Spectroelectrochemical analysis of the Li-ion battery solid electrolyte interphase using simulated Raman spectra

Juni 2020

Spectroelectrochemical analysis

of the Li-ion battery solid electrolyte

interphase using simulated Raman

spectra

Teknisk- naturvetenskaplig fakultet UTH-enheten Besöksadress: Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0 Postadress: Box 536 751 21 Uppsala Telefon: 018 – 471 30 03 Telefax: 018 – 471 30 00 Hemsida: http://www.teknat.uu.se/student

Spectroelectrochemical analysis of the Li-ion battery

solid electrolyte interphase using simulated Raman

spectra

Edvin Andersson

Lithium Ion Batteries (LIBs) are important in today's society, powering cars and mobile devices. LIBs consist of a negative anode commonly made of graphite, and a positive cathode commonly made from transition metal oxides. Between these electrodes are separators and organic solvent based electrolyte. Due to the high potential of LIBs the electrolyte is reduced at the anode. The electrolyte reduction results in the formation of a layer called the Solid Electrolyte Interphase (SEI), which prohibits the further breakdown of the electrolyte. Despite being researched for over 50 years, the composition formation of the SEI is still poorly understood. The aim of this project is to develop strategies for efficient identification and classification of various active and intermediate components in the SEI, to, in turn, gain an

understanding of the reactions taking place, which will help find routes to stabilize and tailor the composition of the SEI layer for long-term stability and optimal battery performance. For a model gold/li-ion battery electrolyte system, Raman spectra will be obtained using Surface Enhanced Raman Spectroscopy (SERS) in a

spectroelectrochemical application where the voltage of the working gold electrode is swept from high to low potentials. Spectra of common components of the SEI as well as similar compounds will be simulated using Density Functional Theory (DFT). The DFT data is also used to calculate the spontaneity of reactions speculated to form the SEI. The simulated data will be validated by comparing it to experimental spectra from pure substances. The spectroelectrochemical SERS results show a clear formation of Li-carbonate at the SERS substrate, as well as the decomposition of the electrolyte into other species, according to the simulated data. It is however shown that there are several issues when modelling spectra, that makes it harder to correlate the simulated spectra with the spectroelectrochemical spectra. These issues include limited knowledge of the structure of the compounds thought to form on the anode surface, and incorrect choices in simulational parameters. To solve these issues, more work is needed in these areas, and the spectroelectrochemical methods used in this thesis needs to be combined with other experimental methods to narrow down the amount of compounds to be modelled. More work is also needed to avoid impurities in the electrolyte. Impurities leads to a thick inorganic layer which prohibits the observation of species in the organic layer.

Examinator: Åsa Kassman

Ämnesgranskare: Tomas Edvinsson

Edvin Andersson

Litiumjon-batterier är en stor del av våra digitala tillvaro, och tillser mycket av vår bärbara vardagse-lektronik med energi. När, på senare år, milöjöfrågor har blivit allt viktigare, har batterierna också börjat dyka upp i våra transportmedel. Anledningen till detta är deras stora energi till vikt- och volymförhållande. Denna egenskap kommer i grunden från att oxidationen av litiumatomen skapar en stor negativ potential, eller spänning. Spänning multiplicerat med laddning är energi, och detta tillsammans med att litium är ett lätt grundämne ger den tidigarenämnda egenskapen.

Litiumjonbatterier består av två elektroder, en negativ (anod) och en positiv (katod). Mel-lan dessa elektroder går strömmen, som består av elektroner. Inuti batteriet går också ström som består av positivt laddade joner. Dessa joner bärs från ena elektroden till den andra med hjälp av elektrolyten. Elektrolyten får ej leda elektronisk ström, eftersom elektronerna måste tvingas genom en krets utanför batteriet för att vi skall kunna utnyttja den energi som de bär på. Den sista delen som är viktig i ett batteri är separatorn, ett membran som ej heller leder elektronisk ström men också håller isär de två elektroderna mekaniskt, så att det ej blir kortslutning. När det kommer till anoden så händer det något väldigt intressant vid dess yta. Som tidigare nämnt så är anoden den negativa elektroden, och det är här som oxidationen av litiumatomer sker. Den negativa spänningen som skapas här är ju väldigt bra för energiinnehållet i batteriet, men den skapar också problem. Anoden är full av elektroner, och ju mer negativ den blir, ju mindre lust har elektronerna att stanna kvar i anoden, vilket får dem att hoppa ut i elektrolyten. Detta startar kemiska reaktioner vilka leder till nedbrytningen av elektrolyten, vilket hade varit väldigt ofördelaktigt i vattenbaserade elektrolyter, då denna nedbrytning hade fortsatt tills batteriet var oanvändbart. I litiumjonbatterier har man dock karbonatelektrolyter, som bildar ett gränsskikt på anoden när de bryts ned. Detta gränsskikt kallas Solid Elektrolyte Interphase (SEI) och förhindrar fortsatt nedbrytning genom att förhindra elektronöverföring från anoden till elektrolyten. Den leder dock fortfarande joner, vilket innebär att batteriet kan fortsätta att användas.

Gränsskiktet tros bestå av ett inorganiskt lager närmast anodens yta, och ett organiskt lager lite längre ut. Skiktets exakta sammansättning och de reaktioner som leder till det är fortfarande relativt okända. I detta arbete utforskas en specifik strategi för att studera detta lager. Strate-gin innefattar användandet av simulerade spektra och termodynamisk data för analysen av data från spektroelektrokemiska experiment, dvs experiment där spektroskopi (metoder som kollar på molekylers vibrationer) används för att analysera prov som utsätts för applicerad potential. I detta arbete används Surface Enhanced Raman Spectroscopy (SERS), vilken förstärker signalen från vanlig Ramanspektroskopi precis vid en yta, vilket är perfekt för undersökning av elektroder, eftersom man är intresserad av fenomen vid ytan. SERS funkar bara om ytan är av metall och har en nanostruktur, så att undersöka en vanlig grafitanod fungerar ej. En guldyta får istället agera anod genom att applicera en potential, och att använda modellytor för att efterlikna riktiga förhållanden kallas operando. För att simulera spektra av de ämnen som förväntas bildas i dessa operandoexperiment så behövs en metod som kan räkna ut ämnenas atomistiska egenskaper. I detta arbete användes en metod som heter täthetsfunktionalteori (DFT), vilken räknar ut ett systems kvantmekaniska egenskaper, vilket är precis vad som behövs. De vibrationer som räknats ut med denna metod sorterades enligt typ och användes som referens vid tolkningen av operandoexperi-mentens SERS-data.

I detta arbete konstaterades att litiumkarbonat, ett vanligt ämne i det inorganiska larget i SEI, hade bildats nära elektrodens yta. Andra oidentifierbara ämnen bildades också. Dessa ämnen hade vibrationer som skulle kunna tillskrivas till karbonatgrupper. På grund av orenheter i elektrolyten så blev troligen detta lager så tjockt att det utanför detta lager ej kunde ses med SERS.

vilka gaser utvecklas under operandoexperimenten, skulle kunna vara till stor nytta, då dessa skulle utesluta vissa reaktionsvägar till vissa ämnen. Mer simulerad data behövs dock också för att denna strategi skall fungera, eftersom den ringa data som fåtts hittills ger för stor osäkerhet.

