolyte and rested for 26 hours, B) formed in 0.7 M LiBOB in 3:7 EC:EMC electrolyte and rested for 39 hours, and C) formed in gen2 + 3% FEC and rested for 36 hrs. For C the best performing (lowest κ) were chosen for comparison; however many areas showed large regions of high reactivity as shown in Figure 4.9. 137 κ Distance (μm) 415 20.0 16.0 332 12.0 249 8.0 166 4.0 83 0 Distance (μm) 415 332 249 166 83 0 0.0 Figure 4.9. Example of the high heterogeneity observed in the SEI formed with gen2 + 3% FEC electrolyte. The white points are “invalid” points where the kappa was infinitely large. This could be due to the probe coming in contact with the SEI if it wasn’t well adhered and fairly conductive. This would mean that the surface had greatly varying height differences. Or these regions could just be highly reactive. CHAPTER 5 MECHANICAL IMPACTS FROM CYCLING ON CALENDAR AGING MEASUREMENTS Calendar aging is typically measured by long periods of OCV that are intermittently interrupted with an RPT to quantify performance and capacity fade. The USABC protocol is to do an RPT once a month with daily voltage pulses to keep the state of charge the same during the rest. If the SEI is unstable and there is substantial volumetric changes during cycling, as is the case with silicon, the frequency and amount of cycling interruption may impact how calendar life is quantified. It was hypothesized that an SEI that equilibrates during a rest may be disrupted enough during RPT cycling that the SEI passivation is decreased and the start of the next rest will result in greater irreversible lithium inventory consumption to rebuild the SEI that was lost during cycling. To test this, the protocol in Figure 2.6 was applied to a variety of silicon and graphite electrodes (Table 2.1) against NMC622. First, the longer rest periods of 1, 2, and 3 months were used for the RPT intervals. The voltage change during the first 3 month rest for the different electrodes is shown is Figure 5.1. A more rapid voltage drop is observed the greater the silicon content consistent with the discharge voltage profile for the given material. The test was about 8 months long and by the end of the test, all of the cells had approximately had the same total amount of rest (but at different intervals) and number of cycles. The test matrix was repeated by collaborators at Argonne National Laboratory and National Renewable Energy Laboratory with comparable results. Figure 5.2A-D shows capacity retention (relative to the third formation cycle) as a function of time for the graphite-2, 10% Si, 15% Si-Gr, and 80% Si. Figure 5.2E-H shows the same capacity retention data, but as a function of cycle number. If cycling was greatly impacting the capacity fade, all the 1, 2, and 3 month cells should overlap on the cycle number graph because only 139 the number of cycles determines the capacity fade. However, the opposite is found here where, accounting for noise, the data overlaps as a function of time indicating the time since assembly determines the capacity fade. For the same number of cycles, but different aging times, the greater aged cells have lower capacities while cells that have been aged the same amount, but cycled differently, have the same capacity loss. This results in all of the cells with a given anode ending at a comparable capacity despite differences in RPT frequency. In Figure 5.2E-H, the jump in cycle number from 15 to 30 represents constant cycling at the end of the OCV-RPT sequence, in all cases the capacity changes minimally during those 15 cycles, further demonstrating that the capacity fade in these cells was driven by time rather than cycling. The pure cycling condition, which was performed for the graphite-2 and 80% silicon, also resulted in substantially lower capacity fade for the same number of cycles. However, the trend of temporal degradation dominating over mechanical was true in both the silicon and the graphite-2 control cells, so this wasn’t a silicon-specific behavior. Silicon does differ from the graphite-2 cells though because the trends are not nearly as clean and clear in the silicon cells suggesting there are multiple degradation pathways taking place. Both the electrolyte and NMC622 cathode were the same between all anode chemistries. The graphite capacity fade is much greater than expected which may be due to stainless steel corrosion on the cathode side or the use of a non graphite optimized electrolyte. It was hypothesized that a longer rest may lead to processes with slow kinetics dominating degradation if given sufficient time to occur. While on shorter time scales, cycling would dominate. The same experiment was repeated with 1, 2, and 4 week rest periods between RPTs for the 80% Si and graphite-2 electrodes to see if the trends observed in Figure 5.2 would change. Figure 5.3 shows the first 5 months of variable OCV-RPT cycling with the shorter rest periods. Similar trends as the longer rests can be observed for both the 80% silicon and graphite-2. For the 80% silicon, the cycling does seem to play a larger role with the shorter rest periods, but time since assembly still appears to dominate capacity fade. The light blue and black curves in Figure 5.2B tend to have a lower capacity fade than the dark blue or red curves with no rest and 1 week of rest between RPTS, respectively. With the exception of the pure cycling cells, the difference isn’t substantial when accounting for cell to cell variability. Figure 5.2D shows a smaller difference between 140 the capacity fade for the same number of cycles for the shorter rest periods, indicating cycling does start to influence degradation. Coulombic efficiency can be a helpful tool for evaluating lithium inventory losses, including losses to SEI formation. The coulombic efficiency of the 1, 2, and 4 week repeating rest periods is shown in Figure 5.4. For the 80% silicon cells (Figure 5.4A), the coulombic efficiency drops during the rest. In the normal 0.1 C cycles following the rest, the coulombic efficiency recovers, by the second cycle, the coulombic efficiency is at or slightly above the coulombic efficiency of the pure cycling cell. For graphite-2 (Figure 5.4B), the coulombic efficiencies drop during the rest, but then recover quickly during the RPT to values similar to the pure cycling case. There isn’t a lag before reaching the pre-rest coulombic efficiency, the coulombic efficiency of the first full cycle after the rest goes back up close to the pure cycling coulombic efficiency. There are a couple possibilities for the differences between graphite-2 and silicon. First, the rest may allow for more homogenous lithiation of the silicon which then may make removal of lithium more difficult leading to lower coulombic efficiency after the rest, but then above 100% on the second RPT cycle when some of that lithium is recaptured after cycling allows it to diffuse and redistribute. Secondly, if the SEI is mechanically damaged during the RPT cycles, lower coulombic efficiencies would occur until the SEI restabilized. Although both the graphite-2 and silicon cells have a large decrease in coulombic efficiency during the rest step, the loss decreases in the graphite-2 case, but remains constant in the silicon case. This means that the mechanism for this behavior is likely different for the two electrodes since one seems to stabilize while the other causes ongoing degradation. The decrease in the coulombic efficiency for the rest does trend with the length of the rest, meaning the longest rest resulted in the greatest drop in coulombic efficiency. To try and understand what causes the delayed recovery in coulombic efficiency of the silicon cells after a rest, the dQ/dV plots are evaluated for the cycles surrounding a rest (Figure 5.5). The blue curve depicts the 3rd formation cycle which shows the starting behavior of the cell. The orange curve shows the lithiation and delithiation before and after the fifth rest, respectively. The grey and yellow curves show the cycles of the RPT after the rest. Since these are full cells, the dQ/dV peaks are a convolution of the cathode and anode processes. For graphite2-NMC cells, both charge and discharge peaks before 141 and after the rest, as well as the RPT cycles are fairly consistent. This is especially true for the large peak at ∼ 3.5 V which is attributed to graphite (de)lithiation. Conversely, the one major peak in the silicon charge curves changes for these cycles. The change in peaks occurs only on lithiation of the silicon and not delithiation. Since the discharge peak remains the same, the change in coulombic efficiency observed in Figure 5.4A is caused by the decreasing ability to lithiate the silicon rather than delithiate the silicon. There is a slight shift in peak position which may indicate lithium inventory loss, but there is also a decrease in peak intensity suggesting active material loss although the contributions from the anode and cathode are convoluted due to the full cell configuration. Figure 5.6 shows the dQ/dV plots of graphite-2- and silicon- NMC cells at cycle 18 (AB) and at approximately 700 hours (C-D) for the cells with varying rest intervals (1, 2, and 4 weeks and pure 0.1 C cycling). For graphite, the graphite peak, indicated by an asterisk, shifts when calendar aged, but not when cycled to the same cycle number (Figure 5.6A) or same time (Figure 5.6C). The graphite peak shifts more when it has a longer rest interval, but the same number of cycles, but when aged for the same amount of time, but cycled different amounts, the peaks are the same. This confirms the trends seen in Figure 5.2A and E and Figure 5.3A and C which clearly indicate that time dominates degradation of the graphite-2 rather than the number of cycles. Another interesting observation from the graphite-2 data is the change in the NMC peak (shown by a circle), the peak changes (a slight shift to higher potentials and a decrease in peak intensity) are different for the pure cycling case over resting cases regardless of being sampled at the same cycle number or time. Because the setup is a two electrode cell, the exact cause of these peak shifts cannot be identified. Although the peak at lower potentials is due to graphite and the higher potential peak is due to NMC, the peak shifts can be caused by difficulties lithiating and delithiating either electrode. Either way, the different behavior between the constant cycling and calendar aging case indicates that there are different degradation pathways for the graphite-NMC cells when cycling constantly versus cycling with intermittent rest. This suggests that performance, including capacity and coulombic efficiency, while cycling constantly will not necessarily act as an indicator as to whether or not a cell will have good calendar life as well. The sharp features of the graphite and NMC make observing these changes easier. 142 The trends for silicon-NMC are not as straight forward. Figure 5.6B shows silicon for the different rest intervals after 18 cycles. The peaks are similar between all cases except in the longest rest interval shown by light blue. So a longer rest does lead to more degradation. Similarly Figure 5.6D shows silicon for the different rest intervals at the similar time of ∼ 700 hours. In this case, all the peaks are similar on the lithiation side with the exception of pure cycling, which shows that the other extreme of pure cycling also is detrimental. The delithiation peak amplitude trends with the amount of cycling, with the cells that were cycled more within the 700 hour period displaying more degradation. Because of the poor cycling capability of this particular silicon electrode, the determination of the impact of SEI disruption on the quantification of calendar life is convoluted. The degradation is definitely a function of both time and the amount of cycling. The longer the rest periods between RPTs, the more likely time since assembly will dominate the degradation rather than how often the SEI is disrupted by cycling. Repeating this experiment on a silicon electrode that achieves a higher coulombic efficiency when cycling, closer to that of graphite, may more clearly indicate the impact of intermittent cycling or at least minimize the impact as shown with graphite. The performance of the graphite control cell is also worse than expected. This is likely because the electrolyte isn’t optimized for graphite. The cell form factor of coin cells rather than pouch cells may also impact performance due to corrosion of the steel. Therefore, even for graphite, only general trends rather than quantitative fade should be considered in this dissertation. Since the calendar aging studies showed a complex interplay between aging time and disruption of the SEI that forms with silicon with RPTs, the heat generation from the same 80% silicon electrode was studied using isothermal microcalorimetry to track changes in heat generation after an RPT. Measurements were taken at OCV, so any heat generation should be dominated by parasitic processes. To simplify the system, the 80% silicon was initially paired with LFP which should be more stable than NMC. The initial electrolyte was gen2. Figure 5.7B shows three data points which represent the average heat generation after 3 formation cycles, after ∼ 3 weeks of OCV rest and before an RPT, and after an RPT. The heat generation decreases slightly during the rest, but then increases to more than the starting heat generation after the RPT. This suggests that the RPT does sufficiently disturb the SEI to set back passivation. The increase in heat generation may also indicate a build-up of a 143 reactive compound (for example, HF) that can only react with the electrode once the SEI is mechanically broken during the RPT. The coulombic efficiency (Figure 5.7A) does decrease after the rest, but recovers by the end of the RPT to a value similar to the end of formation, so that alone can’t explain the increase in heat generation. In one of the sample sets there was a slight increase in capacity during formation which can indicate wetting issues, especially in the thick LFP cathode that was used. However, both sample sets showed the increase in heat generation regardless of having this behavior during formation. The impedance was taken prior to each calorimetry measurement and is shown in Figure 5.8. Due to OCV aging, there is a change in potential, so the EIS was taken at varying potentials, so trends are difficult to track for the pre-RPT curve compared to postformation and post-RPT. The postformation and post-RPT are at similar potentials and thus are a fair comparison. The first semicircle increases in size indicating an increase in capacitance and resistance of the SEI which further suggests the SEI has reformed by the end of the RPT when assuming the model from Figure 1.16. Next, the same experiment was repeated for Gr-1 -NMC622 and 80% Si - NMC622 cells with gen2 + 3% FEC electrolyte, to see if the observed behavior was just a function of LFP. Figure 5.9 shows the comparison between all of the chemistries studied. Figure 5.9A and B show the cycling behavior for the 80% Si - NMC622 and the graphite-2 - NMC622 cells, respectively. Figure 5.9C shows the change in heat generation for all three systems. The 80% silicon - LFP has a lower heat rate which is expected since the LFP is at a lower potential and is generally more stable than NMC622. For the NMC622-silicon cells, the heat generation stays approximately the same as after formation. The same increase in heat generation to a value larger than the initial postformation value is observed after the RPT, suggesting the behavior is real and likely due to the silicon, irrespective of the cathode. The graphite-2 - NMC622 cell heat generation decreases during the OCV rest. It does increase in heat generation after the RPT, suggesting it is at least partially affected by SEI disruptions and/or build up of reactive components in the electrolyte. However, the postRPT heat generation is less than after formation meaning graphite is still becoming more passivating with time, unlike the silicon cell. This agrees with the shorter variable OCVRPT tests of 1, 2, and 3 week OCV rests where cycling did seem to impact degradation. If this experiment was repeated for a longer rest period, the cycling would still likely show 144 an impact to the SEI, but the overall degradation would probably start to be dominated by time since assembly rather than the number of times the SEI was disrupted by RPT cycling. 5.1 Conclusions When compared to a graphite control, the 80% silicon electrode is more greatly affected by cycling at lower rest times. However, as the amount of time between RPTs increases, time since assembly dominates degradation in both the silicon and graphite. The poor performance of the graphite is likely due to an electrolyte that isn’t optimized for graphite and may also point towards the need to do calendar aging in pouch cells to decrease timedependent reactivity with the cell packaging. When graphite is calendar aged, the dQ/dV plots indicate that there is possibly an irreversible loss of lithium inventory (causing a shift in peaks) that does not occur when the cell is cycled nonstop for the same amount of time. This may also point towards incompatibility with the electrolyte and/or the packaging material at rest times. When comparing the dQ/dV plots between graphite and the 80% silicon NMC622 cells, the loss for silicon seems to be dominated by active material loss rather than lithium inventory loss. Zilberman et al. also observed active material loss on low silicon loading anodes dominating capacity fade during storage [96]. The silicon degrades as a function of both cycle number and time since assembly (changes in peak intensity at the extremes of 4 weeks of rest and constant cycling) whereas the graphite peaks all overlap more at constant time than at the constant cycle number(Figure 5.6A and C) indicating degradation dominated by time. Since the 80% silicon electrode used in this study already has poor cycling, it may be expected that it would also have poor calendar life. So, repeating these experiments with a higher performing electrode may be useful. Despite poor cycling, at longer rest times, time since assembly starts to dominate when it would be expected that poor cycling would disrupt the SEI the same for each RPT. This suggests that cycle life is not necessarily an indicator for calendar aging. This supports why there is a large technical gap between the calendar life and cycle life of commercial cells (Figure 5.6). The calorimetry results support that the SEI is indeed disrupted during RPTs and thus protocols testing calendar life must take into account the effect of measuring capacity 145 fade (performing an RPT) will have on the perceived calendar life for silicon cells. The calorimetry results also point towards the need for a better chemical understanding of what occurs during rest since the increase in heat generation after an RPT may be due to changes in electrolyte composition during rest. 146 Figure 5.1. Change in potential during 3 month rest for the 4 anodes tested: Gr-2, 80% Si, 15% Si-Gr, and 10% Si. The cathode is NMC622 and the electrolyte is gen2 + 10 % FEC. 147 A B Graphite 10% Si C E F G 15% Si-Gr D H 80% Si Figure 5.2. Change in capacity retention as a function of time for A) graphite-2, B) 10% Si, C) 15% Si-Gr, and D) 80% Si. Within the noise, all curves behave approximately the same. For capacity retention as a function of cycle number for E) graphite-2, F) 10% Si, G) 15% Si-Gr, and H) 80% Si, there is a clear difference between the different rest intervals. Red is 1 month rest between RPTs, black is 2 months between RPTs, and light blue is 3 months between RPTs. For the same number of cycles, but different lengths of rest intervals in between, the cells with the greatest rest have the greatest amount of capacity fade. For 80% silicon and graphite-2, the pure cycling case is shown in dark blue. 148 Figure 5.3. Graphite-2 and 80% Si with shorter rest times of 1, 2, and 4 weeks as a function of time in A) and B) and cycle number in C) and D). Graphite-2 still overlaps on the time plot showing the low influence of cycling. Silicon starts to show cycling is having more of an impact that in the longer 1, 2, and 3 month rest case. The cathode in both cases is NMC622 with gen2 + 3% FEC. 149 A 100% Coulombic Efficiency 95% 90% 85% 80% Rest 75% Rest 70% 65% 60% 0 5 B 100% 10 Cycle Number 15 20 Coulombic Efficiency (%) 95% 90% 85% Rest 80% Rest 75% 70% 65% 60% 0 5 10 Cycle Number 15 20 Figure 5.4. Coulombic efficiency of A) 80% silicon and B) graphite-2 during variable OCV-RPT testing. Cycles that had a rest between charge and discharge are indicated. The cycles between the rests are the RPT cycles. The first three cycles were formation cycles. The insets show the protocol and corresponding cycles to the given coulombic efficiency data. 150 A Post Formaon 5th OCV 5th OCV Cycle 1 5th OCV Cycle 2 7 5 dQ/dV 3 1 -1 3 3.2 3.4 3.6 3.8 4 4.2 -3 -5 -7 B Post Formaon Full Cell Voltage 5th OCV 5th OCV Cycle 1 5th OCV Cycle 2 5 4 3 dQ/dV 2 1 0 -1 3 3.2 3.4 3.6 3.8 4 4.2 -2 -3 -4 Full Cell Voltage Figure 5.5. dQ/dV plots of A) graphite and B) 80% silicon for the cycles surrounding the 5th OCV rest. 151 7 dQ/dV 7 5 3 3 1 1 -1 3 3.2 3.4 3.6 3.8 4 dQ/dV 5 -3 -5 -5 D4 3 3 2 2 1 1 -1 3 3.2 3.4 3.6 3.8 4 0 -1 -2 -2 -3 -3 -4 Full Cell Voltage 3.2 -7 Full Cell Voltage 4 0 * -1 3 -3 -7 B C o * dQ/dV dQ/dV A -4 3.4 o 3.6 3.8 4 3.8 4 Full Cell Voltage 3 3.2 3.4 3.6 Full Cell Voltage Figure 5.6. dQ/dV plots of A) graphite2-NMC622, and B) 80% silicon-NMC622 cells at cycle 18 and dQ/dV plots of C) graphite2-NMC622, and D) 80% silicon-NMC622 at approximately 700 hours. In A) and C), * indicates the peak due to graphite and o indicates the peak due to NMC. 152 A B Figure 5.7. Coulombic efficiency, delithiation capacity, and heat generation of the 80%Si-LFP cells. A) Coulombic efficiency and delithiation capacity of the 80 % Si - LFP cells input into the calorimeter. The stars indicate when calorimetry measurements were taken. Postformation was after three formation cycles, pre-RPT was after 21-40 days of rest and before the first RPT, and the post-RPT was taken after the three RPT cycles. B) Shows the corresponding heat generation rates for the three data points, again indicated by stars. Each calorimetry measurement is the average of four cells. (Ω) 153 Post Formaon, 3.24 V Pre RPT, 3.15 V Post RPT, 3. 25 V (Ω) Figure 5.8. Representative electrochemical impedance spectroscopy of 80%Si-LFP at the postformation, pre-RPT, and post-RPT conditions. 154 A B Heat Generaon (μW/mAh) C LFP-80% Si NMC622-80% Si NMC622-Gr 3.5 3 2.5 2 1.5 1 0.5 0 0 5 10 15 20 Time (days) 25 30 35 Figure 5.9. Coulombic efficiency and delithiation capacity of A) the 80 % Si - NMC622 and B) the Gr - NMC622 cells. The stars indicate when calorimetry measurements were taken. Postformation was after three cycles, pre-RPT was after 21 days of rest and before the first RPT, and the post-RPT was taken after the three RPT cycles. C) Shows the corresponding heat generation rates for 80%Si-NMC622, Gr-2-NMC622, and 80%Si-LFP cells. CHAPTER 6 SUMMARY AND CONCLUSIONS This dissertation studied the impact of SEI mechanics on calendar aging. In Chapter 3, the development of potentiostatic holds as a rapid qualitative stage gate for calendar aging was discussed. Decreasing the time to evaluate calendar aging will greatly impact the battery field, especially silicon anodes, since the major hurdle left for commercial cells appears to be poor calendar life (Figure 1.1). The additional difficulties silicon creates for potentiostatic holds was discussed and an outline of best practices for potentiostatic experiments was provided. The capacity loss predicted by the voltage hold was consistently greater than the actual capacity fade and the deviation from the actual capacity loss changed with time and SOC. However, electrolyte studies and temperature studies (Figure 3.11) showed the correct qualitative trends. Thus, it was concluded that the protocol, at least in its current form, is not quantitative or semiquantitative, but is qualitative. Turning to more traditional OCV aging, the origin of SEI mechanical failure was investigated. Moire interferometry was first used in an attempt to quantify SEI strain but was discounted due to sensitivity to tip/tilt of the thin film foil electrodes that were necessary to improve resolution. Furthermore, this technique wouldn’t directly measure failure of the SEI. SECM was instead used to directly investigate the change in passivation of silicon thin film during OCV rest (Chapter 4) to determine if the SEI fails locally (cracking) or globally (SEI pore stretching, SEI dissolution, etc.). It was found that passivation decreases with increasing rest time and that passivation is better under lithiated conditions where rest started at 100 mV versus Li/Li+ . When delithiated to 1.5 V versus Li/Li+ , the surface was completely nonpassivating, possibly due to oxidation of the SEI. If the thin film was only delithiated to 0.75 V versus Li/Li+ , the surface was still less passivated than the lithiated state but was much more passivated than the 1.5 V Li/Li+ delithiated surface. In all cases, at the resolution limit of ∼ 12.5 µm, the failure of the SEI was global. Although 156 there were some local “hot spots” of reactivity, the surface changes were fairly large scale across the surface. The effect of fluorine in the electrolyte (and likely HF concentration) was investigated by looking at passivation after ∼ 1 day of rest in a LiBOB, gen2, and gen2 + 3% FEC. The passivation was found to decrease in order from LiBOB > gen2 > gen2 + 3% FEC. The FEC- containing electrolyte seemed to create a poorly passivated surface with rest, with large regions of high reactivity. This could have been due to poor adhesion to the silicon surface and removal of the SEI with removal of the separator after cycling or it could indicate that FEC, although beneficial for cycling, is not beneficial for calendar aging of silicon. The decrease in passivation with time observed with SECM is counterintuitive. For example, from Chapter 3, during a voltage hold, after lithiation relaxes, the current decreases as the surface stabilizes. This would imply that the surface is becoming more passivated. If passivation decreases with time, then we would expect an increase in heat generation during rest. Although the heat generation decreased slightly in the 80% Si - LFP cells (Figure 5.7B), in one of the replicates of the 80%Si - NMC622 cells (Figure 5.9B), the heat increased with time. None of the silicon-containing cells decreased in heat as greatly as the graphite cell. So, the calorimetry results may actually agree with the SECM data, especially considering the increase in complexity with the composite electrodes (addition of carbon and binder) in the calorimetry experiments. Either way, the instability and continued reactivity of silicon-containing systems demonstrates the need for more studies on the calendar aging of silicon-containing batteries. Chapter 5 investigated the impact of RPT frequency on mechanically disrupting the SEI during the calendar aging of silicon anodes. At shorter time scales, the number of cycles does seem to impact the capacity fade of the cells, confirming the mechanical disruption of the SEI. However, at longer time periods (1, 2, and 3 months), time since assembly starts to dominate capacity fade. Overall, this dissertation has demonstrated the unique challenges that the integration of silicon into next generation lithium ion batteries poses for reaching sufficient calendar life, specifically how the mechanics of the SEI are relevant during both cycle and calendar aging. The qualitative potentiostatic stage gate provides the community with a relatively quick method for determining if a material should undergo long-term calendar aging, 157 assuming the protocol is carefully followed. This research will impact the battery research community by describing how calendar aging of silicon-containing electrodes is different than graphite and thus may require modified protocols to be studied effectively. Lastly, the SECM study indicates that silicon does not stabilize with time, giving credence to the need for more studies to understand why silicon has poor calendar life. Furthermore, the decreased passivation observed at high potentials of 1.5 V versus Li/Li+ shows that half cells should not be operated above about 0.75 V versus Li/Li+ . Future work that can build off this dissertation include more studies to determine at exactly what potential silicon goes from passivated to nonpassivated. This could motivate voltage cutoff values for full cells to improve calendar aging. This work also helps motivate issues with SEI mechanics with calendar aging, but there is also clearly major chemical degradation during rest that needs to be studied more carefully to fully understand the poor silicon calendar life. APPENDIX CHAPTER 3 VOLTAGE HOLD SUPPLEMENTAL This section was adapted with permission from a published manuscript: M. C. Schulze, M.T.F. Rodrigues, J.D. McBrayer, et al., “Critical Evaluation of Potentiostatic Holds as Accelerated Predictors of Capacity Fade during Calendar Aging,” Journal of The Electrochemical Society, vol. 169, Art. no. 5, 2022, doi: 10.1149/1945-7111/ac6f88. A.1 Relationship Between Measured and Parasitic Currents The thought experiments below assume that voltage and current can be measured with perfect accuracy and time resolution. A discussion about the effect of imperfect sensitivity is included later in the text. To simplify the interpretation of the phenomena, all quantities are represented by their magnitude, and signs are explicitly introduced in the equations to make the direction of change clear. Thus, q ≡ |q| for both oxidation and reduction processes, and lithiation or delithiation of the anode will change the potential by -∆V or +∆V, respectively. Consider the zoomed-in view of electrode profiles for hypothetical positive and negative electrodes in Figure A.1A. The instantaneous value of the cell voltage V is given by the difference between the instantaneous potentials of each electrode: Vcell = Vpos + Vneg (A.1) During a potentiostatic hold experiment, the cell voltage remains fixed at the value Vhold but electrode potentials can vary depending on processes occurring in the cell. For example, electron losses at the negative electrode due to reduction processes at the SEI could lead to an increase in anode potential, due to a decrease in electrode SOC (i.e., Li+ is lost). 