I stort har dessa experiment, alltså att använda med DFT simulerade spektra för att tolka spektroelektrokemisk operando SERS-data visat att analysen av spektroelektroemiska metoder med hjälp av simulerade spektra har potential att vara väldigt användbart, om metoderna förfinas mer och mer data simuleras.

Examensarbete 30 hp på civilingenjörsprogrammet Teknisk fysik med materialvetenskap

1 Introduction 1

1.1 Lithium ion batteries . . . 1

1.1.1 Anodes and Electrolyte materials . . . 2

1.1.2 Solid Electrolyte Interphase . . . 3

1.2 Aim of the thesis . . . 6

1.3 Strategy . . . 6

2 Theory 8 2.1 Raman spectroscopy . . . 8

2.1.1 Surface Enhanced Raman Spectroscopy . . . 8

2.1.2 Spectroelectrochemistry . . . 9

2.2 Density Functional Theory . . . 9

2.2.1 Vibrational spectroscopy simulations using DFT . . . 10

2.2.2 Electrochemistry and thermodynamics calculation . . . 10

3 Experimental setups and procedures 11 3.1 Raman measurements . . . 11

3.2 Density functional theory calculations . . . 11

3.2.1 Thermodynamic system . . . 12

3.3 SERS . . . 13

4 Infrastructure for SERS analysis 14 4.1 Analysis using modelled data . . . 14

4.2 Thermodynamic calculations . . . 16

4.3 Validation . . . 18

4.3.1 Comparing real and simulated data . . . 19

5 Operando Surface Enhanced Raman Spectroscopy results 22

6 Conclusions, general discussion and future work 24

1

Introduction

1.1

Lithium ion batteries

Lithium ion (Li-ion) batteries are a common battery technology found in numerous electric devices, like cars or cell phones. Cell phones are able to have long battery life owing to Li-ion batteries, and electric cars, which are very much on the rise due to environmental concerns regarding fossil fuels, can only compete with regular cars thanks to Li-ion batteries. A Li-ion battery consist of two electrodes, with electrolyte impregnated in a polymer-based separator between the two different electrodes, as seen in figure 1.

Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Li+ Anode Catho de Separator

e-e

-

+

Figure 1: Schematic of a typical Li-ion battery during discharge.

There is an electric potential between the electrodes (cell potential), due to their different composi-tions. During the discharge, electrons will flow from the negatively charged electrode (anode) to the positively charged electrode (cathode) via an external circuit, while ions will flow inside the battery to compensate for the charge. During charging, this process is reversed, i.e. the ions and electrons travel from the cathode to the anode1_{. The electrolyte should have a a high ionic conductivity and} a low electronic conductivity. It should also preferably be stable at the cell potential, as well as be able to soak into the separator. The task of the separator is to separate the two electrodes so that they do not short circuit. It also needs to be able to let the electrolyte through with the ions, as well as not be conducting to electrons, like the electrolyte. Common materials for the anode and cathode are graphite and transition metal oxides (TMO), respectively, and the electrode reactions are exemplified by:

Anode: C6Li −−→ C6+e –_+Li+ Cathode: 2 Li0.5CoO2+e

– _+Li+_{−−→2 LiCoO} 2 Full cell: C6Li + 2 Li0.5CoO2−−→C6+2 LiCoO2

The Li-ion battery is a full cell, with a cathode and an anode. This is in contrast with half cells that are commonly used in Li-ion battery research. A half cell is one of these electrodes (the anode or cathode) that are of interest to research, with a counter electrode of Li-metal.

The benefits of Li-ion batteries, which makes them suitable to these applications, is the high

1_{It should be noted that this designation of anode and cathode which is commonly used in battery research is in}

contrast to designations used in regular electrochemistry, where the electrodes switch places depending if discharge or charge is considered.

specific energy. Current Tesla technology reaches about 0.4 MJ/kg[1]. This high specific energy originates in the low electrochemcial potential of Li+/Li couple and low weight of the lithium atom. As an example, the Li-atom in pure Li metal has a standard electrochemical potential of -3.06 vs standard hydrogen electrode[2][3], giving anodes using Li-atoms (e.g graphite in Li-ion batteries) the most negative electrode potentials. There are however downsides and safety issues to Li-ion batteries. The reactivity of lithium atoms makes it necessary to isolate it from oxygen and water, which means that if it is exposed to air the cell can spontaneously combust. Another example is that short circuiting of the cell can lead to over-heating, which can lead to combustion of the flammable electrolyte and possibly even explosion, like what happened with the Samsung Galaxy Note 7[4][5]. Other downsides to the use of Li-ion batteries that are of geopolitical concern is the fact that many of them use cobalt in the cathode material. A large procentage of the global cobalt production is based in the Democratic Republic of Congo (DRC), and its mining is the cause of countless human rights abuses[6].

1.1.1 Anodes and Electrolyte materials

The anode of the Li-ion battery is of large interest in the effort to improve battery performance. At the anode during discharge Lithium atoms are oxidized to Li-ions and then dissolved into the electrolyte. To get the most out of this process (i.e. have the most negative potential at the anode possible) Lithium metal can be used. They are however NOT used, as pure metal anodes exhibit extensive dendrite growth which can eventually lead to short circuiting of the cell.

Due to these safety hazards, other anode materials have to be used. Commonly, graphite is used, where the Lithium atoms are intercalated in the graphite’s layered structure when charged and are oxidized by giving an electron (C6Li −−→ C6+e

– _+Li+_{) to the graphite as well as} de-intercalating from the graphite into the electrolyte during discharge. The graphite anode also experience degradation, in the form of exfoliation during intercalation and dendrite formation among other processes. These degradation processes are however much milder than the degradation processes of a cell with a Li-metal anode.

The electrolytes commonly used in Li-ion batteries are non-aqueous carbonate based electrolytes. A typical example of such an electrolyte is LP40. It consist of ethylene carbonate (EC) and diethyl carbonate (DEC) in a 1:1 weight ratio, with as the salt LiPF6 dissolved in it with a concentration of 1 M. In LP40 the EC is used due to its high dielectric constant as well as its ability to form the Solid Electrolyte Interphase (SEI) [7], while DEC acts to decrease the electrolytes viscosity. These electrolyte are not stable at the normal Li-ion battery operating potential, as illustrated in figure 2 (C6Li −−→ C6+e

–_+Li+ _{has a very low potential (only around 0.2V vs Li(metal)/Li+)).}

This instability leads to reduction processes in (breakdown of) the electrolyte. One might think that electrolyte reduction will be detrimental for prolonged battery operation. The evidence has shown that while the electrolyte is initially broken down, the reduction reactions are hindered by an interphase formed between the electrode and electrolyte, the so-called Solid Electrolyte Interphase (SEI) layer.

Energy of el

ectron

Anode Electrolyte Cathode

Stability window Reduction e-Oxidati on

Figure 2: Stability window of the electrolyte. The electrons at the anode of a charged battery have a higher energy than they have in a reduced electrolyte, making the reduction thermodynamically favourable. The opposite is true for the cathode.[8]

1.1.2 Solid Electrolyte Interphase

A stable SEI, crudely illustrated in figure 3, is formed when EC-based electrolytes breaks down at low potentials, which is as mentioned one reason that EC is used. The SEI prohibits further breakdown by not conducting electrons, so that charge transfer reactions can not take place. The other important aspect of it is that it still allows for the transfer of Li-ions, keeping the battery functional [7].

Figure 3: A sketch of the SEI at the graphite electrode, as it is currently understood [9].

Being studied vigorously since the 80’s[10], the structure of the SEI is well studied ex situ. It consists of an inorganic layer close to the electrode surface, and an outer organic layer [9] as seen in figure 3, but the processes and reactions forming these layers are not well understood, and there are several different theories[7]. Some probable breakdown products of the reduction process at the anode are the organic Lithium ethylene dicarbonate, Lithium ethylene monocarbonate, Lithium ethyl carbonate, and the inorganic Lithium carbonate, Lithium oxide, Lithium peroxide, Lithium hydroxide and flouride. They could be formed as a result of the reactions presented in table 1.

Table 1: Probable breakdown products of the reactions at the anode. In the pictures showing the structures, lithium is purple, oxygen is red, carbon is black and hydrogen is white.

Lithium carbonate (Li₂CO₃)

Formed either by the reduction of car-bonates by 2 electrons[7] according to the following reaction:

2 Li + EC −−→ Li2CO3+C2H4

Or by complex multistep chemical and elec-trochemical processes initiated by electrolyte impurities, in the following ways:

2 CO2+2 Li −−→ Li2CO3+CO[11]

Where the CO2 is from contamination. Alternatively, Li-carbonate can form from this reaction: 2 H2O + e – _{−−→2 OH}– _+H 2 2 OH– _{+2 CO} 2+2 Li −−→ 2 Li2CO3+H2

Where the CO2 can come from either

contamination or the attack by another OH– on EC[12], OH -O O O O -OH + + CO₂ Lithium carbonate

Lithium Ethylene Dicarbonate (LEDC, Li2C4O6H4)

Formed as the breakdown product of EC. Thought to be the primary component of the SEI [7], according to the following reaction:

O -O O O O -O Li+ Li+ O O O 2 + 2e -Li+ + C₂H₄ Recent studies have shown that this might not be the case, and that LEDC could in fact be an intermediate product of electrolyte breakdown, breakdown[13].

Lithium Ethylene Monocarbonate (LEMC, LiC3O4H5 or Li2C3O4H4)

In 2019 Wang et al. proposed that

LEMC (LiC3O4H5) is the main product of EC breakdown, instead of LEDC. The lewis structure of LEMC according to Wang et al. [14] is: O HO O -O Li+

Following up on their study, Henshchel et al. proposed the following reaction path:

LEDC −−→ LEMC + CO2

using a slightly different LEMC, with Li on both ends. The lewsi structure of LEMC according to Henschel et al. [13] is:

O LiO

O

-O Li+

Both LEMCs will be modelled in this thesis, as it is unclear which is the correct one, and any difference in the Raman spectra could be used to resolve this question. However, for thermodynamic caclulations, the LEMC considered will be version 2, as it is the one with a clear reaction path.

LEMC version 1

LEMC version 2

Lithium Ethyl Carbonate (LEC, LiC₃O₃H₅)

Possible breakdown product of DEC (C5H10O3) according to Xu’s review on

Li-ion batteries [7].

2 DEC + 2 Li −−→ 2 LEC + C4H10

LEC

Lithium fluoride (LiF)

Component of the inorganic SEI layer, as seen in figure 3, formed in many possible ways, one of which is:

LiPF6+2 Li −−→ PF3+3 LiF[11]

LiF is Raman inactive, but it can still be discussed as a part of the system of reactions that take place at the anode. It can also serve as a clear indicator to the accuracy of simulated Raman spectra, and is therefore of interest to this thesis.

Lithium Oxide (Li2O)

Formed as a product of oxygen impurities[11] in the following way

O2+4 Li −−→ 2 Li2O

or from extensive Argon ion etching[9], and probably not part of the SEI proper. Lithium Hydroxide (LiOH)

Formed as a product of water impurities[11] by the following reaction:

2 H2O + 2 Li −−→ 2 LiOH + H2and probably not part of the SEI proper.

These last two compounds are, as mentioned, not thought to be part of the SEI, but formed from impurities. They have however been found in experiments researching the SEI [7], so for the purposes of this thesis they are still interesting.

Li2O

LiOH

The list presented in table 1 is just a fraction of the many intermediate substances and components thought to form the SEI, and as can be seen some of these reaction pathways are still hypothetical.

1.2

Aim of the thesis

The aim of this project is to develop strategies for efficient identification and classification of various active components in SEI layers, and, in turn, gain an understanding of the reactions taking place. This will help find routes to stabilize and tailor the composition of the SEI layer for long-term stability and optimal battery performance. The specific strategy used and evaluated in this thesis was that of using simulated spectra for the evaluation of spectroelectrochemical results.

1.3

Strategy

To be able to classify the active and transient components in the SEI, a quick experimental method of analysis able to see chemical bonds which can be operated in situ or operando is needed. Spec-troscopic methods, that provide information on optically active vibrations, the smallest timescale in figure 4, are often useful in such applications, of which two popular methods are Raman and IR spectroscopy. The idea of this project is to use Surface Enchanced Raman Spectroscopy (SERS) spec-troelectrochemically (in operando with an applied variable voltage) in order to study the behavior of relevant Li-ion electrolyte molecules, reduction products and SEI components (organic carbon-ates, lithium salts, etc) on gold model surfaces. These methods will be described in the next chapter. The operando spectra obtained in these experiments are often very complex, consisting of several different spectra from different compounds, and for many of these compounds there are no spectra for the pure substances. It is also the case that spectra might change due to coordinating differently in solution or when applying potential. In order to analyse it more efficiently, simulated spectra of several SEI compounds, electrolyte reduction products and compounds with similar structure will be used. These recorded and calculated spectra will be used to find correlations between each peak of the simulated compounds spectrum to one or several vibrational modes, attributing the peaks in

the SERS spectra to these vibrations, and from there try to build knowledge of what compounds are detected by the experiment.

Length scale

Electronic

structure

Macrosopic

forces

Interfacial

Chemistry

Time sc

ale

Molec

ular

Vibra

tions

Structur

al

Vibra

tions

Diffusi

on

Figure 4: Simulation methods at different scales. The smaller the systems and timescales get the more complex the calculations get, and therefor the less complex the models get.