159 Similarly, extraction of electrons from the cathode (and transference to the anode through the external circuit) could cause an increase in cathode potential due to an increase in SOC. Consider the simple case in which electron losses to the SEI is the only parasitic process. Assume that the cell is in equilibrium at the voltage where the potentiostatic hold is being carried out, with cathode and anode at the initial potentials V( pos,i) and V(neg,i) , respectively. When reduction reactions consume an amount of charge qneg at the SEI, the anode selfdischarges and its instantaneous potential increases as: Vneg = Vneg,i + [qneg x (qneg x (| dVneg |y )] = Vneg,i + (qneg xδneg ) dQneg (A.2) where δneg is the magnitude of the local slope of the voltage profile of the anode at its instantaneous SOC y. When that happens, the cell voltage becomes smaller than Vhold , as the cathode potential remains constant while the anode potential increases by (qneg δneg ). Thus, electrons must be transferred from the cathode to the anode through the external circuit to resume the condition Vcell = Vhold . When an amount of charge q pos is transferred to the anode, the instantaneous electrode potentials become: Vneg = Vneg,i + (qneg δneg ) − (q pos δneg ) (A.3) Vpos = Vpos,i + (q pos δpos ) (A.4) Note that q pos does not need to be equal to qneg , and that the amount of charge being transferred must be such that the cumulative changes to the electrode potentials lead to Vcell = Vhold being met, even though the final and initial values of Vneg and Vpos may differ. Then, we have: Vcell = Vpos,i + (q pos δpos ) − [Vneg,i + (qneg δneg ) − (q pos δneg )] = Vpos,i + (q pos δpos ) − Vneg,i − (qneg δneg ) + (q pos δneg′ ) (A.5) = (Vpos,i − Vneg,i ) + (q pos δpos ) + [(q pos − qneg )δneg ] = Vhold + (q pos δpos ) + [(q pos − qneg )δneg ] = Vhold 160 q pos δpos = (qneg − q pos )δneg (qneg − q pos ) δpos = δneg q pos qneg = −1 q pos (A.6) qneg δpos = +1 q pos δneg (A.7) q pos R = qneg ( R + 1) (A.8) Making R ≡ δneg /δpos , we get: Since qneg is the charge lost to the SEI and q pos is the measured charge flowing through the external circuit, it can be rewritten as: qhold R = qSEI ( R + 1) (A.9) If we assume that the relationship expressed by the equation above remains valid for all times t and considering that q≡q(t), we can take derivatives with respect to time to arrive at an expression correlating the measured and actual parasitic currents: ihold = iSEI ( R ) → ihold = iSEI κ R R+1 (A.10) The consequence of this relationship is that, in theory, the current measured during the potentiostatic hold could quantitatively describe the rate of parasitic reactions at the SEI when κ R approaches unity. That is the case for R ≫ 1, when the local slope of the cathode profile is much smaller than the magnitude of the local slope of the anode. Figure A.2 shows that κ R grows very quickly with R and that the offset between ihold andiSEI becomes 161 smaller than 1% for R≥99. This condition can be fulfilled when the slope of the voltage profile of the positive electrode is very small (such as in LiFePO4 at intermediate SOCs) and when that of the anode is not negligible (such as in the single-phase regions of Li-graphite compounds, away from the plateaus, and in many SOCs of silicon anodes). Conversely, large values of R are less likely to be achieved when cathodes with sloped profiles are used, such as NMCs. The phenomenology behind the relationship expressed by Equation A.10 becomes clearer if we consider an example in which R=1, that is, when δpos = δneg , for which κ R =0.5. In that case, Equation A.10 indicates that half of the reduction reactions will not produce measurable current during the voltage hold. Figure A.1 attempts to illustrate the electron transfers during a voltage hold, assuming that SEI-forming reactions at the anode are the only parasitic processes in the cell. Filled circles represent electrons in the cathode (blue) and anode (orange) and are assumed to be traceable with perfect time resolution. When a reduction side-reaction consumes an electron from the negative electrode, the anode potential increases, causing the cell voltage to drop below Vhold (Figure A.1B). Consequently, an electron is transferred from the cathode to the anode through the external circuit (thus being measured), replenishing the anode SOC but increasing the cathode potential; the electron transfer actually makes Vcell > Vhold (Figure A.1C). The next time a reduction reaction knocks an electron out of the anode, the resulting rise in anode potential will restore Vcell = Vhold (Figure A.1D), prompting no measurable current flow through the external circuit in response to the parasitic process. The dynamic shown in Figure A.1 will repeat for as long as the slopes of both voltage profiles remain unchanged, in agreement with κ R =0.5. Whenever there is a net increase in anode potential (and decrease in SOC) during the hold, more electrons will have been lost to side-reactions then replenished through the external circuit, causing a mismatch between ihold and iSEI . Equation A.10 indicates how much the measured current underestimates the rate of side-reactions. For completeness, we note that when the rate of oxidation at the cathode is much higher than that of reduction at the anode, the sensitivity of the voltage hold for oxidation is improved by lower values of R: 162 ihold = ioxi ( R ) R+1 (A.11) This has been explored in previous publications, in which electrolyte oxidation at high voltage in NMC-based cathodes is measured against “flat” anodes (LTO or a voltage plateau of graphite) [97, 224, 239]. A.2 The Inverse Polarization Experiment The manuscript describes how the inverse polarization experiment can be used to estimate how long is needed before parasitic processes become the main contributors to the currents measured during the voltage hold. Here we include additional details about this test and discuss how experimental variables can affect the outcome. The experiment uses an anode half-cell and generally consists of three parts: i) an initial lithiation (which can involve voltage holds followed by rest at open circuit for equilibration/depolarization); ii) a partial delithiation (which could be either voltage- or capacitylimited); iii) a voltage hold at the final state of step ii. As discussed below, the test involves a change of sign in measured current during step iii, and thus must be carried out using a potentiostat (as cyclers typically operate strictly at positive or negative currents within a given step). At the beginning of step iii, the cell has been partially delithiated and is being held at its initial open circuit voltage. As the voltage hold initiates, the cell depolarizes and continues to delithiate. Over time, the delithiation current decreases (Figure A.3A), similar to observed for Irev of full-cells, as described in the main manuscript. Note that delithiation currents are measured as “positive” values by the potentiostat. Consider now that, at all times, parasitic reduction reactions are happening at the anode, with rate ISEI (Iirrev in the main text). These reactions consume electrons from the anode, requiring additional electrons to be injected into the electrode to maintain a constant voltage. This lithiation current is measured as a “negative” value by the potentiostat. By holding the cell voltage constant and observing the time needed for the current to switch from positive to negative, we can estimate how long it takes for parasitic processes to dominate the measured currents. The current measured at any point of step iii is the difference between the magnitudes of the residual delithiation current (Idelit ) and ISEI . Figure A.3A shows schematically that, early during the hold, delithiation currents are much larger than the parasitic ones. At 163 some point, however, both currents become equal, resulting in the measured current being zero. As discussed in the main manuscript, the reversible currents appear to vanish after sufficiently long holds for most materials. When that happens, Idelit ∼ 0 and the measured current is approximately equal to ISEI (the minimum in Figure A.3A). From this point onwards, the magnitude of the current should decrease, as parasitic processes tend to be self-limiting. The reason why these experiments can be useful is that by determining when ISEI ≫ Idelit , we can identify how long voltage holds must be before they start yielding the desired information. The current decay during the inverse polarization test of a Gr-1 half-cell is shown in Figure A.3B (test details can be found in figure caption). The current rapidly relaxes within the first few tens of hours, switches signs before reaching a minimum, then decays towards zero magnitude, as illustrated in Figure A.3A. When an 80% Si electrode is used, however, the current does not become negative until hundreds of hours into the experiment, and a minimum is never reached, indicating sluggish relaxation kinetics. The observation of prompt relaxation of graphite and slow kinetics of silicon agrees with Figure 3.5 and highlights how the inverse polarization can guide the design of voltage hold tests. Nevertheless, inverse polarization tests will only be useful if properly planned. Figure A.4A and B show the current measured during the inverse polarization of 15% Si-Gr electrodes versus Li. Both were first lithiated to 110 mV and held at that voltage for 2 hours. They were then delithiated to different potentials and held at this new potential for 720 hours. The tests in panels a and b are for holds at 150 and 300 mV, respectively. The results are markedly different: while Figure A.4A appears to display nearly immediate relaxation, Figure A.4B shows a coexistence between “positive” and “negative” currents, thus indicating a lack of clear-cut relaxation features. This difference is an artifact caused by the improper choice of hold potential in the former. At 150 mV, the hysteresis of the Si component of the electrode makes it able to lithiate, but not to delithiate [264]. Consequently, the experiment cannot measure how long it would take for the delithiation current of Si to subside; rather, it measures the relaxation of the graphite component and any additional lithiation of Si that could occur at 150 mV (which is also associated with a “negative” current). On the other hand, the test in Figure A.4B was carried out at a potential high enough for Si to be able to delithiate, and the existence of “positive” 164 delithiation currents long into the experiment indicates that the kinetics of this process is rather slow. Figure A.4B correctly shows that Si would likely continue to exhibit reversible lithiation during voltage hold experiments, requiring extended times for parasitic processes to become dominant, in agreement with Figure 3.5. Finally, we would like to point out that the poor kinetics of Si is not exclusive to composite electrodes. Figure A.4C shows the current measured during the inverse polarization of a ∼50 nm Si thin film, which was previously lithiated to 80 mV, delithiated to 285 mV and then held at this potential. This experiment indicates that the parity between reversible and irreversible currents does not occur before ∼ 600 hours have elapsed, showing that kinetic limitations appear to be intrinsic to the (de)lithiation of silicon. A.3 Hardware Specifications and Sensing Error Based on the information from Figure A.5, we estimate that, for 1.2 Ah cylindrical cells, the expected uncertainty for measuring C/10 capacity is < 0.15% of the cell capacity when using a Maccor 4100. For recording the capacity exchanged during the hold, the error of this cycler is < 0.01% of cell capacity. Thus, deviations between actual capacity loss and the capacity exchanged during the voltage hold in Figure 3.10B are unlikely to originate from sensing limitations. A.4 Measurement Noise An interesting consequence of Equation A.10 is that, even if the voltage hold experiment was to be performed using hardware with perfect time and signal (current and voltage) resolution, and if the parasitic current varies smoothly over time, the measured current could still present an oscillation in its instantaneous amplitude, creating a distribution in values that could be perceived as noise. As discussed above, electron transfers at the cathode and at the anode will momentarily expose the cells to conditions in which Vcell ̸= Vhold , and this gives rise to interesting scenarios depending on the specific capabilities of the instrumentation used in the test. Specific examples are shown in Figure A.6. Hence, currents during the voltage hold that are recorded from “ideal” hardware could still be a combination of null (if Vcell = Vhold , or if Vcell > Vhold when a cycler is used), “negative” (when Vcell > Vhold in a potentiostat) and “positive” values. This fundamental spread in 165 the measured currents, which would be observed even with ideal instrumentation, would only be avoided if R → ∞, such as if a perfectly flat cathode is used; in this case, charge transfer through the circuit would always restore Vcell = Vhold . Naturally, real instrumentation has finite sensitivity, affecting how the measured current relates to the rate of parasitic processes in the cell. Indeed, limitations in time, voltage and current resolution of the sensing circuitry could result in outputs presenting a rather broad distribution of values. Nonetheless, we did not observe a clear correlation between current magnitude and current “noise”. An interesting example is presented in Figure A.7 which compares the currents measured as Gr-1 versus Li (black) and Cu versus Li (copper) cells were held at 100 mV using a Maccor 4100 cycler in the “discharge” mode. Although the magnitude of currents measured with a bare copper foil were generally smaller by an order of magnitude, the spread in the values recorded for the Gr cell was significantly larger. The breadth of the distribution of recorded currents (the “noise”) appears to depend not only on the current resolution, but also on the shape of the voltage profiles of the electrodes used in the experiment. We also evaluated how the accuracy in voltage and current measurements during the hold could affect the test outcomes. To that end, LFP versus Gr-1 single-layer pouch cells were tested using either a Maccor 4100 cycler or a custom-made high-precision cycler (HPC); the sensing limits for both instruments are compared in Figure A.5. Figure A.8A compares the voltages recorded during a potentiostatic hold at 3.335 V. Note that each instrument used different criteria to control the cell voltage during the hold (absolute value versus the last value measured prior to the hold), justifying the offset from the target value measured with the HPC. Nevertheless, using the HPC resulted in a narrower and much smoother voltage distribution, though the maximum offset observed for the Maccor was only < 0.2 mV, with most values remaining within 0.1 mV of the set hold voltage. The improved voltage control achieved with the HPC did not result in a narrower distribution of measured currents (Figure A.8B). Note that, differently from the Maccor, the HPC had the capability of applying “negative” currents during the hold (just as in the case of a potentiostat). Assuming that both instruments are tracking the same underlying rate of parasitic reactions, the occurrence of “negative” currents must be compensated by applying “positive” currents of higher magnitude, which is observed in