When it comes to the simulations, different methods and entities are considered depending on the scale of the problem, as can bee seen in figure 4. At larger time and length scales, like the calculations of heat transfer according to the heat equation for example, Finite Element Method (FEM) calculations are applicable, as they require only material properties on a macroscopic scale and the geometric shape of the structure. At intermediary scales the calculation of large molecular systems over a long time, like the diffusion of atoms through a membrane, force field calculations are applicable. These calculations require no information on the actual electronic structure, but use potentials regarding the bending and twisting of molecules as well as interaction between molecules. These potentials can come from simulations on smaller scales or empirical data. At the small scale, electronic structure and vibrational modes of molecules can be calculated using quantum mechanics. As the experimental methods used in this thesis are spectroscopic, which probes the state of chemical bonds and charge fluctuations in these, the simulation methods in this thesis need to be able to simulate the electronic structure as well as the vibrations of the atoms, they need to be quantum mechanical which allows us to compute the electronic structure, as well as the atomic vibrations, of the considered molecules and materials, c.f. figure 4. In this thesis the density functional theory (DFT) is used for this purpose. The DFT data will also be utilised to narrow down the search of probable reduction products, by calculating the free en-ergy of the reactions leading to these products and find which are the most thermodynamically stable. Before applying the strategy to the spectroelectrochemical measurments, it is of vital impor-tance to validate the strategy, i.e. comparing theory and model experiments, is also necessary to investigate. Regarding the DFT simulated spectra, several pure compounds (Li-carbonate,

oxalate, peroxide, DEC, hydroxide monohydrate, LiPF6, LiBOB, methoxide and Li-acetate dihydrate) will be measured and simulated, and the differences will be noted and discussed. Measurements will be made to see if laser wavelength has an effect on the Raman spectra, using two different wavelengths.

2

Theory

2.1

Raman spectroscopy

Raman spectroscopy is a method that uses a monochrome laser to excite the electrons in a substance (molecule, crystal etc) to virtual states, which then collapse back to regular energy states, as illustrated in figure 5. Raman measures the interactions where the light either looses (Stokes scattering) or gains energy (anti-Stokes scattering), while filtering out the elastically scattered light (Rayleigh scattering)[15].

Energy

Anti-Stokes Rayleigh Stokes

Ground state Vibrational energy states Virtual energy states

Figure 5: Picture explaining the main principle of Raman scattering.

This interaction is stronger the more the polarisability of the substance is changed with the interac-tion. The polarisability of a substance is in essence how easy it is to induce an electric dipole when applying a field.

As the interactions result in a spectra of intensities at wavelengths relative to the incoming laser wavelength, the laser wavelength can be constant. This is unlike other spectroscopic methods like IR, that needs to either sweep a range of laser wavelengths using a variable laser or use Fourier transform and light with many wavelengths.

The penetration depth d of Raman can be determined according to the following equation[16]: d = (2λ

µσ)

1

2 (1)

Where λ is the wavelength of the laser, µ the permeability of the material and σ the conductivity. As can be seen, σ is in the denominator, meaning that for non conducting substances the surface penetration is high, while with conducting surfaces the penetration is extremely low.

2.1.1 Surface Enhanced Raman Spectroscopy

The surface enhanced part of SERS comes from the fact that it detects only the molecules close to the surface of a conducting nanoparticle. A field is induced in and close to the nano particle surface by the field from the incoming Raman light source, which is stronger than the incoming field is in vacuum. As the intensity of the Raman spectrum is higher the more intense the applied

field is, the intensities of the molecules near the surface, where the field is stronger, are enhanced. The enhancement factor can be as high as 1010_{, but decreases with r}−10_{, where r is the distance} from the surface [17]. This enhancement is due to something called Localised Surface Plasmon Resonance (LSPR), which occurs when the electrons on the surface of the conducting nanoparticle oscillate with the incoming field. For the LSPR to be significant in magnitude, it is important that the electrons of the metal used are easily polarised, which is why metals like gold and silver are commonly used. For this to occur, the particle must be small enough for the field not to vary significantly along the length of the particle. The particle also has to be large enough for the electrons in the particle to arrange themselves like they would in a regular metal sphere, i.e. along the surface. This balance leads to different laser wavelengths requiring different particle sizes to achieve optimal enhancement. For Au particles and a 785 nm laser, which are used in this thesis, a particle size of 46-74 nm is optimal [18].

2.1.2 Spectroelectrochemistry

Spectroelectrochemistry is the direct observation of electrochemical processes using spectroscopic methods. For example it might be continuously using IR on a cell electrode as the potential is being cycled. Spectroelectrochemical methods are very useful as they have the potential to provide direct observations of reactions, but for many applications, especially regarding Li-ion batteries, the implementation of these methods can be quite tricky. This is because the chemistry of Li-ion systems is quite sensitive to oxygen and water, and canth not be exposed to a normal atmosphere. To use SERS specifically in a spectroelectrochemical application like Radjenovic[19], Mozhzhukhina[20] and others have done, a special gold SERS substrate is used. This substrate has a nano-structured surface, which means that it can be used as a substrate in SERS. Electrolyte is applied to the substrate and then the potential is varied to emulate the environment at an electrode surface, specifically lowering the potential to emulate the anode.

2.2

Density Functional Theory

Density functional theory has become a standard tool in modern materials science. In particlar in cases where experimental techniques are to obtuse to capture the details of structure-property relationships, for example when it comes to correlating certain spectral features to specific chemical bonds. In the following, I will briefly go through the basis of DFT. Ref. [21] gives a full review of the method.

In DFT, the total energy is a unique functional of the ground state electron density. The to-tal energy of any system of electrons and nuclei is in DFT expressed as:

E[n(r)] = Ekin[n(r)] + Eext[n(r)] + EH[n(r)] + Exc[n(r)] + EII (2) This equation describe 5 different contributions to the total energy of a system. The energy from the interaction between the positive nuclei EII, the approximated kinetic energy of the electrons Ekin, the nuclei and electron interaction Eext, Hartree interaction EH, i.e. the interaction between the electrons in the system, and the exchange correlation energy Exc. The electron density n(r) is described by the sum of the square of the orbitals describing the electrons.

Given a position r and the electronic density, the first 4 terms of equation 2 can be calculated. The last term however is unknown, and needs to be approximated. There are several approximations of the Exc available, each having different applications in modelling. In this thesis, the RPBE functional will be used [22]. This functional is part of a class of functionals called Generalised Gradient Approximation (GGA), in which the gradient of the electronic density at r is taken into account, as well as the value of the electronic density[23]. In standard DFT, dispersion effects, which are very important for large molecules, as well as other molecules exhibiting large Van der Waals forces, are not accounted for. In this work, we correct for such interactions using the Grimme D3 model (RPBE-D3) [22].

DFT in practice works by calculating the electronic density according to these equations. Starting with a basis set, and a “guess” for the coefficients of this basis set, it calculates the density and changes the coefficients. This process is repeated until the program finds the electronic density with the lowest energy for this specific arrangement of the atoms, and is called the self-consistent field (SCF) approach. The interatomic forces are then calculated from the derivative of the total energy with respect to position, according to which the atoms are moved. At this new position, the SCF cycle to calculate the electronic density is performed again, followed again by the movement of the atoms according to the forces. This entire process is repeated until the change in energy is small enough. At this point, the electronic density is calculated again, and this electronic density is assumed to be close to the actual electronic density. Often the basis sets used for the initial guesses are based on atomic orbitals or plane waves. Atomic orbitals are appropriate in the case of small molecules with a definite end, and plane waves are best in the case of crystalline structures, where periodic boundary conditions might apply.

DFT is faster than other modeling techniques like wave function based methods such as Hartree Fock (HF), especially for larger systems, as it scales with the N3_{and (compared to HF, that scales} with N4_{), N being the size of the system [21].}

2.2.1 Vibrational spectroscopy simulations using DFT

DFT is a powerful tool when it comes to modelling spectra, as it can model individual vibrations and energy levels of molecules. At first, the structure is geometrically optimised according to the description above, the iterated process. The theoretical frequencies according to the harmonic approximation are then obtained by performing a linear response DFT calculation on the converged structure. This means displacing the atoms in all 3N (where N is the number of atoms, 3N arises from the number of ways a system of N atoms can be displaced) directions and calculating the electronic density of each of these displacements. To obtain which of the calculated vibrational modes are Raman active, the derivative of the polarisability of the vibrations is calculated[24].

2.2.2 Electrochemistry and thermodynamics calculation

With the energies given from the DFT calculations, it is useful and interesting to be able to calculate the change in Gibbs free energy (∆G) of the reactions given in table 1. This change is given by:

∆Gr= ∆H − T ∆S (3)

Where H is the enthalpy, T is temperature and S is entropy. ∆G is however also given by Nernst equation, when electron transfer reactions are considered:

∆Gr= −nqE (4)

Where n is the number of electrons transferred in the reaction, q is the electron fundamental charge, and E is the potential. To calculate how ∆G changes when applying a potential, the ∆G from equation 2 is added to the one in equation 1:

∆Gr= ∆H − T ∆S − nqEapplied (5)

This gives a change in ∆G depending how many electrons are transferred at potential E, i.e. how ∆Gis changed when applying potential compared to ∆G at no applied potential. In this equation the assumption that Nernst equation can be applied in this manner was made (i.e. the other way

around, with a set potential and an unknown Gibbs energy). To simplify the calculations in the scope of this study, several assumptions will be made. The first is that the enthalpy H is equal to the internal energy U, and that the contribution to H from the product of the pressure P and volume V is negligible. Another assumption is that contribution from S is only dependent on the vibrational states of the system and their occupation at 0 Kelvin, included as the energy ∆Evib. This assumption leads to the following equation:

∆Gr= ∆U + ∆Evib− nqEapplied (6)

The applied potential during spectroelectrochemical experiments is always in relation to something, usually Li/Li+ _{when it is Li-ion batteries being investigated. Due to the U and E}

vib not being calculated with this in mind, it is not possible to use the applied potential vs Li/Li+ _{as E}

applied. To calculate the applied potential it is necessary to first calculate E it for a reaction with a known standard potential, use this as a reference to calculate the Li/Li+ _{potential, and add this potential} to the applied potential vs Li/Li+_:

Eapplied= ELi/Li+_{vs ref} − E_{applied vs Li/Li}+ (7)

3

Experimental setups and procedures

In this section the experimental setup of the different parts will be discussed.

3.1

Raman measurements

The Raman experiments were performed in the Renishaw InVia 2013 Raman Spectrometer using the Wire 4.2 software.

For the experiment testing the effect of laser wavelength an airtight chamber was introduced into the glove box where the EC is stored. The EC was introduced into the chamber and the chamber was tightly sealed and then taken out of the glove box. Raman measurements were then performed using a 5x lens in the spectrometer and two different lasers, 532 and 785 nm, with a 1800 and 1200 lines/mm grating respectively.

These measurements were performed as follows: three extended scans from 100-3200 cm−1 _{with an} acquisition time of 30 seconds, and a laser power of 1% of 100 or 300 mW for the 532 and 785 nm respectively, with cosmic ray removal and background subtraction. The intensity of two spectra was then normalised by dividing every data point with the maximum value of the data.

The experimental procedure for the Raman spectra of pure compounds was similar to the EC laser experiment. The laser used in this case was 532 nm, with a 1800 lines/mm grating, cosmic ray removal, 20 second acquisition time, 5% of 100 mW laser power, and 2 acquisitions. The lens used in this case however was a long-focus 15x. The preparation of the samples were done by simply putting the samples in an aluminium cell using a spatula cleaned with ethanol. To be sure that there was no ethanol left on the spatula it was air dried for 2 minutes. The sample preparation was performed in a fumehood.

3.2

Density functional theory calculations

For the Density Functional Theory calculations, the Vienna Ab Initio Simulation Package (VASP) was used. VASP is a plane-wave pseudopotential implementation of DFT, which is suitable for simulations of molecules and solids. The plane-wave basis set were truncated using a kinetic energy cut-off of 600 eV. The Brillouin zone of our super cells were sampled in the Gamma point for

molecules and by a converged set (5x5x5) of k-points for crystals. The Monkhorst-Pack scheme were used for the k-point sampling. If available, structures of the compounds were taken from the Materials project [25]. Otherwise, the starting structure was sketched using tools provided in the python based Atomic Simulation Environment software. Geometry optimizations were performed using the preconditioned conjugated gradient algorithm and the systems were considered converged when the forces on each atom were less than 0.002 eV/Å. For crystals the atomic position and the unit cell was optimised, for molecules only the atomic position were optimised.

With a geometrically optimised structure a finite displacement calculation is performed to obtain the vibrational modes. The inputs for the INCAR files for the geometric optimisation and the finite displacement can be found in the appendix. Lastly, the sc-Raman code [24] is used to calculate the polarisability of the vibrational modes from the Gamma point phonons and the macroscopic deilectric tensor, thus obtaining the Raman spectrum.

3.2.1 Thermodynamic system

All molecules were geometry optimized and the total energy for all species were collected to evaluate their relative thermodynamic stability. The DFT calculations for the thermodynamic system is performed just like the calculations for the Raman spectra in the previous chapter, but they are stopped after the finite displacement, as only the energies are needed. The equation system from section 1.5 is the used, in which U is obtained as the energy of the geometrically optimised system, and Evibas the zero point vibrational energy divided by 2. They are first used to find the potential of a reference reaction, which is compared to tabulated data. The reaction used as reference was the electrolysis of water:

2 H2O −−→ 2 H2+O2

With a standard cell potential of -1.23 V. After this calculation, ∆G of the following reactions from section 1.1.2 were calculated:

2 Li + EC −−→ Li2CO3+C2H4 2 CO2+2 Li −−→ Li2CO3+CO 2 Li + 2 EC −−→ LEDC + C2H4 LEDC −−→ LEMC + CO2 2 DEC + 2 Li −−→ 2 LEC + C4H10 LiPF6+2 Li −−→ PF3+3 LiF O2+4 Li −−→ 2 Li2O 2 H2O + 2 Li −−→ 2 LiOH + H2

The reason that these are the reactions modelled was that they were easily available as clearly written reaction formulae, and thus possible to model in the scope of this project.