shortfigrefS62B. 166 A histogram of the values of current measured after 40 hours of hold is presented in Figure A.8C, showing that the HPC provided a broader current distribution even if only the “positive” values are considered. Regardless of the differences observed in the instantaneous values of measured current, the trends exhibited by the integrated capacity curves were identical (Figure A.8D), indicating that both instruments provided equivalent output. Identical aging trends obtained from integrated capacities originated from dissimilar current distributions were also observed in tests with Li versus Gr-1 coin cells in a Maccor 4100 cycler using either a fixed current range (red) or autoranging (black) at the voltage hold step. Figure A.9 overlays data from triplicates for each case, showing that a narrower current distribution obtained from tests with a fixed range did not result in substantial differences in the capacity curves. We note that, in this example, the measured current after 10 hours of hold was always well below the maximum threshold of the lowest current range, so major changes in sensing accuracy were not expected during the experiment. All in all, given the invariance of capacity evolution obtained from different current distributions, the discussion within this section reveals that inferring aging trends from integrated capacity curves appears to be significantly more robust than inspecting values of the measured current. A.5 Quantitative Correction of Aging Data from Commercial LFP-Gr Cells In section S1, we discuss how Equation A.10 relates the current measured during the voltage hold with the theoretical parasitic current when reduction reactions are assumed to be the only parasitic process in the cell. This assumption reasonably holds for LFP-based cells, as the low operating potentials of this cathode (< 3.45VversusLi/Li+ ) warrants that negligible electrolyte oxidation takes place. In this section, we use Equation A.10 to adjust the rates of aging as a function of SOC that is reported in Figure 3.10 for LFP-Gr cells. Equation A.10 relies on the instantaneous slopes of the voltage profiles for the cathode and the anode. Thus, an important question is: what is the true slope of LFP during the voltage hold? Figure A.10A shows a portion of the voltage profile for a LFP versus Li metal coin cell while it was charged at a C/4000 rate for 90 days; the absolute current was 167 1 µA, which is within the same order of magnitude as most of the data points recorded during our voltage hold tests with this cell format (see Figure A.7 and Figure A.9, for example). Even at this minute rate, the LFP profile was surprisingly slanted, varying by ∼ 6 mV within the studied range. Interestingly, the profile appeared much flatter if the OCV was recorded after 6 hours of rest following small charging increments (Figure A.10B), closer to what is used in analytical expressions for the theoretical open circuit curve of LFP, such as the one proposed by Safari et al [257]. (Figure A.10C). As we show below, the latter appears to be more representative of the behavior of the LFP in the commercial 18650-format cells. It is important to consider that the characteristics of the voltage profile of LFP are dependent on particle size distribution, causing variability of outcomes across different material sources [265, 266]. Differential voltage analysis (DVA) [267] was used to estimate the instantaneous slopes of both LFP and graphite in the commercial cells at all SOCs. Since half-cell data from the same electrodes used in the full-cells was not available, the LFP profile was approximated by either using data from a LFP versus Li metal coin cells tested at C/100 or the analytical expression by Safari et al. (Figure A.10C). The profile of graphite was approximated from a Gr-1 versus Li metal coin cell tested at C/100. Figure A.11 compares the experimental voltage profile for the 18650 cell with the obtained from DVA using different LFP input. Clearly, assuming a flatter LFP by using the analytical expression for the OCV results in a better description of the cell behavior at all SOCs and will be used to correct the rate of aging as a function of SOC. Importantly, comparison with Figure 3.10A shows that the aging SOCs are reasonably well-captured by the fitted profile. The rate of aging at all SOCs was estimated by assuming that capacities evolved following a sqrtt dependency. Fits were performed at t > 400 h to eliminate the effects of capacity relaxation at certain SOCs. As shown in Figure A.12, satisfactory fits were obtained for all but the cells at the lowest SOCs; these latter cells exhibited very low rates of aging, which could affect our ability to accurately measure their behavior. Based on the fitted half-cell profiles from the analysis shown in Figure A.11B, local slopes of the voltage profile were obtained, and the values of κ R were computed (Figure A.13A). Given the flatness of LFP, significant departure from unity was only observed at the highest SOC, when the cell was at the curved tip of LFP (Figure 3.10A). Using the 168 calculated κ to correct the rate of aging obtained by experiments resulted in Figure A.13B. Although the correction made the rate of aging grow monotonically with SOC (fixing the anomaly observed in Figure 3.10C and D for high SOCs), the correction seemed unreasonably high compared with expected values in Figure 3.10D. Although the correction helps explain some of the unexpected trends observed in the data, this example suggests that qualitative trends can only be reasonably obtained at intermediate LFP SOCs, when the slope in the voltage profile is minimal. Correctly accounting for the curvatures at the extremes of the LFP profile would require knowledge of the profile slopes with extremely high accuracy and/or a more rigorous treatment than allowed by Equation A.10. Figure A.14 shows the current relaxation if the cell is allowed to equilibrate at OCV before starting a voltage hold. This should minimize reversible current contributions and it does indeed show a similar trend once the reversible portion of the capacity during a typical voltage hold is subtracted. Figure A.15 shows the fitting of the capacity passed during the voltage hold of a LFP-Gr 18650 to different functional forms. Figure A.16 shows how the voltage chosen for the voltage hold changes the recorded capacity. 169 Figure A.1. Theoretical electron transfers during a voltage hold when reactions at the SEI are the only parasitic processes in the cell. The images represent a zoomed-in portion of the voltage profiles of a cathode (blue) and anode (orange), with slopes that are identical in magnitude but opposite in sign. Circles represent filled and unfilled electron sites and are used to indicate how the measured current during the voltage hold relates to the parasitic reduction processes at the anode. Each panel depicts a slice “frozen” in time of rapid electron transfers. A) initial state, assuming that the cell is in equilibrium at Vcell = Vhold . Cell voltage is defined by the state of cathode delithiation and anode lithiation. B) A parasitic reduction reaction occurs at the expense of an electron in the anode, decreasing anode SOC but with no direct effect in cathode SOC. Since the electron loss increases the anode potential, Vcell < Vhold . C) In an attempt to restore the Vcell = Vhold condition, an electron is transferred from the cathode to the anode through the external circuit, restoring the initial anode SOC, but increasing the cathode SOC in the process. Now, the increase in cathode potential makes Vcell > Vhold . D) A new reduction side-reaction consumes another electron from the anode, but the cathode is unaffected. The parasitic process restores Vcell = Vhold , and hence no electron is transferred through the external circuit, i.e., there is no measured current associated with this side-reaction. The theoretical rate of this inaccuracy depends on the ratio between the local of slopes of the electrode profiles, as described by Equation A.10. 170 Figure A.2. Relationship between the ratio of the magnitudes of local slopes of anode and cathode profiles (R) and the accuracy factor κ. Voltage holds can, in principle, provide a more quantitative description of cell aging due to reduction reactions when the cathode profile is much flatter than that of the anode. 171 Figure A.3. Inverse polarization experiments of half-cells. A) Schematic illustration of the evolution of measured currents during an inverse polarization test. The minimum indicates the point when residual delithiation becomes negligible. Voltage hold experiments using this anode should be much longer than this time to mitigate the interference from Irev . B) Gr-1 versus Li held at 105 mV after prior lithiation to 100 mV followed by an additional lithiation of 5% of the total anode capacity (reaching some maximum SOC in the ∼ 89 mV versus Li/Li+ plateau). C) 80% Si held at 300 mV after prior lithiation to 100 mV. All tests were performed at 30 °C in a coin cell format and using a Solartron 1470E potentiostat. Points are partially transparent, so darker regions indicate more frequent occurrences. See Experimental section for details about electrode composition. 172 Figure A.4. Currents measured during inverse polarization experiments of half-cells. A) 15% Si-Gr held at 150 mV after prior lithiation to 110 mV. B) 15% Si-Gr held at 300 mV after prior lithiation to 110 mV. C) ∼50 nm Si thin film held at 285 mV after prior lithiation to 80 mV. All tests were performed at 30 °C in a coin cell format and using a Solartron 1470E potentiostat. Points are partially transparent, so darker regions indicate more frequent occurrences. Figure A.5. Technical specifications for hardware used in this work. For reference, the recorded currents long into the hold are typically < 100µA for 1.2 Ah cylindrical cells, < 10µA for single-layer pouch cells and < 1µA for coin cells. 173 Figure A.6. Correlation between instantaneous voltage and observed current assuming different instrumentation capabilities. Reduction reactions at the anode are assumed to be the sole parasitic process. A positive current is defined as one that increases cell voltage (i.e., a “charging” current). Figure A.7. Currents measured during a potentiostatic hold of Gr (black) and Cu foil (copper) cells versus Li metal, tested at 100 mV using a Maccor 4100 cycler. The measured currents are presented in both log (A) and linear (B) scales. Points are partially transparent, so darker regions indicate more frequent occurrences. For example, data points for graphite cells extend over many orders of magnitude but are mostly concentrated above 0.1 µA. 174 Figure A.8. Comparing potentiostatic holds from identical cells being tested in a Maccor 4100 (cycler) and a high precision cycler (HPC). A) voltages recorded during the hold. B) currents. C) histogram showing all current data points recorded at times t > 40 h. D) Capacities integrated from the measured currents. The capacity curves were vertically shifted to present identical values at t = 50h, to eliminated differences in Qrev . The legend in panel b applies to all panels. 175 Figure A.9. Testing the effect of current auto ranging using Li versus Gr-1 coin-cells tested in a Maccor 4100 cycler: A) currents and B) integrated capacities. The legend in panel B applies to all panels, and data for triplicates are overlaid in the plots. Figure A.10. Open circuit curves of LiFePO4 : A) portion of a LFP versus Li coin cell cycled at C/4000 (1 µA); this current is comparable to the observed long into the hold in our experiments with coin cells. B) OCV values for two cells similar to those used in panel A, but with voltages recorded after 6 hours of rest following incremental C/25 charges. C) Analytical OCV curve for LFP proposed by Safari et al. [257]. 176 Figure A.11. Differential voltage analysis of a C/10 profile of an 18650 cell, using different cathode OCV data. A) LFP half-cell cycled at C/100. B) LFP OCV curve from Safari et al., assuming an 18 mV overpotential. In both cases, the anode profile was obtained from a half-cell cycled at C/100. Although the fit in panel B is not perfect, it captures cell behavior at the SOCs of interest (compare with Figure 3.10A). The legend in panel A applies to both panels. 177 Figure A.12. Determining the rate of aging of √ LFP-Gr cylindrical cells (18650) at all SOCs. Integrated capacities are shown in blue, and t fits are presented as dashed lines. Fits were performed at t > 400 h. Hold voltages and fitting parameters are included in the panels. Note that the assumed functional form is just used as proxy to extract general trends and does not necessarily represent the true long-term behavior of the cells. 178 Figure A.13. Using local slopes to correct aging trends. A) Values of κ R calculated at each SOC using slopes derived from the DVA analysis in Figure A.11B. B) Aging rates corrected using Equation A.10 and the values of κ R shown in panel A. Original aging rates were defined from the fits shown in Figure A.12. SOC is calculated based on the charge capacity prior to the hold added by the Qrev for each cell. Figure A.14. Allowing the cell to relax at open circuit and then performing the voltage hold at the relaxed voltage decreases the reversible lithiation capacity contribution. A) Shows the difference between the capacity passed during a voltage hold at 3.32 V and a voltage hold at 3.314 V after a /sim50 hour OCV step that relaxed to 3.314 V for commercial Gr-LFP 18650s. B) Shows the voltage hold at 3.32 V with the Qrev subtracted. C) Shows the two capacity plots arbitrarily shifted to compare trends. The two plots show similar trends once accounting for the reversible capacity passed when there is not an OCV step prior to the voltage hold. 179 Figure A.15. Fitting the capacity data from the voltage hold of an LFP-Gr 18650-format cell (black line) held at 3.35 V (∼95% SOC) at 25°C. Dashed lines indicate the trends obtained after fitting portions of the data with various functional forms √ encountered in the literature 0.5 ( Q = a t + b ), tx (Q = at x + b), linear [95, 241, 242] for Li+ loss during calendar aging: t √ (Q = at + b), t0.5 + linear (Q = a t + bt + c), and logarithmic (Q = a ln t + b). The data interval used for fitting is delimited by gray squares: A) 200 - 600 hours. B) 200 - 1000 hours. C) 600 - 1000 hours. When only the latter portion of the plot is considered (panel C), the fits tend to converge within the remaining shown portion of the data. Panel C also present values for the calendar life obtained from the fits to 600 -1000 h of the data, which is the number of years in which the cell would be predicted to lose 20% of its initial capacity. The 4-6 years obtained for most of the fits largely underestimates the reported projected life of LFP-Gr cells under similar conditions, which is likely > 9 years [242, 252, 268]. For all three examples, assuming a t0.5 dependency consistently captured the behavior of the remaining portion of the data with relative accuracy. This assumed dependency is used throughout the paper to estimate aging rates for the cells. 