3.3

SERS

All SERS measurements were performed in a custom SERS operando cell, as represented in figure 6 below.

Figure 6: Cross section SERS operando cell. 1. Teflon O-ring, 2. Metal mesh, 3. SERS substrate with gold side pointing up, 4. Separators and electrolyte, 5. Lithium, 6. Plastic cylinder, 7. polymer seal. Figure

from Mozhzhukhina paper[20], supplementary info

The part of the cell in contact via the metal mesh to the gold substrate is then connected to the working electrode cable of a potentiostat equipment, and the Li part is connected to the reference cable of the potentiostat equipment.

The cell is assembled in a glovebox with an inert atmosphere. The all of the cell components, as well as the tools, are dried before entering the glovebox. Care was taken never to touch the components without an extra pair of gloves inside the glovebox, and tweezers that had been cleaned inside the glovebox using methanol.

The cell is assembled by first assembling the top metal part, the glass and the metal mesh. 200 µl of electrolyte is put into the cell. 50 µl before the SERS substrate is put in, and then 50 µl before each separator, which number 3 in total. The Li metal is cut out using a punch and then inserted into the cell very carefully, as to avoid short circuit the cell. The cell is then sealed, put into a glove and taken out of the glovebox for the analysis.

The electrolyte used in these tests was a 1M solution of LiPF6 in EC, prepared on a hot plate in a glove box. Two tests were performed. In the first the working electrode was set to 3.0 V, stepped down by 0.1 V to 0.5 V, with a step time of 5 minutes. It was then kept at 0.5 V for 50 minutes and then the program ended. In the second, the steps were of 0.025 V instead, resulting in a test taking roughly 4 times longer to complete. Raman measurements was started immediately after 3.0 V had been applied. The Raman measurements were then collected every 5 minutes for enough time to cover all the voltage steps, as well as taking some measurements after all the steps had completed. The settings for the Raman measurements were as follow:

785 laser, 1200 lines/mm grating, 50x lens, 30s acquisition time, 1% of 300 mW laser power, and cosmic ray removal.

4

Infrastructure for SERS analysis

Conventional analysis of spectra includes identifying molecular vibrations using reference spectra. This requires a large amount of experience on the part of the analyser, and also much data from several different experiments.

The advantages of this analysis are that the spectra are exact, i.e. if the spectrum of a compound can be seen clearly in the spectroelectrochemical results then the presence of this compound is very likely. The disadvantages of this analysis are however that if there are no clear spectra of reference compounds present, the analysis relies on, as previously mentioned, identifying the vibrations of peaks seen in the reference spectra, and using this knowledge to identify the peaks of the spectroelectrochemical data. From there a picture can be built of the compounds present in the experiment. The uncertainty of the correct vibrations were identified correctly identified is always present, and there is also always the chance that the frequency of vibrations might shift due to chemical environment and structure. There is also the problem of building a database of reference spectra. It is of course good to have reference spectra of compounds as closely resembling the compounds expected to be found in the experiment. As there are not always spectra available, these compounds have to be synthesised and tested using several different methods, first to validate the synthesis and then to obtain the reference spectrum.

4.1

Analysis using modelled data

In this thesis, as mentioned in the introduction, simulated reference spectra from DFT calculations will be used to replace experimental spectra. The advantage of this is twofold. First, the vibrations are directly calculated by the program, so the only part left of the identification of the reference spectrum is labeling the vibrations. Second, the need to synthesise and validate the synthesis is completely removed. This means that the modelled spectra can be used without as much experience or work. The spectrum is simply simulated, labeled, and then the analysis continues as regular analysis of spectral data. The disadvantages of this kind of analysis is that the simulated spectra can be shifted, and that the intensities can be incorrect.

Shown in figure 7 and table 2 are the vibrations of the DEC molecule from the simulation, labeled by going through the vibrations manually, compared to the experimental Raman spectrum of liquid DEC.

Figure 7: DEC experimental and modelled spectra. DEC data supplied by Nataliia Mozhzhukhina. The boxes presents groups of vibrations present in the molecule, to illustrate how the groups differ between the simulated and experimental spectrum.

Table 2: DEC vibrational modes in figure 7

1 338_{375 C-O-C’s and O-C-C’s asymmetric scissoring bends}C-O-C’s and O-C-C’s symmetric scissoring bends

2 489 O-C-C and O-C-C’s asymmetric scissoring bends

3 785 H2 and H3 asymmetric rocking delta

4 847 H3 asymmetric wagging delta and O-C-C asymmetric stretch

5 877 H3 symmetric wagging delta and C-O-C symmetric stretch

6 ₁₀₁₇986 O-C-O symmetric stretch and C-C’s symmetric stretch_{C-C-O’s asymmetric stretch} 7 10971108 C-C’s stretch, C-O-C scissoring bends and H3 wagging deltaC-O-C’s asymmetric stretch and H3 wagging delta

1134 H2 and H3 rocking delta and twisting delta and O-C-C wagging delta 8

1252 H2 twisting delta and H3 rocking delta and end C’s rocking

1341 H2 and H3 wagging with each other (H2 most active) and O-C-C asymmetric stretch 1369 H2 and H3 wagging against each other (H2 most active) and O-C-C asymmetric stretch 1380 H2 and H3 wagging against each other, C-C stretch and O-C-O asymmetric stretch

9 1436 H3 asymmetric twisting delta

1449 H2 and H3 scissoring delta with each other

1468 H2 and H3 scissoring delta against each other

10 1733 Carbonate asymmetric stretch (O-C-O symmetric stretch against top C-O stretch) A similar analysis has been performed in the spectra for LEMC, LEDC, LEC, oxalate, Li-carbonate and Li-peroxide. These vibrations have then been plotted in a single graph (figure 8), with the goal of being able to gain knowledge of the types of vibrations, if there are any trends pertaining to the vibrations and their atomic environment and so on. The differences between the simulated and experimental spectra will be discussed in section 4.3.1.

Figure 8: Raman active vibrations sorted by type

As can be seen in figure 8 there are some interesting trends. The symmetric stretch of the carbonate is centered around 1000 cm−1_{. The C-O-C scissoring bend is concentrated at lower wavenumbers} and C-C-O asymmetric stretch is located roughly at 1000-1050 cm−1_{. As expected from the light}

H atom many of the vibrations involving it center around higher wave numbers, and the H-C-H scissoring bend is only seen in a very narrow range at about 1450 cm−1_{. Some of the asymmetric} vibrations of carbonate is centered around 1550 cm−1 _{in molecules where it is coordinated with Li.} As can be seen in the following figure it is not as much that the vibrations are shifted, but that there are new modes added due to the coordination with Li.