180 Figure A.16. Integrated capacities recorded during holds at various voltages. The large variation observed at early stages of the hold is due to Qrev and shows strong dependence on the chosen voltage. This dataset is discussed in detail in Figure 3.10. REFERENCES [1] A. F. Gonzalez, N. H. Yang, and R. S. Liu. Silicon anode design for lithium-ion batteries: Progress and perspectives. Journal of Physical Chemistry C, 121(50):27775–27787, 2017. [2] J. Ryu, D. Hong, H. W. Lee, and S. Park. Practical considerations of si-based anodes for lithium-ion battery applications. Nano Research, 10(12):3970–4002, 2017. [3] W. Luo, X. Q. Chen, Y. Xia, M. Chen, L. J. Wang, Q. Q. Wang, W. Li, and J. P. Yang. Surface and interface engineering of silicon-based anode materials for lithium-ion batteries. Advanced Energy Materials, 7(24), 2017. [4] M. T. McDowell, S. W. Lee, W. D. Nix, and Y. Cui. 25th anniversary article: Understanding the lithiation of silicon and other alloying anodes for lithium-ion batteries. Advanced Materials, 25(36):4966–4984, 2013. [5] T.B. Reddy. Linden’s Handbook of Batteries. The McGraw-Hill Companies, New York City, 1325 Avenue of the Americas, 4 edition, 2011. [6] C. H. Chen, E. Chason, and P. R. Guduru. Measurements of the phase and stress evolution during initial lithiation of sn electrodes. Journal of the Electrochemical Society, 164(4):A574–A579, 2017. [7] D. Aurbach, B. Markovsky, M. D. Levi, E. Levi, A. Schechter, M. Moshkovich, and Y. Cohen. New insights into the interactions between electrode materials and electrolyte solutions for advanced nonaqueous batteries. Journal of Power Sources, 81:95– 111, 1999. [8] E. Peled, D. Golodnitsky, and G. Ardel. Advanced model for solid electrolyte interphase electrodes in liquid and polymer electrolytes. Journal of the Electrochemical Society, 144(8):L208–L210, 1997. [9] A. P. Wang, S. Kadam, H. Li, S. Q. Shi, and Y. Qi. Review on modeling of the anode solid electrolyte interphase (sei) for lithium-ion batteries. NPJ Computational Materials, 4, 2018. [10] M. Haruta, Y. Kijima, R. Hioki, T. Doi, and M. Inaba. Artificial lithium fluoride surface coating on silicon negative electrodes for the inhibition of electrolyte decomposition in lithium-ion batteries: visualization of a solid electrolyte interphase using in situ afm. Nanoscale, 10(36):17257–17264, 2018. [11] R. Kumar, P. Lu, X. C. Xiao, Z. Q. Huang, and B. W. Sheldon. Strain-induced lithium losses in the solid electrolyte interphase on silicon electrodes. ACS Applied Materials & Interfaces, 9(34):28406–28417, 2017. 182 [12] F. Sun, H. Markotter, K. Dong, I. Manke, A. Hilger, N. Kardjilov, and J. Banhart. Investigation of failure mechanisms in silicon based half cells during the first cycle by micro x-ray tomography and radiography. Journal of Power Sources, 321:174–184, 2016. [13] F. F. Shi, P. N. Ross, G. A. Somorjai, and K. Komvopoulos. The chemistry of electrolyte reduction on silicon electrodes revealed by in situ atr-ftir spectroscopy. Journal of Physical Chemistry C, 121(27):14476–14483, 2017. [14] M. Gu, Z. G. Wang, J. G. Connell, D. E. Perea, L. J. Lauhon, F. Gao, and C. M. Wang. Electronic origin for the phase transition from amorphous lixsi to crystalline li15si4. Acs Nano, 7(7):6303–6309, 2013. [15] Y. Son, S. Jaekyung, Y. Son, and J. Cho. Recent progress of analysis of techniques for silicon-based andoe of lithium-ion batteries. Elsevier, 6:77–83, 2017. [16] D. Liu, O. Lord, and P. E. J. Flewitt. Calibration of raman spectroscopy in the stress measurement of air-plasma-sprayed yttria-stabilized zirconia. Applied Spectroscopy, 66(10):1204–1209, 2012. [17] H. Gao, L. S. Xiao, I. Plume, G. L. Xu, Y. Ren, X. B. Zuo, Y. Z. Liu, C. Schulz, H. Wiggers, K. Amine, and Z. H. Chen. Parasitic reactions in nanosized silicon anodes for lithium-ion batteries. Nano Letters, 17(3):1512–1519, 2017. [18] G. M. Veith, M. Doucet, J. K. Baldwin, R. L. Sacci, T. M. Fears, Y. Q. Wang, and J. F. Browning. Direct determination of solid-electrolyte interphase thickness and composition as a function of state of charge on a silicon anode. Journal of Physical Chemistry C, 119(35):20339–20349, 2015. [19] Z. Q. Zhuo, P. Lu, C. Delacourt, R. M. Qiao, K. Xu, F. Pan, S. J. Harris, and W. L. Yang. Breathing and oscillating growth of solid-electrolyte-interphase upon electrochemical cycling. Chemical Communications, 54(7):814–817, 2018. [20] B. Cunningham. Silicon and intermetallic anode portfolio strategy overview. Annual Merit Review, 2020. [21] I. Zilberman, S. Ludwig, and A. Jossen. Cell-to-cell variation of calendar aging and reversible self-discharge in 18650 nickel-rich, silicon-graphite lithium-ion cells. Journal of Energy Storage, 26:100900, 2019. [22] L. De Sutter, G. Berckmans, M. Marinaro, J. Smekens, Y. Firouz, M. WohlfahrtMehrens, J. Van Mierlo, and N. Omar. Comprehensive aging analysis of volumetric constrained lithium-ion pouch cells with high concentration silicon-alloy anodes. Energies, 11(11), 2018. [23] W. Lu, L. Zhang, Y. Qin, and A. Jansen. Calendar and cycle life of lithium-ion batteries containing silicon monoxide anode. Journal of The Electrochemical Society, 165(10):A2179–A2183, 2018. [24] M. P. Manoharan, S. Kumar, M. A. Haque, R. Rajagopalan, and H. C. Foley. Room temperature amorphous to nanocrystalline transformation in ultra-thin films under tensile stress: an in situ tem study. Nanotechnology, 21(50), 2010. 183 [25] X. J. Wang, F. F. Fan, J. W. Wang, H. R. Wang, S. Y. Tao, A. Yang, Y. Liu, H. B. Chew, S. X. Mao, T. Zhu, and S. M. Xia. High damage tolerance of electrochemically lithiated silicon. Nature Communications, 6, 2015. [26] B. Key, R. Bhattacharyya, M. Morcrette, V. Seznec, J. M. Tarascon, and C. P. Grey. Real-time nmr investigations of structural changes in silicon electrodes for lithium-ion batteries. Journal of the American Chemical Society, 131(26):9239–9249, 2009. [27] B. Key, M. Morcrette, J. M. Tarascon, and C. P. Grey. Pair distribution function analysis and solid state nmr studies of silicon electrodes for lithium ion batteries: Understanding the (de)lithiation mechanisms. Journal of the American Chemical Society, 133(3):503–512, 2011. [28] J. H. Trill, C. Q. Tao, M. Winter, S. Passerini, and H. Eckert. Nmr investigations on the lithiation and delithiation of nanosilicon-based anodes for li-ion batteries. Journal of Solid State Electrochemistry, 15(2):349–356, 2011. [29] S. Tardif, E. Pavlenko, L. Quazuguel, M. Boniface, M. Marechal, J. S. Micha, L. Gonon, V. Mareau, G. Gebel, P. Bayle-Guillemaud, F. Rieutord, and S. Lyonnard. Operando raman spectroscopy and synchrotron x-ray diffraction of lithiation/delithiation in silicon nanoparticle anodes. ACS Nano, 11(11):11306–11316, 2017. [30] Z. D. Zeng, N. A. Liu, Q. S. Zeng, S. W. Lee, W. L. Mao, and Y. Cui. In situ measurement of lithiation-induced stress in silicon nanoparticles using micro-raman spectroscopy. Nano Energy, 22:105–110, 2016. [31] S. T. Amimoto, D. J. Chang, and A. D. Birkitt. Stress measurements in mems using raman spectroscopy. Materials and Device Characterization in Micromachining, 3512:123– 129, 1998. [32] T. Wermelinger, C. Charpentier, M. D. Yüksek, and R. Spolenak. Measuring stresses in thin metal films by means of raman microscopy using silicon as a strain gage material. Journal of Raman Spectroscopy, 40(12):1849–1857, 2009. [33] K. H. Wu, X. Q. Yan, and M. W. Chen. In situ raman characterization of reversible phase transition in stress-induced amorphous silicon. Applied Physics Letters, 91(10), 2007. [34] J. Nanda, M. K. Datta, J. T. Remillard, A. O’Neill, and P. N. Kumta. In situ raman microscopy during discharge of a high capacity silicon-carbon composite li-ion battery negative electrode. Electrochemistry Communications, 11(1):235–237, 2009. [35] S. Q. Huang, L. Z. Cheong, D. Y. Wang, and C. Shen. Thermal stability of solid electrolyte interphase of lithium-ion batteries. Applied Surface Science, 454:61–67, 2018. [36] B. Breitung, P. Baumann, H. Sommer, J. Janek, and T. Brezesinski. In situ and operando atomic force microscopy of high-capacity nano-silicon based electrodes for lithium-ion batteries. Nanoscale, 8(29):14048–14056, 2016. [37] L. Y. Beaulieu, T. D. Hatchard, A. Bonakdarpour, M. D. Fleischauer, and J. R. Dahn. Reaction of li with alloy thin films studied by in situ afm. Journal of the Electrochemical Society, 150(11):A1457–A1464, 2003. 184 [38] B. K. Seidlhofer, B. Jerliu, M. Trapp, E. Huger, S. Risse, R. Cubitt, H. Schmidt, R. Steitz, and M. Ballauff. Lithiation of crystalline silicon as analyzed by operando neutron reflectivity. ACS Nano, 10(8):7458–7466, 2016. [39] S. K. Soni, B. W. Sheldon, X. C. Xiao, M. W. Verbrugge, D. Ahn, H. Haftbaradaran, and H. J. Gao. Stress mitigation during the lithiation of patterned amorphous si islands. Journal of the Electrochemical Society, 159(1):A38–A43, 2012. [40] S. K. Soni, B. W. Sheldon, X. C. Xiao, and A. Tokranov. Thickness effects on the lithiation of amorphous silicon thin films. Scripta Materialia, 64(4):307–310, 2011. [41] V. A. Sethuraman, M. J. Chon, M. Shimshak, V. Srinivasan, and P. R. Guduru. In situ measurements of stress evolution in silicon thin films during electrochemical lithiation and delithiation. Journal of Power Sources, 195(15):5062–5066, 2010. [42] D. E. Arreaga-Salas, A. K. Sra, K. Roodenko, Y. J. Chabal, and C. L. Hinkle. Progression of solid electrolyte interphase formation on hydrogenated amorphous silicon anodes for lithium-ion batteries. Journal of Physical Chemistry C, 116(16):9072–9077, 2012. [43] M. J. Chon, V. A. Sethuraman, A. McCormick, V. Srinivasan, and P. R. Guduru. Real-time measurement of stress and damage evolution during initial lithiation of crystalline silicon. Physical Review Letters, 107(4), 2011. [44] M. Boniface, L. Quazuguel, J. Danet, D. Guyomard, P. Moreau, and P. BayleGuillemaud. Nanoscale chemical evolution of silicon negative electrodes characterized by low-loss stem-eels. Nano Letters, 16(12):7381–7388, 2016. [45] K. Ogata, S. Jeon, D. S. Ko, I. S. Jung, J. H. Kim, K. Ito, Y. Kubo, K. Takei, S. Saito, Y. H. Cho, H. Park, J. Jang, H. G. Kim, J. H. Kim, Y. S. Kim, W. Choi, M. Koh, K. Uosaki, S. G. Doo, Y. Hwang, and S. Han. Evolving affinity between coulombic reversibility and hysteretic phase transformations in nano-structured silicon-based lithium-ion batteries. Nature Communications, 9, 2018. [46] K. W. Schroder, A. G. Dylla, S. J. Harris, L. J. Webb, and K. J. Stevenson. Role of surface oxides in the formation of solid-electrolyte interphases at silicon electrodes for lithium-ion batteries. ACS Applied Materials & Interfaces, 6(23):21510–21524, 2014. [47] A. A. Hubaud, Z. Z. Yang, D. J. Schroeder, F. Dogan, L. Trahey, and J. T. Vaughey. Interfacial study of the role of sio2 on si anodes using electrochemical quartz crystal microbalance. Journal of Power Sources, 282:639–644, 2015. [48] H. Bulter, M. Sternad, E. D. Sardinha, J. Witt, C. Dosche, M. Wilkening, and G. Wittstock. Investigation of the electron transfer at si electrodes: Impact and removal of the native sio2 layer. Journal of the Electrochemical Society, 163(3):A504–A512, 2016. [49] S. L. Zhang, K. J. Zhao, T. Zhu, and J. Li. Electrochemomechanical degradation of high-capacity battery electrode materials. Progress in Materials Science, 89:479–521, 2017. [50] S. F. Lux, I. T. Lucas, E. Pollak, S. Passerini, M. Winter, and R. Kostecki. The mechanism of hf formation in lipf6 based organic carbonate electrolytes. Electrochemistry Communications, 14(1):47–50, 2012. 185 [51] A. N. Dey. Film formation on lithium anode in propylene carbonate. Journal of the Electrochemical Society, 117(8):C248–&, 1970. [52] E. Peled. The electrochemical-behavior of alkali and alkaline-earth metals in nonaqueous battery systems - the solid electrolyte interphase model. Journal of the Electrochemical Society, 126(12):2047–2051, 1979. [53] G. M. Veith, L. Baggetto, R. L. Sacci, R. R. Unocic, W. E. Tenhaeff, and J. F. Browning. Direct measurement of the chemical reactivity of silicon electrodes with lipf6-based battery electrolytes. Chemical Communications, 50(23):3081–3084, 2014. [54] H. Yamin and E. Peled. Solid-electrolyte interphase (sei) electrode .8. lithium-cyclic ether systems. Journal of the Electrochemical Society, 130(8):C307–C307, 1983. [55] E. Peled, R. Tulman, and A. Meitav. Solid electrolyte interphase (sei) electrodes .6. calcium-ca(alcl4)2-sulfuryl chloride system. Journal of Power Sources, 10(2):125–135, 1983. [56] E. Peled, A. Meitav, and R. Tulman. Solid electrolyte interphase (sei electrodes .5. calcium-ca(alcl4)2 thionyl chloride system. Journal of the Electrochemical Society, 129(8):C336–C336, 1982. [57] A. Meitav and E. Peled. Solid electrolyte interphase (sei) electrode .3. deposition dissolution process of lithium in thionyl-chloride solution. Journal of Electroanalytical Chemistry, 134(1):49–63, 1982. [58] E. Elster, R. Cohen, and E. Peled. Solid electrolyte interphase (sei) electrodes .7. the electrochemical-behavior of calcium in ba(alcl4)2 and sr(alcl4)2 thionyl chloride solutions. Journal of Power Sources, 26(3-4):423–425, 1989. [59] A. Meitav and E. Peled. Solid electrolyte interphase (sei) electrode .2. the formation and properties of the sei on magnesium in soci2-mg(aici)4)2 solutions. Journal of the Electrochemical Society, 128(4):825–831, 1981. [60] R. Fong, U. Vonsacken, and J. R. Dahn. Studies of lithium intercalation into carbons using nonaqueous electrochemical-cells. Journal of the Electrochemical Society, 137(7):2009–2013, 1990. [61] D. Aurbach, M. L. Daroux, P. W. Faguy, and E. Yeager. Identification of surface-films formed on lithium in propylene carbonate solutions. Journal of the Electrochemical Society, 134(7):1611–1620, 1987. [62] G. Nazri and R. H. Muller. Composition of surface-layers on li electrodes in pc, liclo4 of very low water-content. Journal of the Electrochemical Society, 132(9):2050–2054, 1985. [63] K. Kalaga, I. A. Shkrob, R. T. Haasch, C. Peebles, J. Bareno, and D. P. Abraham. Auger electrons as probes for composite micro- and nanostructured materials: Application to solid electrolyte interphases in graphite and silicon-graphite electrodes. Journal of Physical Chemistry C, 121(42):23333–23346, 2017. [64] D. A. Dalla Corte, A. C. Gouget-Laemme, K. Lahlil, G. Caillon, C. Jordy, J. N. Chazalviel, T. Gacoin, M. Rosso, and F. Ozanam. Molecular grafting on silicon anodes: artificial solid-electrolyte interphase and surface stabilization. Electrochimica Acta, 201:70–77, 2016. 186 [65] S. Zhang, M. N. He, C. C. Su, and Z. C. Zhang. Advanced electrolyte/additive for lithium-ion batteries with silicon anode. Current Opinion in Chemical Engineering, 13:24–35, 2016. [66] A. M. Haregewoin, A. S. Wotango, and B. J. Hwang. Electrolyte additives for lithium ion battery electrodes: progress and perspectives. Energy & Environmental Science, 9(6):1955–1988, 2016. [67] R. Nolle, A. J. Achazi, P. Kaghazchi, M. Winter, and T. Placke. Pentafluorophenyl isocyanate as an effective electrolyte additive for improved performance of siliconbased lithium-ion full cells. ACS Applied Materials & Interfaces, 10(33):28187–28198, 2018. [68] S. W. Song and S. W. Baek. Silane-derived sei stabilization on thin-film electrodes of nanocrystalline si for lithium batteries. Electrochemical and Solid State Letters, 12(2):A23–A27, 2009. [69] Y. G. Ryu, S. Lee, S. Mah, D. J. Lee, K. Kwon, S. Hwang, and S. Doo. Electrochemical behaviors of silicon electrode in lithium salt solution containing alkoxy silane additives. Journal of the Electrochemical Society, 155(8):A583–A589, 2008. [70] A. Rezqita, M. Sauer, A. Foelske, H. Kronberger, and A. Trifonova. The effect of electrolyte additives on electrochemical performance of silicon/mesoporous carbon (si/mc) for anode materials for lithium-ion batteries. Electrochimica Acta, 247:600–609, 2017. [71] G. M. Veith, M. Doucet, R. L. Sacci, B. Vacaliuc, J. K. Baldwin, and J. F. Browning. Determination of the solid electrolyte interphase structure grown on a silicon electrode using a fluoroethylene carbonate additive. Scientific Reports, 7, 2017. [72] M. Y. Nie, D. P. Abraham, Y. J. Chen, A. Bose, and B. L. Lucht. Silicon solid electrolyte interphase (sei) of lithium ion battery characterized by microscopy and spectroscopy. Journal of Physical Chemistry C, 117(26):13403–13412, 2013. [73] B. T. Young, D. R. Heskett, C. C. Nguyen, M. Y. Nie, J. C. Woicik, and B. L. Lucht. Hard x-ray photoelectron spectroscopy (haxpes) investigation of the silicon solid electrolyte interphase (sei) in lithium-ion batteries. ACS Applied Materials & Interfaces, 7(36):20004–20011, 2015. [74] C. Xu, F. Lindgren, B. Philippe, M. Gorgoi, F. Bjorefors, K. Edstrom, and T. Gustafsson. Improved performance of the silicon anode for li-ion batteries: Understanding the surface modification mechanism of fluoroethylene carbonate as an effective electrolyte additive. Chemistry of Materials, 27(7):2591–2599, 2015. [75] C. C. Nguyen and B. L. Lucht. Comparative study of fluoroethylene carbonate and vinylene carbonate for silicon anodes in lithium ion batteries. Journal of the Electrochemical Society, 161(12):A1933–A1938, 2014. [76] J. B. Chen, A. K. Thapa, and T. A. Berfield. In-situ characterization of strain in lithium battery working electrodes. Journal of Power Sources, 271:406–413, 2014. 187 [77] J. S. Kim, D. Byun, and J. K. Lee. Electrochemical characteristics of amorphous silicon thin film electrode with fluoroethylene carbonate additive. Current Applied Physics, 14(4):596–602, 2014. [78] S. Uchida, M. Yamagata, and M. Ishikawa. Effect of electrolyte additives on nonnano-si negative electrodes prepared with polyimide binder. Journal of the Electrochemical Society, 162(3):A406–A412, 2015. [79] K. Schroder, J. Avarado, T. A. Yersak, J. C. Li, N. Dudney, L. J. Webb, Y. S. Meng, and K. J. Stevenson. The effect of fluoroethylene carbonate as an additive on the solid electrolyte interphase on silicon lithium-ion electrodes. Chemistry of Materials, 27(16):5531–5542, 2015. [80] M. Sina, J. Alvarado, H. Shobukawa, C. Alexander, V. Manichev, L. Feldman, T. Gustafsson, K. J. Stevenson, and Y. S. Meng. Direct visualization of the solid electrolyte interphase and its effects on silicon electrochemical performance. Advanced Materials Interfaces, 3(20), 2016. [81] H. Wang, N. Y. Sa, M. N. He, X. Liang, L. F. Nazar, M. Balasubramanian, K. G. Gallagher, and B. Key. In situ nmr observation of the temporal speciation of lithium sulfur batteries during electrochemical cycling. Journal of Physical Chemistry C, 121(11):6011– 6017, 2017. [82] A. Bordes, K. Eom, and T. F. Fuller. The effect of fluoroethylene carbonate additive content on the formation of the solid-electrolyte interphase and capacity fade of liion full-cell employing nano si-graphene composite anodes. Journal of Power Sources, 257:163–169, 2014. [83] V. Etacheri, O. Haik, Y. Goffer, G. A. Roberts, I. C. Stefan, R. Fasching, and D. Aurbach. Effect of fluoroethylene carbonate (fec) on the performance and surface chemistry of si-nanowire li-ion battery anodes. Langmuir, 28(1):965–976, 2012. [84] S. Dalavi, P. Guduru, and B. L. Lucht. Performance enhancing electrolyte additives for lithium ion batteries with silicon anodes. Journal of the Electrochemical Society, 159(5):A642–A646, 2012. [85] Y. T. Jin, N. J. H. Kneusels, L. E. Marbella, E. Castillo-Martinez, P. C. M. M. Magusin, R. S. Weatherup, E. Jonsson, T. Liu, S. Paul, and C. P. Grey. Understanding fluoroethylene carbonate and vinylene carbonate based electrolytes for si anodes in lithium ion batteries with nmr spectroscopy. Journal of the American Chemical Society, 140(31):9854–9867, 2018. [86] B. Han, C. Liao, F. Dogan, S. E. Trask, S. H. Lapidus, J. T. Vaughey, and B. Key. Using mixed salt electrolytes to stabilize silicon anodes for lithium-ion batteries via in situ formation of li-m-si ternaries (m = mg, zn, al, ca). ACS Applied Material Interfaces, 11(33):29780–29790, August 2019. [87] X. Li, J. A. Gilbert, S. E. Trask, R. Uppuluri, S. H. Lapidus, S. Cora, N. Sa, Z. Yang, I. D. Bloom, F. Dogan, J. T. Vaughey, and B. Key. Investigating ternary li-mg-si zintl phase formation and evolution for si anodes in li-ion batteries with mg(tfsi)2 electrolyte additive. Chemistry of Materials, 33(13):4960–4970, July 2021. 188 [88] D. E. Galvez-Aranda, A. Verma, K. Hankins, J. M. Seminario, P. P. Mukherjee, and P. B. Balbuena. Chemical and mechanical degradation and mitigation strategies for si anodes. Journal of Power Sources, 419:208–218, 2019. [89] H. C. Kolb, M. G. Finn, and K. B. Sharpless. Click chemistry: Diverse chemical function from a few good reactions. Angewandte Chemie-International Edition, 40(11):2004–+, 2001. [90] Y. Jin, S. Li, A. Kushima, X. Q. Zheng, Y. M. Sun, J. Xie, J. Sun, W. J. Xue, G. M. Zhou, J. Wu, F. F. Shi, R. F. Zhang, Z. Zhu, K. P. So, Y. Cui, and J. Li. Self-healing sei enables full-cell cycling of a silicon-majority anode with a coulombic efficiency exceeding 99.9%. Energy & Environmental Science, 10(2):580–592, 2017. [91] J. Jiang, H. Zhang, J. H. Zhu, L. P. Li, Y. N. Liu, T. Meng, L. Ma, M. W. Xu, J. P. Liu, and C. M. Li. Putting nanoarmors on yolk-shell si@c nanoparticles: A reliable engineering way to build better si-based anodes for li-ion batteries. ACS Applied Materials & Interfaces, 10(28):24157–24163, 2018. [92] I. Yoon, D. P. Abraham, B. L. Lucht, A. F. Bower, and P. R. Guduru. In situ measurement of solid electrolyte interphase evolution on silicon anodes using atomic force microscopy. Advanced Energy Materials, 6(12), 2016. [93] P. P. R. M. L. Harks, F. M. Mulder, and P. H. L. Notten. In situ methods for li-ion battery research: A review of recent developments. Journal of Power Sources, 288:92– 105, 2015. [94] P. Keil, S.F. Schuster, J. Wilhelm, J. Travi, A. Hauser, R.C. Karl, and A. Jossen. Calendar aging of lithium-ion batteries. Journal of The Electrochemical Society, 163(9):A1872– A1880, 2016. [95] F. Single, A. Latz, and B. Horstmann. Identifying the mechanism of continued growth of the solid-electrolyte interphase. ChemSusChem, 11(12):1950–1955, June 2018. [96] I. Zilberman, J. Sturm, and A. Jossen. Reversible self-discharge and calendar aging of 18650 nickel-rich, silicon-graphite lithium-ion cells. Journal of Power Sources, 425:217– 226, 2019. [97] M. T. F. Rodrigues, K. Kalaga, S. E. Trask, I. A. Shkrob, and D. P. Abraham. Anodedependent impedance rise in layered-oxide cathodes of lithium-ion cells. Journal of The Electrochemical Society, 165(9):A1697–A1705, 2018. [98] M. Dubarry, N. Qin, and P. Brooker. Calendar aging of commercial li-ion cells of different chemistries - a review. Current Opinion in Electrochemistry, 9(Complete):106– 113, 2018. [99] Z. Deng, Z. Huang, Y. Shen, Y. Huang, H. Ding, A. Luscombe, M. Johnson, J. E. Harlow, R. Gauthier, and J. R. Dahn. Ultrasonic scanning to observe wetting and “unwetting” in li-ion pouch cells. Joule, 4(9):2017–2029, 2020. [100] M. Naumann, M. Schimpe, P. Keil, H.C. Hesse, and A. Jossen. Analysis and modeling of calendar aging of a commercial lifepo4/graphite cell. Journal of Energy Storage, 17:153–169, 2018. 189 [101] Y. Ding, Z. P. Cano, A. Yu, J. Lu, and Z. Chen. Automotive li-ion batteries: Current status and future perspectives. Electrochemical Energy Reviews, 2:1–28, 2019. [102] K. Kalaga, M. T. F. Rodrigues, S. E. Trask, I. A. Shkrob, and D. P. Abraham. Calendarlife versus cycle-life aging of lithium-ion cells with silicon-graphite composite electrodes. Electrochimica Acta, 280:221–228, 2018. [103] D. Schneier, Y. Shaham, G. Ardel, L. Burstein, Y. Kamir, and E. Peled. Elucidation of the spontaneous passivation of silicon anodes in lithium battery electrolytes. Journal of The Electrochemical Society, 166(16):A4020–A4024, 2019. [104] Y. Yin, E. Arca, L. Wang, G. Yang, M. Schnabel, L. Cao, C. Xiao, H. Zhou, P. Liu, J. Nanda, G. Teeter, B. Eichhorn, K. Xu, A. Burrell, and C Ban. Nonpassivated silicon anode surface. ACS Applied Materials Interfaces, 12(23):26593–26600, June 2020. [105] R. T. Pekarek, A. Affolter, L. L. Baranowski, J. Coyle, T. Hou, E. Sivonxay, B. A. Smith, R. D. McAuliffe, K. A. Persson, B. Key, C. Apblett, G. M. Veith, and N. R. Neale. Intrinsic chemical reactivity of solid-electrolyte interphase components in silicon-lithium alloy anode batteries probed by ftir spectroscopy. Journal of Materials Chemistry A, 8(16):7897–7906, 2020. [106] C. L. Seitzinger, R. L. Sacci, J. E. Coyle, C. A. Apblett, A. Hays, K, R. R. Armstrong, A. M. Rogers, B. L. Armstrong, T. H. Bennet, N. R. Neale, and G. M. Veith. Intrinsic chemical reactivity of silicon electrode materials: Gas evolution. Chemistry of Materials, 32(7):3199–3210, April 2020. [107] D. Aurbach, K. Gamolsky, B. Markovsky, Y. Gofer, M. Schmidt, and U. Heider. On the use of vinylene carbonate (vc) as an additive to electrolyte solutions for li-ion batteries. Electrochimica Acta, 47(9):1423–1439, 2002. [108] I. Yoon, D. P. Abraham, B. L. Lucht, A. F. Bower, and P. R. Guduru. In situ measurement of solid electrolyte interphase evolution on silicon anodes using atomic force microscopy. Advanced Energy Materials, 6(12):1600099, June 2016. [109] I. Hasa, A. M. Haregewoin, L. Zhang, W. Tsai, J. Guo, G. M. Veith, P. N. Ross, and R. Kostecki. Electrochemical reactivity and passivation of silicon thin-film electrodes in organic carbonate electrolytes. ACS Applied Materials Interfaces, 12(36):40879–40890, September 2020. [110] I. A. Shkrob, J. F. Wishart, and D. P. Abraham. What makes fluoroethylene carbonate different? Journal of Physical Chemistry C, 119(27):14954–14964, July 2015. [111] R. Jung, M. Metzger, D. Haering, S. Solchenbach, C. Marino, N. Tsiouvaras, C. Stinner, and H. A. Gasteiger. Consumption of fluoroethylene carbonate (fec) on sic composite electrodes for li-ion batteries. Journal of The Electrochemical Society, 163(8):A1705–A1716, 2016. [112] H. Bryngelsson, M. Stjerndahl, T. Gustafsson, and K. Edström. How dynamic is the sei? Journal of Power Sources, 174(2):970–975, 2007. [113] C. Stetson, Y. Yin, C. S. Jiang, S. C. DeCaluwe, M. Al-Jassim, N. R. Neale, C. Ban, and A. Burrell. Temperature-dependent solubility of solid electrolyte interphase on silicon electrodes. ACS Energy Letters, 4(12):2770–2775, December 2019. 190 [114] T. Hou, G. Yang, N. N. Rajput, J. Self, S. Park, J. Nanda, and K. A. Persson. The influence of fec on the solvation structure and reduction reaction of lipf6/ec electrolytes and its implication for solid electrolyte interphase formation. Nano Energy, 64:103881, 2019. [115] M. Stich, M. Göttlinger, M. Kurniawan, U. Schmidt, and A. Bund. Hydrolysis of lipf6 in carbonate-based electrolytes for lithium-ion batteries and in aqueous media. Journal of Physical Chemistry C, 122(16):8836–8842, April 2018. [116] S. Wiemers-Meyer, S. Jeremias, M. Winter, and S. Nowak. Influence of battery cell components and water on the thermal and chemical stability of lipf6 based lithium ion battery electrolytes. Electrochimica Acta, 222:1267–1271, 2016. [117] J. C. Burns, N. N. Sinha, Gaurav Jain, Hui Ye, Collette M. VanElzen, Erik Scott, A. Xiao, W. M. Lamanna, and J. R. Dahn. The impact of intentionally added water to the electrolyte of li-ion cells. Journal of The Electrochemical Society, 160(11):A2281– A2287, 2013. [118] Y. Ha, C. Stetson, S. P. Harvey, G. Teeter, B. J. Tremolet de Villers, C. S. Jiang, M. Schnabel, P. Stradins, A. Burrell, and S. Han. Effect of water concentration in lipf6-based electrolytes on the formation, evolution, and properties of the solid electrolyte interphase on si anodes. ACS Applied Materials Interfaces, 12(44):49563–49573, November 2020. [119] C. Jung, K. Kim, and S. Hong. Stable silicon anode for lithium-ion batteries through covalent bond formation with a binder via esterification. ACS Applied Materials Interfaces, 11(30):26753–26763, July 2019. [120] J. Bareño, I. A. Shkrob, J. A. Gilbert, M. Klett, and D. P. Abraham. Capacity fade and its mitigation in li-ion cells with silicon-graphite electrodes. Journal of Physical Chemistry C, 121(38):20640–20649, September 2017. [121] A. Tokranov, B. W. Sheldon, C. Z. Li, S. Minne, and X. C. Xiao. In situ atomic force microscopy study of initial solid electrolyte interphase formation on silicon electrodes for li-ion batteries. ACS Applied Materials & Interfaces, 6(9):6672–6686, 2014. [122] J. Y. Zheng, H. Zheng, R. Wang, L. B. Ben, W. Lu, L. W. Chen, L. Q. Chen, and H. Li. 3d visualization of inhomogeneous multi-layered structure and young’s modulus of the solid electrolyte interphase (sei) on silicon anodes for lithium ion batteries. Physical Chemistry Chemical Physics, 16(26):13229–13238, 2014. [123] A. Mukhopadhyay, A. Tokranov, X. C. Xiao, and B. W. Sheldon. Stress development due to surface processes in graphite electrodes for li-ion batteries: A first report. Electrochimica Acta, 66:28–37, 2012. [124] H. Tavassol, E. M. C. Jones, N. R. Sottos, and A. A. Gewirth. Electrochemical stiffness in lithium-ion batteries. Nature Materials, 15(11):1182–+, 2016. [125] R. Kumar, A. Tokranov, B. W. Sheldon, X. C. Xiao, Z. Q. Huang, C. Z. Li, and T. Mueller. In situ and operando investigations of failure mechanisms of the solid electrolyte interphase on silicon electrodes. ACS Energy Letters, 1(4):689–697, 2016. 191 [126] Y. Liu, K. Guo, C. Wang, and H. Gao. Wrinkling and ratcheting of a thin film on cyclically deforming plastic substrate: Mechanical instability of the solid-electrolyte interphase in li-ion batteries. Journal of the Mechanics and Physics of Solids, 123:103–118, 2019. [127] X. M. Cheng and M. Pecht. In situ stress measurement techniques on li-ion battery electrodes: A review. Energies, 10(5), 2017. [128] D. Post, B. Han, and P. Ifju. High Sensitivity Moire: Experimental Analysis for Mechanics and Materials. Mechanical Engineering Series. Springer, Verlag New York, 1994. [129] W. Callister and D. Rethwisch. Materials Science and Engineering: An Introduction. John Wiley & Sons, Inc., Hoboken, NJ, River St, 8 edition, 2010. [130] K. Guo, R. Kumar, X. C. Xiao, B. W. Sheldon, and H. J. Gao. Failure progression in the solid electrolyte interphase (sei) on silicon electrodes. Nano Energy, 68, 2020. [131] G. G. Stoney. The tension of metallic films deposited by electrolysis. Proceedings of the Royal Society of London Series a-Containing Papers of a Mathematical and Physical Character, 82(553):172–175, 1909. [132] X. Feng, Y. Huang, and A. J. Rosakis. On the stoney formula for a thin film/substrate system with nonuniform substrate thickness. Journal of Applied Mechanics-Transactions of the Asme, 74(6):1276–1281, 2007. [133] L. B. Freund and S. Suresh. Thin Film Materials: Stress, Defect Formation and Surface Evolution. Cambridge University Press, Cambridge, Shaftesbury Road, 2004. [134] E. Ventosa, P. Wilde, A.