Figure 9: Asymmetric and symmetric stretching of the carbonate group

This data has to be confirmed by the simulation of more molecules like DEC, with a carbonate group without a lithium, as well as the validation of the simulated spectra of molecules like LEDC, LEC, LEMC and so on by experimental Raman. It might well be the case that crystalline arrangements of these molecules behave more like Li-carbonate, with peaks around

4.2

Thermodynamic calculations

Another way to aid the analysis of the spectroelectrochemical data is to calculate the spontaneity of the proposed reactions as a function of the applied potential, the result of which is plotted in figure 10. This is done as described in the experimental setups and procedures section, and will be demonstrated here, using the formation of Li2O as an example:

The energy U + Evib of the reactants is subtracted from U − Evib of the products. For the reaction 1

2O2+2Li −−→ Li2O, this means:

(ULi2O+ Evib,Li2O) − 1/2(U2O+ Evib,Li) − 2(ULi+ Evib,Li)

This is then added to −nqEapplied. As q = 1 (eV units), and the number of electrons trans-ferred in this reaction is 2 (n = 2), this gives the total equation:

∆G = −nqEapplied+ (ULi2O+ Evib,Li2O) − 1/2(U2O+ Evib,Li) − 2(ULi+ Evib,Li)

Figure 10: Calculated Gibbs free energy and its dependence on potential using the equations from section 1.2 and the reactions from section 1.1.2. The reactions are notated by the main product they produce in regards to the SEI.

As can be seen in figure 10, the reactions leading to lithium carbonate, Li2O and lithium hydroxide all have quite low ∆G, compared to the reactions leading to the organic molecules like LEDC, LEMC and LEC. This is in keeping with the general principle that crystalline substances are more stable than the organic molecules. The only outlier is the reaction leading to LiF, which above 1.25 V vs Li/Li+_{has a higher ∆G than the organic molecules. The comparison between the two reaction} paths to Li2CO3 is particularly interesiting, as if the 2 electron reduction path to Li2CO3 was the most stable, which is unexpected, as when impurities are present lithium carbonate is observed to be formed from these. This highlights the importance of having a complete reaction path, and including all of the reaction paths. The specific comparison of LEDC and LEMC highlights a large issue with the assumptions made in the calculations. The energies rise at the same rate, so according to the calculation LEDC will be more favourable than LEMC at all potentials, and suggests according to these simulations that it should not be found in the SEI.

The problem here is that one of the products of the reaction leading to LEMC is carbon dioxide, a gas. As a gas has a very high entropy compared to liquids or solids, and this transitional entropy is completely ignored by the assumption that the entropy contribution is completely vibrational and at 0 Kelvin. It is also assumed that there are no molecule-molecule interactions (which are present in gasses), due to the way the molecules were modelled. With the added entropy from the gas transition the energy of LEMC migh very well be lower than that of LEDC.

Another reason for the discrepancy could be that the wrong LEMC has been used, and that it is in fact the first version proposed in 2019 by Wang et al. that is correct, in the sense that the reactions leading to Wang’s LEMC have a lower ∆G than the reaction leading to Henschel’s LEMC, which was the one calculated in this thesis.

It should be noted that kinetic limitations are not considered in these calculations and might change the conclusions drawn from the results of the calculation.

4.3

Validation

To be able to use the data obtained in figure 8 the results and methods used to obtain it needs to be validated. This was done by modelling relevant spectra and comparing these with each other and experimental data. Relevant spectra are spectra of molecules with similar bond structure to the molecules of interest in the SEI, that were available in the laboratory.

Comparing the two spectra of the LEMC molecules (figure 11), to ascertain internal consistency, the two different LEMC molecules give similar spectra, except when it comes to the vibrations involving hydrogen, as can be seen in figure11 in the box labeled H-C-H vibrations. This is likely due to the way the atom bonded to the oxygen (i.e. hydrogen or lithium) interacts with the rest of the atoms. In the first version of LEMC, with the hydrogen on the end, the C-O-H shows a scissoring mode as its primary vibration, while in the other LEMC, with the lithium on the end, the C-O-Li group shows a stretching mode as its primary way of vibrating. This last mode is seen clearly in figure 11 at about 850 cm−1_.

Another interesting point to make is that it shares two distinct modes related to the carbon-ate with LEDC. These modes are the Li stretching symmetrically with its two closest oxygen neighbors, and then the symmetric stretch of the carbonate group. These modes seem very stable in regards to the rest of the molecule, and could be used as identification of carbonate groups coordinated with lithium.

Figure 11: Comparison between the two different LEMC molecules simulated in this project, as well as LEDC, as further comparison. The boxes presents groups of vibrations present in the molecules, to illustrate how the groups differ between the different molecules.

A source of error for this is the effect of the laser wavelength on the intensity spectra, by obtaining spectra from EC using two different lasers, as described in section 3.1. As shown in figure 12, the laser used can have a significant effect on the intensity of the peaks. With higher laser wavelengths the peaks at higher wave numbers seem to get much lower intensities.

Figure 12: Measurement of the same EC sample with different laser wavelengths, showing the effect that wavelength can have on the Raman spectrum.

This effect could possibly be explained by the heating of the EC due to the laser. If this effect is present in all compounds and if there are other effects cannot be deduced from this brief comparison.

4.3.1 Comparing real and simulated data

As seen in figure 7 and table 2, where the simulated spectrum of molecular DEC and the experi-mental spectrum of liquid DEC are presented, the simulated Raman spectrum matches well with the experimental spectrum, except for the peaks nr 7 and 9. The extreme heights of the 7 and 9 peaks could be attributed to many factors. For the 7 peak there are in the simulation 3 separate peaks, which are broadened by putting the appropriate peak shape at each peak position. As the peaks have shifted and are closer together in the simulated spectrum, this will make the peak look much higher, as it is an addition of three separate peaks. For 9 the explanation might be that these vibrations are mostly from hydrogen, which will be discussed at the end of this section.

Another compound that was simulated and compared to experimental data was lithium car-bonate, as seen in figure 13. In this, the match is expected to be better than for DEC, as both the simulated and experimental compounds are crystalline.

The lithium carbonate spectra matches very well, and the only discrepancies are the low peaks, at 100-300 cm−1_{, which do not show up in the simulation. This could very well be due to the} way the simulations are performed. Phonons with wavelengths that are larger than the unit cell does not show up in the simulated spectrum, since periodic boundary conditions are used. Vibrations that are larger than the unit cell would tend to be slower, i.e. have a lower wave number. The crystalline compounds are also assumed be bulk crystals, which disregard break-ing of symmetry at the surface, which is also a source of error. Overtones, combination modes and differential modes are other possible sources of error, as they are not covered by the calculations. A third compound used to validate the simulations in this way was lithium oxalate, and as seen in figure 14, the lithium oxalate spectra matches very well except for some shifts in frequency

Figure 14: Lithium oxalate experimental and modelled spectra.