-H. Zinn, M. Trautmann, A. Ludwig, and W. Schuhmann. Understanding surface reactivity of si electrodes in li-ion batteries by in operando scanning electrochemical microscopy. Chemical Communications, 52(41):6825–6828, 2016. [135] C. Lefrou and R. Cornut. Analytical expressions for quantitative scanning electrochemical microscopy (secm). ChemPhysChem, 11(3):547–556, February 2010. [136] C. Lefrou. A unified new analytical approximation for positive feedback currents with a microdisk secm tip. Journal of Electroanalytical Chemistry, 592(1):103–112, 2006. [137] R. Cornut and C. Lefrou. New analytical approximations for negative feedback currents with a microdisk secm tip. Journal of Electroanalytical Chemistry, 604(2):91–100, 2007. [138] R. Cornut and C. Lefrou. New analytical approximation of feedback approach curves with a microdisk secm tip and irreversible kinetic reaction at the substrate. Journal of Electroanalytical Chemistry, 621(2):178–184, 2008. [139] I. Yoon, S. Jurng, D. P. Abraham, B. L. Lucht, and P. R. Guduru. In situ measurement of the plane-strain modulus of the solid electrolyte interphase on lithium-metal anodes in ionic liquid electrolytes. Nano Letters, 18(9):5752–5759, 2018. 192 [140] M. Steinhauer, M. Stich, M. Kurniawan, B. K. Seidlhofer, M. Trapp, A. Bund, N. Wagner, and K. A. Friedrich. In situ studies of solid electrolyte interphase (sei) formation on crystalline carbon surfaces by neutron reflectometry and atomic force microscopy. ACS Applied Materials & Interfaces, 9(41):35794–35801, 2017. [141] E. Radvanyi, E. De Vito, W. Porcher, and S. J. S. Larbi. An xps/aes comparative study of the surface behaviour of nano-silicon anodes for li-ion batteries. Journal of Analytical Atomic Spectrometry, 29(6):1120–1131, 2014. [142] C. Y. Tang, Y. H. Ma, R. T. Haasch, J. H. Ouyang, and S. J. Dillon. Insights into solid-electrolyte interphase induced li-ion degradation from in situ auger electron spectroscopy. Journal of Physical Chemistry Letters, 8(24):6226–6230, 2017. [143] T. Y. Ma, X. N. Yu, H. Y. Li, W. G. Zhang, X. L. Cheng, W. T. Zhu, and X. P. Qiu. High volumetric capacity of hollow structured sno2@si nanospheres for lithium-ion batteries. Nano Letters, 17(6):3959–3964, 2017. [144] E. Allcorn and A. Manthiram. Thermal stability of sb and cu2sb anodes in lithiumion batteries. Journal of the Electrochemical Society, 162(9):A1778–A1786, 2015. [145] C. K. Chan, R. Ruffo, S. S. Hong, and Y. Cui. Surface chemistry and morphology of the solid electrolyte interphase on silicon nanowire lithium-ion battery anodes. Journal of Power Sources, 189(2):1132–1140, 2009. [146] K. Dokko, Y. Fujita, M. Mohamedi, M. Umeda, I. Uchida, and J. R. Selman. Electrochemical impedance study of li-ion insertion into mesocarbon microbead single particle electrode part ii. disordered carbon. Electrochimica Acta, 47(6):933–938, 2001. [147] S. Z. Xiong, K. Xie, Y. Diao, and X. B. Hong. On the role of polysulfides for a stable solid electrolyte interphase on the lithium anode cycled in lithium-sulfur batteries. Journal of Power Sources, 236:181–187, 2013. [148] M. Itagaki, N. Kobari, S. Yotsuda, K. Watanabe, S. Kinoshita, and M. Ue. In situ electrochemical impedance spectroscopy to investigate negative electrode of lithiumion rechargeable batteries. Journal of Power Sources, 135(1-2):255–261, 2004. [149] N. Shpigel, M. D. Levi, S. Sigalov, L. Daikhin, and D. Aurbach. In situ real-time mechanical and morphological characterization of electrodes for electrochemical energy storage and conversion by electrochemical quartz crystal microbalance with dissipation monitoring. Accounts of Chemical Research, 51(1):69–79, 2018. [150] K. Kwon, F. P. Kong, F. McLarnon, and J. W. Evans. Characterization of the sei on a carbon film electrode by combined eqcm and spectroscopic ellipsometry. Journal of the Electrochemical Society, 150(2):A229–A233, 2003. [151] F. Wang, J. Graetz, M. S. Moreno, C. Ma, L. J. Wu, V. Volkov, and Y. M. Zhu. Chemical distribution and bonding of lithium in intercalated graphite: Identification with optimized electron energy loss spectroscopy. ACS Nano, 5(2):1190–1197, 2011. [152] H. L. Zhang, F. Li, C. Liu, J. Tan, and H. M. Cheng. New insight into the solid electrolyte interphase with use of a focused ion beam. Journal of Physical Chemistry B, 109(47):22205–22211, 2005. 193 [153] J. Coyle, C. Apblett, M. Brumbach, J. Ohlhausen, and C. Stoldt. Structural and compositional characterization of rf magnetron cosputtered lithium silicate films: From li2si2o5 to lithium-rich li8sio6. Journal of Vacuum Science & Technology A, 35(6), 2017. [154] S. Perez-Villar, P. Lanz, H. Schneider, and P. Novak. Characterization of a model solid electrolyte interphase/carbon interface by combined in situ raman/fourier transform infrared microscopy. Electrochimica Acta, 106:506–515, 2013. [155] G. M. A. Girard, M. Hilder, N. Dupre, D. Guyomard, D. Nucciarone, K. Whitbread, S. Zavorine, M. Moser, M. Forsyth, D. R. MacFarlane, and P. C. Howlett. Spectroscopic characterization of the sei layer formed on lithium metal electrodes in phosphonium bis(fluorosulfonyl)imide ionic liquid electrolytes. ACS Applied Materials & Interfaces, 10(7):6719–6729, 2018. [156] T. Schwieters, M. Evertz, M. Mense, M. Winter, and S. Nowak. Lithium loss in the solid electrolyte interphase: Lithium quantification of aged lithium ion battery graphite electrodes by means of laser ablation inductively coupled plasma mass spectrometry and inductively coupled plasma optical emission spectroscopy. Journal of Power Sources, 356:47–55, 2017. [157] H. Tavassol, D. G. Cahill, and A. A. Gewirth. Potential induced structural changes and solid electrolyte interphase (sei) decomposition in sn anodes for li ion batteries. Abstracts of Papers of the American Chemical Society, 245, 2013. [158] J. A. Dura, J. E. Owejan, S. C. DeCaluwe, and J. P. Owejan. In situ neutron reflectometry measurements of the solid electrolyte interphase in li-ion batteries. Abstracts of Papers of the American Chemical Society, 245, 2013. [159] M. C. Smart, B. V. Ratnakumar, S. Surampudi, Y. Wang, X. Zhang, S. G. Greenbaum, A. Hightower, C. C. Ahn, and B. Fultz. Irreversible capacities of graphite in lowtemperature electrolytes for lithium-ion batteries. Journal of the Electrochemical Society, 146(11):3963–3969, 1999. [160] L. Cabo-Fernandez, F. Mueller, S. Passerini, and L. J. Hardwick. In situ raman spectroscopy of carbon-coated znfe2o4 anode material in li-ion batteries - investigation of sei growth. Chemical Communications, 52(20):3970–3973, 2016. [161] R. Schmitz, R. A. Muller, R. W. Schmitz, C. Schreiner, M. Kunze, A. Lex-Balducci, S. Passerini, and M. Winter. Sei investigations on copper electrodes after lithium plating with raman spectroscopy and mass spectrometry. Journal of Power Sources, 233:110–114, 2013. [162] H. Bulter, F. Peters, J. Schwenzel, and G. Wittstock. Spatiotemporal changes of the solid electrolyte interphase in lithium-ion batteries detected by scanning electrochemical microscopy. Angewandte Chemie-International Edition, 53(39):10531–10535, 2014. [163] R. L. Sacci, J. Black, N. Balke, and N. J. Dudney. In situ electrochemical tem for quantitative nanoscale imaging dynamics of solid electrolyte interphase and lithium electrodeposition. Microscopy Microanalysis, 21(S3):2437–2438, 2015. [164] B. L. Mehdi, J. Qian, E. Nasybulin, C. Park, D. A. Welch, R. Faller, H. Mehta, W. A. Henderson, W. Xu, C. M. Wang, J. E. Evans, J. Liu, J. G. Zhang, K. T. Mueller, and N. D. 194 Browning. Observation and quantification of nanoscale processes in lithium batteries by operando electrochemical (s)tem. Nano Letters, 15(3):2168–2173, 2015. [165] S. Hy, Felix, Y. H. Chen, J. Y. Liu, J. Rick, and B. J. Hwang. In situ surface enhanced raman spectroscopic studies of solid electrolyte interphase formation in lithium ion battery electrodes. Journal of Power Sources, 256:324–328, 2014. [166] S. Tang, Y. Gu, J. Yi, Z. C. Zeng, S. Y. Ding, J. W. Yan, D. Y. Wu, B. Ren, Z. Q. Tian, and B. W. Mao. An electrochemical surface-enhanced raman spectroscopic study on nanorod-structured lithium prepared by electrodeposition. Journal of Raman Spectroscopy, 47(9):1017–1023, 2016. [167] H. Li, Y. J. Mo, N. Pei, X. X. Xu, X. J. Huang, and L. Q. Chen. Surface-enhanced raman scattering study on passivating films of ag electrodes in lithium batteries. Journal of Physical Chemistry B, 104(35):8477–8480, 2000. [168] G. F. Li, H. Li, Y. J. Mo, L. Q. Chen, and X. J. Huang. Further identification to the set film on ag electrode in lithium batteries by surface enhanced raman scattering (sers). Journal of Power Sources, 104(2):190–194, 2002. [169] T. Yoon, M. S. Millen, B. S. Parimalam, and B. L. Lucht. Thermal decomposition of the solid electrolyte interphase (sei) on silicon electrodes for lithium ion batteries. Chemistry of Materials, 29(7):3237–3245, 2017. [170] C. H. Doh, D. H. Kim, J. H. Lee, D. J. Lee, B. S. Jin, H. S. Kim, S. I. Moon, Y. Hwang, and A. Veluchamy. Thermal behavior of lixcoo2 cathode and disruption of solid electrolyte interphase film. Bulletin of the Korean Chemical Society, 30(4):783–786, 2009. [171] S. Q. Shi, P. Lu, Z. Y. Liu, Y. Qi, L. G. Hector, H. Li, and S. J. Harris. Direct calculation of li-ion transport in the solid electrolyte interphase. Journal of the American Chemical Society, 134(37):15476–15487, 2012. [172] A. Augustsson, M. Herstedt, J. H. Guo, K. Edstrom, G. V. Zhuang, P. N. Ross, J. E. Rubensson, and J. Nordgren. Solid electrolyte interphase on graphite li-ion battery anodes studied by soft x-ray spectroscopy. Physical Chemistry Chemical Physics, 6(16):4185–4189, 2004. [173] C. Yogi, D. Takamatsu, K. Yamanaka, H. Arai, Y. Uchimoto, K. Kojima, I. Watanabe, T. Ohta, and Z. Ogumi. Soft x-ray absorption spectroscopic studies with different probing depths: Effect of an electrolyte additive on electrode surfaces. Journal of Power Sources, 248:994–999, 2014. [174] A. Jarry, S. Gottis, Y. S. Yu, J. Roque-Rosell, C. Kim, J. Cabana, J. Kerr, and R. Kostecki. The formation mechanism of fluorescent metal complexes at the lixni0.5mn1.5o4delta/carbonate ester electrolyte interface. Journal of the American Chemical Society, 137(10):3533–3539, 2015. [175] S. Chattopadhyay, A. L. Lipson, H. J. Karmel, J. D. Emery, T. T. Fister, P. A. Fenter, M. C. Hersam, and M. J. Bedzyk. In situ x-ray study of the solid electrolyte interphase (sei) formation on graphene as a model li-ion battery anode. Chemistry of Materials, 24(15):3038–3043, 2012. 195 [176] J. Zhang, R. Wang, X. C. Yang, W. Lu, X. D. Wu, X. P. Wang, H. Li, and L. W. Chen. Direct observation of inhomogeneous solid electrolyte interphase on mno anode with atomic force microscopy and spectroscopy. Nano Letters, 12(4):2153–2157, 2012. [177] N. Weadock, N. Varongchayakul, J. Y. Wan, S. Lee, J. Seog, and L. B. Hu. Determination of mechanical properties of the sei in sodium ion batteries via colloidal probe microscopy. Nano Energy, 2(5):713–719, 2013. [178] T. Miyatani and M. Fujihira. Calibration of surface stress measurements with atomic force microscopy. Journal of Applied Physics, 81(11):7099–7115, 1997. [179] H. M. Xie, W. Qiu, H. B. Song, and J. H. Tian. In situ measurement of the deformation and elastic modulus evolution in si composite electrodes during electrochemical lithiation and delithiation. Journal of the Electrochemical Society, 163(13):A2685–A2690, 2016. [180] S. J. Lee, J. K. Lee, S. H. Chung, H. Y. Lee, S. M. Lee, and H. K. Baik. Stress effect on cycle properties of the silicon thin-film anode. Journal of Power Sources, 97-8:191–193, 2001. [181] J. J. Brown, S. H. Lee, J. L. Xiao, and Z. C. Wu. Observations of stress accumulation and relaxation in solid-state lithiation and delithiation of suspended si microcantilevers. Physica Status Solidi a-Applications and Materials Science, 213(8):2156–2168, 2016. [182] J. M. Rosolen and F. Decker. Stress in carbon film electrodes during li+ electrochemical intercalation. Journal of the Electrochemical Society, 143(8):2417–2421, 1996. [183] X. J. Zhang and D. G. Cahill. Measurements of interface stress of silicon dioxide in contact with water-phenol mixtures by bending of microcantilevers. Langmuir, 22(21):9062–9066, 2006. [184] Y. Qi and S. J. Harris. In situ observation of strains during lithiation of a graphite electrode. Journal of the Electrochemical Society, 157(6):A741–A747, 2010. [185] C. Y. Dai, C. S. Li, H. Y. Huang, Z. Wang, X. X. Zhu, X. B. Liao, X. Chen, Y. Pan, and D. N. Fang. In situ strain measurements and stress analysis of sio@c composite electrodes during electrochemical cycling by using digital image correlation. Solid State Ionics, 331:56–65, 2019. [186] E. M. C. Jones, M. N. Silberstein, S. R. White, and N. R. Sottos. In situ measurements of strains in composite battery electrodes during electrochemical cycling. Experimental Mechanics, 54(6):971–985, 2014. [187] W. Yang, H. M. Xie, B. Q. Shi, H. B. Song, W. Qiu, and Q. Zhang. In-situ experimental measurements of lithium concentration distribution and strain field of graphite electrodes during electrochemical process. Journal of Power Sources, 423:174–182, 2019. [188] M. Seo, M. Aomi, and K. Yoshida. A combined piezoelectric and eqcm study of underpotential deposition of silver on gold electrodes. Electrochimica Acta, 39(89):1039–1044, 1994. 196 [189] C. J. Bae, A. Manandhar, P. Kiesel, and A. Raghavan. Monitoring the strain evolution of lithium-ion battery electrodes using an optical fiber bragg grating sensor. Energy Technology, 4(7):851–855, 2016. [190] N. A. Padilla, M. T. Rea, M. Foy, S. P. Upadhyay, K. A. Desrochers, T. Derus, K. A. Knapper, N. H. Hunter, S. Wood, D. A. Hinton, A. C. Cavell, A. G. Masias, and R. H. Goldsmith. Tracking lithium ions via widefield fluorescence microscopy for battery diagnostics. ACS Sensors, 2(7):903–908, July 2017. [191] C. Chen, T. Zhou, D. L. Danilov, L. Gao, S. Benning, N. Schön, S. Tardif, H. Simons, F. Hausen, T. U. Schülli, R. A. Eichel, and P. H. L. Notten. Impact of dual-layer solidelectrolyte interphase inhomogeneities on early-stage defect formation in si electrodes. Nature Communications, 11(1), 2020-07-01. [192] Q. Zhang, X. Xiao, W. Zhou, Y. Cheng, and M. W. Verbrugge. Toward high cycle efficiency of silicon-based negative electrodes by designing the solid electrolyte interphase. Advanced Energy Materials, 5(5):1401398, March 2015. [193] B. Han and D. Post. Immersion interferometer for microscopic moire interferometry. Experimental Mechanics, 32(1):38–41, 1992. [194] J. Morton, D. Post, B. Han, and M. Y. Tsai. A localized hybrid method of stressanalysis - a combination of moire interferometry and fem. Experimental Mechanics, 30(2):195–200, 1990. [195] D. Post and B. Han. Moiré Interferometry. Springer, New York City, 42nd Street, 2008. [196] K. J. Zhao, G. A. Tritsaris, M. Pharr, W. L. Wang, O. Okeke, Z. G. Suo, J. J. Vlassak, and E. Kaxiras. Reactive flow in silicon electrodes assisted by the insertion of lithium. Nano Letters, 12(8):4397–4403, 2012. [197] M. Pharr, Z. G. Suo, and J. J. Vlassak. Variation of stress with charging rate due to strain-rate sensitivity of silicon electrodes of li-ion batteries. Journal of Power Sources, 270:569–575, 2014. [198] N. S. Choi, J. G. Han, S. Y. Ha, I. Park, and C. K. Back. Recent advances in the electrolytes for interfacial stability of high-voltage cathodes in lithium-ion batteries. RSC Advances, 5(4):2732–2748, 2015. [199] C. J. Yu, X. Li, T. Ma, J. P. Rong, R. J. Zhang, J. Shaffer, Y. H. An, Q. Liu, B. Q. Wei, and H. Q. Jiang. Silicon thin films as anodes for high-performance lithium-ion batteries with effective stress relaxation. Advanced Energy Materials, 2(1):68–73, 2012. [200] V. A. Sethuraman, N. Van Winkle, D. P. Abraham, A. F. Bower, and P. R. Guduru. Real-time stress measurements in lithium-ion battery negative-electrodes. Journal of Power Sources, 206:334–342, 2012. [201] G. D. Sun, M. Zhang, Z. Y. Xue, Q. L. Guo, D. Chen, Z. Q. Mu, L. X. Dong, X. Wang, and Z. F. Di. Strain redistribution in free-standing bridge structure released from strained silicon-on-insulator. Applied Physics Letters, 105(19), 2014. 197 [202] Q. Wang, Y. Li, B. Bai, W. Mao, Z. G. Wang, and N. F. Ren. Effects of silicon dioxide surface roughness on raman characteristics and mechanical properties of graphene. RSC Advances, 4(98):55087–55093, 2014. [203] P. Zorabedian and F. Adar. Measurement of local stress in laser-recrystallized lateral epitaxial silicon films over silicon dioxide using raman-scattering. Applied Physics Letters, 43(2):177–179, 1983. [204] G. Zatryb, J. Misiewicz, P. R. J. Wilson, J. Wojcik, P. Mascher, and A. Podhorodecki. Stress transition from compressive to tensile for silicon nanocrystals embedded in amorphous silica matrix. Thin Solid Films, 571:18–22, 2014. [205] L. Jaeckel, G. Lang, and K. E. Heusler. A new interferometric method to investigate surface energies at solid electrodes. Electrochimica Acta, 39(8-9):1031–1038, 1994. [206] S. I. Pyun, J. Y. Go, and T. S. Jang. An investigation of intercalation-induced stresses generated during lithium transport through li1-delta coo2 film electrode using a laser beam deflection method. Electrochimica Acta, 49(25):4477–4486, 2004. [207] J. L. Langer, J. Economy, and D. G. Cahill. Absorption of water and mechanical stress in immobilized poly(vinylbenzyltrialkylammonium chloride) thin films. Macromolecules, 45(7):3205–3212, 2012. [208] E. Pomerantseva, H. Jung, M. Gnerlich, S. Baron, K. Gerasopoulos, and R. Ghodssi. A mems platform for in situ, real-time monitoring of electrochemically induced mechanical changes in lithium-ion battery electrodes. Journal of Micromechanics and Microengineering, 23(11), 2013. [209] C. O’Mahony, M. Hill, M. Brunet, R. Duane, and A. Mathewson. Characterization of micromechanical structures using white-light interferometry. Measurement Science and Technology, 14(10):1807–1814, 2003. [210] S. Wolf, N. A. Canas, and K. A. Friedrich. In situ x-ray diffraction and stress analysis of solid oxide fuel cells. Fuel Cells, 13(3):404–409, 2013. [211] S. Misra, N. Liu, J. Nelson, S. S. Hong, Y. Cui, and M. F. Toney. In situ x-ray diffraction studies of (de)lithiation mechanism in silicon nanowire anodes. ACS Nano, 6(6):5465–5473, 2012. [212] F. Wang, L. J. Wu, B. Key, X. Q. Yang, C. P. Grey, Y. M. Zhu, and J. Graetz. Electrochemical reaction of lithium with nanostructured silicon anodes: A study by in-situ synchrotron x-ray diffraction and electron energy-loss spectroscopy. Advanced Energy Materials, 3(10):1324–1331, 2013. [213] M. Winter, B. Barnett, and K. Xu. 118(23):11433–11456, December 2018. Before li ion batteries. Chemical Reviews, [214] G. He, R. Ciez, P. Moutis, S. Kar, and J. F. Whitacre. The economic end of life of electrochemical energy storage. Applied Energy, 273:115151, 2020. [215] Z. Liu, J. Song, J. Kubal, N. Susarla, K. W. Knehr, E. Islam, P. Nelson, and S. Ahmed. Comparing total cost of ownership of battery electric vehicles and internal combustion engine vehicles. Energy Policy, 158:112564, 2021. 198 [216] K. Xu. Nonaqueous liquid electrolytes for lithium-based rechargeable batteries. Chemical Reviews, 104(10):4303–4417, 2004. [217] E. V. Thomas, I. Bloom, J. P. Christophersen, and V. S. Battaglia. Rate-based degradation modeling of lithium-ion cells. Journal of Power Sources, 206:378–382, 2012. [218] J. E. Harlow, X. Ma, J. Li, E. Logan, Y. Liu, N. Zhang, L. Ma, S. L. Glazier, M. M. E. Cormier, M. Genovese, S. Buteau, A. Cameron, J. E. Stark, and J. R. Dahn. A wide range of testing results on an excellent lithium-ion cell chemistry to be used as benchmarks for new battery technologies. Journal of The Electrochemical Society, 166(13):A3031–A3044, 2019. [219] U.S. Department of Energy. U.s. department of energy vehicle technologies program. United States Advanced Battery Consortium Battery Test Manual For Electric Vehicles, Revision 3.1, 2015. [220] T. Gewald, M. Lienkamp, D. Lehmkuhl, and A. Hahn. 14th international conference on ecological vehicles and renewable energies (ever). Institute of Electrical and Electronics Engineers Inc., 2019. [221] B. Han, Y. Zhang, C. Liao, S. E. Trask, X. Li, R. Uppuluri, J. T. Vaughey, B. Key, and F. Dogan. Probing the reactivity of the active material of a li-ion silicon anode with common battery solvents. ACS Applied Materials Interfaces, 13(24):28017–28026, June 2021. [222] J. D. McBrayer, C. A. Apblett, K. L. Harrison, K. R. Fenton, and S. D. Minteer. Mechanical studies of the solid electrolyte interphase on anodes in lithium and lithium ion batteries. Nanotechnology, 32, Sep 2021. [223] A. Tornheim, S. E. Trask, and Z. Zhang. Evaluation of electrolyte oxidation stability on charged lini0.5co.2mn0.3o cathode surface through potentiostatic holds. Journal of The Electrochemical Society, 163(8):A1717–A1722, 2016. [224] A. Tornheim, J. C. Garcia, R. Sahore, H. Iddir, I. Bloom, and Z. Zhang. Decomposition of phosphorus-containing additives at a charged nmc surface through potentiostatic holds. Journal of The Electrochemical Society, 166(4):A440–A447, 2019. [225] N. R. Vadivel, S. Ha, M. He, D. Dees, S. Trask, B. Polzin, and K. G. Gallagher. On leakage current measured at high cell voltages in lithium-ion batteries. Journal of The Electrochemical Society, 164(2):A508–A517, 2017. [226] M. Lewerenz, J. Münnix, J. Schmalstieg, S. Käbitz, M. Knips, and D. U. Sauer. Systematic aging of commercial lifepo4—graphite cylindrical cells including a theory explaining rise of capacity during aging. Journal of Power Sources, 345:254–263, 2017. [227] M. Theiler, C. Endisch, and M. Lewerenz. Float current analysis for fast calendar aging assessment of 18650 li(nicoal)o2/graphite cells. Batteries, 7(2), 2021. [228] A. Zülke, Y. Li, P. Keil, R. Burrell, S. Belaisch, M. Nagarathinam, M. P. Mercer, and H. E. Hoster. High-energy nickel-cobalt-aluminium oxide (nca) cells on idle: Anodeversus cathode-driven side reactions. Batteries and Supercaps, 4(6):934–947, June 2021. 199 [229] A. Tornheim, S. Sharifi-Asl, J. C. Garcia, J. Bareño, H. Iddir, R. Shahbazian-Yassar, and Z. Zhang. Effect of electrolyte composition on rock salt surface degradation in nmc cathodes during high-voltage potentiostatic holds. Nano Energy, 55:216–225, 2019. [230] M. Lewerenz, S. Käbitz, M. Knips, J. Münnix, J. Schmalstieg, A. Warnecke, and D. U. Sauer. New method evaluating currents keeping the voltage constant for fast and highly resolved measurement of arrhenius relation and capacity fade. Journal of Power Sources, 353:144–151, 2017. [231] S. S. Zhang. Understanding of performance degradation of lini0.80co0.10mn0.10o2 cathode material operating at high potentials. Journal of Energy Chemistry, 41:135–141, 2020. [232] H. Zhang, H. Liu, L. F. J. Piper, M. S. Whittingham, and G. Zhou. Oxygen loss in layered oxide cathodes for li-ion batteries: Mechanisms, effects, and mitigation. Chemical Reviews, 122(6):5641–5681, March 2022. [233] M. C. Schulze, G. M. Carroll, T. R. Martin, K. Sanchez-Rivera, F. Urias, and N. R. Neale. Hydrophobic versus hydrophilic interfacial coatings on silicon nanoparticles teach us how to design the solid electrolyte interphase in silicon-based li-ion battery anodes. ACS Applied Energy Materials, 4(2):1628–1636, February 2021. [234] M. T. F. Rodrigues, J. A. Gilbert, K. Kalaga, and D. P. Abraham. Insights on the cycling behavior of a highly-prelithiated silicon-graphite electrode in lithium-ion cells. Journal of Physics: Energy, 2(2):024002, 2020. [235] M. Wetjen, S. Solchenbach, D. Pritzl, J. Hou, V. Tileli, and H. A. Gasteiger. Morphological changes of silicon nanoparticles and the influence of cutoff potentials in silicon-graphite electrodes. Journal of The Electrochemical Society, 165(7):A1503–A1514, 2018. [236] W. M. Dose, V. A. Maroni, M. J. Piernas-Muñoz, S. E. Trask, I. Bloom, and C. S. Johnson. Assessment of li-inventory in cycled si-graphite anodes using lifepo4 as a diagnostic cathode. Journal of The Electrochemical Society, 165(10):A2389–A2396, 2018. [237] A. Tornheim and D. C. O’Hanlon. What do coulombic efficiency and capacity retention truly measure? a deep dive into cyclable lithium inventory, limitation type, and redox side reactions. Journal of The Electrochemical Society, 167(11):110520, 2020. [238] R. Jung, M. Metzger, F. Maglia, C. Stinner, and H. A. Gasteiger. Oxygen release and its effect on the cycling stability of linixmnycozo2 (nmc) cathode materials for li-ion batteries. Journal of The Electrochemical Society, 164(7):A1361–A1377, 2017. [239] A. Tornheim, M. He, C. Su, and Z. Zhang. The role of additives in improving performance in high voltage lithium-ion batteries with potentiostatic holds. Journal of The Electrochemical Society, 164(1):A6366–A6372, 2017. [240] J. Schmalstieg, S. Käbitz, M. Ecker, and D. U. Sauer. A holistic aging model for li(nimnco)o2 based 18650 lithium-ion batteries. Journal of Power Sources, 257:325–334, 2014. 200 [241] P.M. Attia, W.C. Chueh, and S.J. Harris. Revisiting the t¡sup¿0.5¡/sup¿ dependence of sei growth. Journal of The Electrochemical Society, 167(9):090535, 2020. [242] P. Gasper, K. Gering, E. Dufek, and K. Smith. Challenging practices of algebraic battery life models through statistical validation and model identification via machinelearning. Journal of The Electrochemical Society, 168(2):020502, 2021. [243] B. R. Long, S. G. Rinaldo, K. G. Gallagher, D. W. Dees, S. E. Trask, B. J. Polzin, A. N. Jansen, D. P. Abraham, I. Bloom, J. Bareño, and J. R. Croy. Enabling highenergy, high-voltage lithium-ion cells: Standardization of coin-cell assembly, electrochemical testing, and evaluation of full cells. Journal of The Electrochemical Society, 163(14):A2999–A3009, 2016. [244] B. Gyenes, D. A. Stevens, V. L. Chevrier, and J. R. Dahn. Understanding anomalous behavior in coulombic efficiency measurements on li-ion batteries. Journal of the Electrochemical Society, 162(3):A278–A283, 2015. [245] R. Burrell, A. Zulke, P. Keil, and H. Hoster. Communication–identifying and managing reversible capacity losses that falsify cycle ageing tests of lithium-ion cells. Journal of The Electrochemical Society, 167(13):130544, 2020. [246] C. Delacourt and M. Safari. Life simulation of a graphite/lifepo¡sub¿4¡/sub¿cell under cycling and storage. Journal of The Electrochemical Society, 159(8):A1283–A1291, 2012. [247] R. Spotnitz. Simulation of capacity fade in lithium-ion batteries. Journal of Power Sources, 113(1):72–80, 2003. [248] F. Leng, C. M. Tan, and M. Pecht. Effect of temperature on the aging rate of li ion battery operating above room temperature. Scientific Reports, 5(1):12967, 2015. [249] Yao Wu, Peter Keil, Simon F. Schuster, and Andreas Jossen. Impact of temperature and discharge rate on the aging of a licoo2lini0.8co0.15al0.05o2 lithium-ion pouch cell. Journal of The Electrochemical Society, 164(7):A1438–A1445, 2017. [250] I. Bloom, B. W. Cole, J. J. Sohn, S. A. Jones, E. G. Polzin, V. S. Battaglia, G. L. Henriksen, C. Motloch, R. Richardson, T. Unkelhaeuser, D. Ingersoll, and H. L. Case. An accelerated calendar and cycle life study of li-ion cells. Journal of Power Sources, 101(2):238–247, 2001. [251] C. Stetson, T. Yoon, J. Coyle, W. Nemeth, M. Young, A. Norman, S. Pylypenko, C. M. Ban, C. S. Jiang, M. Al-Jassim, and A. Burrell. Three-dimensional electronic resistivity mapping of solid electrolyte interphase on si anode materials. Nano Energy, 55:477– 485, 2019. [252] S. Grolleau, A. Delaille, H. Gualous, P. Gyan, R. Revel, J. Bernard, E. RedondoIglesias, and J. Peter. Calendar aging of commercial graphite/lifepo4 cell - predicting capacity fade under time dependent storage conditions. Journal of Power Sources, 255:450–458, 2014. [253] M. Ecker, J. B. Gerschler, J. Vogel, S. Käbitz, F. Hust, P. Dechent, and D. U. Sauer. Development of a lifetime prediction model for lithium-ion batteries based on extended accelerated aging test data. Journal of Power Sources, 215:248–257, 2012. 201 [254] S. L. Hahn, M. Storch, R. Swaminathan, B. Obry, J. Bandlow, and K. P. Birke. Quantitative validation of calendar aging models for lithium-ion batteries. Journal of Power Sources, 400:402–414, 2018. [255] X. Li, A. M. Colclasure, D. P. Finegan, D. Ren, Y. Shi, X. Feng, L. Cao, Y. Yang, and K. Smith. Degradation mechanisms of high capacity 18650 cells containing si-graphite anode and nickel-rich nmc cathode. Electrochimica Acta, 297:1109–1120, 2019. [256] D. Li, D.L. Danilov, J. Xie, L. Raijmakers, L. Gao, Y. Yang, and P.H.L. Notten. Degradation mechanisms of c6/lifepo4 batteries: Experimental analyses of calendar aging. Electrochimica Acta, 190:1124–1133, 2016. [257] M. Safari and C. Delacourt. Aging of a commercial graphite/lifepo4 cell. Journal of The Electrochemical Society, 158(10):A1123, 2011. [258] P. G. Ifju and B. Han. Recent applications of moiré interferometry. Experimental Mechanics, 50(8):1129–1147, 2010. [259] M Luo, M. T. F. F. Rodrigues, L. L. Shaw, and D. P. Abraham. Examining effects of negative to positive capacity ratio in three-electrode lithium-ion cells with layered oxide cathode and si anode. ACS Applied Energy Materials, May 2022. [260] S. J. An, J. Li, C. Daniel, D. Mohanty, S. Nagpure, and D. L. Wood. The state of understanding of the lithium-ion-battery graphite solid electrolyte interphase (sei) and its relationship to formation cycling. Carbon, 105:52–76, 2016. [261] S. S. Zhang. A review on electrolyte additives for lithium-ion batteries. Journal of Power Sources, 162(2):1379–1394, 2006. [262] H. Lin, H. Noguchi, and K. Uosaki. Effects of hf on the lithiation behavior of the silicon anode in lipf6 organic electrolyte solution. ACS Omega, 5(5):2081–2087, February 2020. [263] Y. Okuno, K. Ushirogata, K. Sodeyama, and Y. Tateyama. Decomposition of the fluoroethylene carbonate additive and the glue effect of lithium fluoride products for the solid electrolyte interphase: an ab initio study. Physical Chemistry Chemical Physics, 18(12):8643–8653, 2016. [264] K. P. C. Yao, J. S. Okasinski, K. Kalaga, J. D. Almer, and D. P. Abraham. Operando quantification of (de)lithiation behavior of silicon-graphite blended electrodes for lithium-ion batteries. Advanced Energy Materials, 9(8):1803380, February 2019. [265] A. T. Phan, A. E. Gheribi, and P. Chartrand. Modeling of coherent phase transformation and particle size effect in lifepo4 cathode material and application to the charging/discharging process. Electrochimica Acta, 295:632–644, 2019. [266] W. Zhang, T. H. Cai, and B. W. Sheldon. The impact of initial sei formation conditions on strain-induced capacity losses in silicon electrodes. Advanced Energy Materials, 9(5), 2019. [267] J. Yang, M. T. F. Rodrigues, S. Son, J. C. Garcia, K. Liu, J. Gim, H. Iddir, D. P. Abraham, Z. Zhang, and C. Liao. Dual-salt electrolytes to effectively reduce impedance rise of high-nickel lithium-ion batteries. ACS Applied Materials Interfaces, 13(34):40502–40512, September 2021. 202 [268] E. Sarasketa-Zabala, I. Gandiaga, L. M. Rodriguez-Martinez, and I. Villarreal. Calendar ageing analysis of a lifepo4/graphite cell with dynamic model validations: Towards realistic lifetime predictions. Journal of Power Sources, 272:45–57, 2014.