There is however a huge caveat: the first time modelling this substance the match was inaccurate and in stark contrast with the experimental data, as shown in figure 15. This was because an incorrect starting position was assumed, and a false minimum which did not match the structure of lithium oxalate.

Figure 15: Lithium oxalate when optimised to a false (incorrect) local minimum vs when optimised to the correct.

The only difference in the structure however was the way the molecules were arranged, in the incorrect one the two oxalate molecules in the unit cell were parallel to one another. In the correct structure the molecules were perpendicular. This highlights an important factor when modelling these substances: that the structure of whatever is being modelled is incredibly important. This will be discussed at the end of this chapter.

In all these comparisons there is a tendency for the vibrations involving C-O bonds to shift down about 20-50 cm−1_{. There are however also instances where there is no shift, as seen with} DEC in figure 7 regarding the peaks labeled 6, which are not shifted. This makes it impossible to just set a general offset to all the wavenumbers of the simulated spectra, and it makes the analysis using the spectra less reliable.

From these comparisons it can be concluded that while it might not be as important to have several molecules interacting when modelling systems like DEC, where it is liquid and the single molecule modelled almost without interaction with its neighbors, it is very important when modelling ionic species like the products of electrolyte breakdown, as these will more likely arrange themselves in some kind of crystalline or semicrystalline structure. The crystalline structure hugely affects the resulting Raman spectra, which makes SERS an extremely versatile tool, but it makes the modelling challenge much harder. You either have to model many structures of the same substance or provide prior knowledge about the structure of your experiments. Amorphous structures or molecular structures like liquids were disregarded in the calculations. These structures will have different spectra to the crystalline structures, but cannot be modelled using a simple unit cell, further emphasising the issues with using simulated spectra.

Another effect that can be seen is that the peaks at about 3000 cm−1_{, originating from O-h} and C-H vibrations, always seem to be much higher than their experimental counterparts. This might to some degree be explained by the effect observed in figure 12, that the laser might have an effect on the higher wavelengths. More likely is however that the hydrogen atoms that these vibrations originate from are light, and therefore more sensitive to weaker interactions. This will affect the result twofold:

• First, if the molecule is assumed to have no molecules around it and interacting with it (reflected in modelling with a large unit cell in these cases) the weak interactions that might

affect the intensities of these vibrations are basically assumed to be close to zero.

• Second, DFT does not describe some of these interactions completely, to be specific the exchange-correlation forces. These effects are dependent on the kind of functional used, and could vary if using another functional. It is not however not that big of a problem when modelling these systems, as many of the interesting peaks appear in the range of 500-2000 cm−1_.

5

Operando Surface Enhanced Raman Spectroscopy results

In this section the results from the SERS experiments as described in section 3.3 are presented and discussed. In figure 16 the voltage was lowered by 0.1 V every 5 minutes, and a spectrum was obtained for each step.

Figure 16: Results from the normal paced run as described in section 3.3.

From this result (figure 16), apart from the LP40 spectrum the formation of lithium carbonate can clearly be seen at 1100 cm−1 _{when reaching 1.5 V, by just comparing the results with the lithium} carbonate spectrum. There is also the disappearance of lithium carboxylates at 1600 cm−1_{, at 2} V. Below 2 V almost everything seems to have been reduced to lithium carbonate, and the only other thing present is the EC signal from bulk, indicating that the reactions are initiated by the reduction of water, as found in the work of Mozhzhukhina et al. [20]. The other peaks, before 2.0 V, are probably noise, since they only show up an one measurement at one voltage step.

In figure 17 the voltage was lowered at a slower pace, 0.25 V every 5 minutes. The reasoning behind this was to obtain a higher resolution.

Figure 17: Results from the slow paced run as described in section 3.3.

due to the substrate sliding out of focus, as it did in the other case as well, only that the other case took much shorter time to complete so the effect isn’t as noticeable.

In this experiment much higher resolution was indeed achieved in the beginning of the exper-iment. Like in the previous experiment Li-oxalates above 1600 were detected. However, the rest of the appearing peaks above 2.0 V has not been attributed to any specific simulated spectrum, and it should be noted that no reduction of the electrolyte should be taking place at this point, but it is clear that these lines are not simply noise. It could be that molecules of the electrolyte are coordinating to the surface at these potentials, causing the new peaks. Below 2 V the formation of lithium carbonate is apparent like in the previous experiment, but also the formation of something else, with peaks at about 850, 1300, 1375, 1515 and 1560 cm−1_{, as seen in figure 18.}

Figure 18: Spectrum at 1.9 V vs Li/Li+, with the interesting peaks at 850, 1300, 1375, 1515 and 1560 cm−1 highlighted

The peaks at 1515 and 1560 cm−1 _{could be attributed to carbonate asymmetric stretching, as seen} in systems where the carbonate is coordinated with lithium, as seen in figure 8, possibly indicating decomposition of EC into something different than Lithium carbonate, which has its peak at about 1460 cm−1_{, as seen by its formation in figure 16. The peaks at 1300 and 1375 cm}−1_{are possibly from} either H-C-H wagging or scissoring or from carbonate asymmetric stretching again. The peak at 875 cm−1 _{could be attributed to carbonate symmetric stretching, if a shift in the simulated spectra of} about -50 cm−1_{is assumed. It could also be attributed to a H-C-H2 wagging motion, but this seems} unlikely since it would imply the creation of a CH3 group, which is not often seen in reactions re-garding the breakdown of EC, and only EC was used, meaning that the CH3group is not from DEC. In both cases, the peak attributed to PF –

6 in the work of Mozhzhukhina et al. [20] has its highest intensity between 2.7 and 2.5 V, indicating a higher concentration of this ion near the surface at this point.

From these two experiments it is quite clear that most spectral features disappear below 2 V, seemingly being reduced to lithium carbonate by the breakdown of water impurities. As this inorganic carbonate layer forms and grows, it is quite possible that it grows thick enough for the SERS effect to disappear at its surface. This could explain why there are no organic species seen in the experiments, as the organic layer would be outside the range of the SERS effect. Another reason that there are no indications of an organic layer is that the structure of the simulated organic compounds does not correspond well with the structure of the compounds present in the

experiments, as some compounds assumed to be molecular in the simulations might actually be crystalline. It is possible that some of the spectra in the second SERS experiment could belong to an organic species. Potentiostat data from both SERS experiments are available in the appendix.

6

Conclusions, general discussion and future work

The goal of this thesis was to use and evaluate the strategy of using simulated spectra to evaluate spectroelectrochemical results, using DFT and SERS.

From the SERS spectra it can be concluded that Li-carbonate and other reduction products with carbonates appear below 2 V, probably due to reactions involving water contaminants. To improve upon this data, more carefully designed SERS experiments are required. To be able to confirm the presence of the organic compounds like LEDC, very pure electrolytes must be used, and contamination must be avoided in all ways possible. Other salts resulting in other chemistries could also be used to minimise the interaction with impurities. But even if a pure enough electrolyte is achieved, it is also possible that the inorganic layer will too thick, and that the SERS effect does not extend far enough out into the SEI. However, the reactions to create the inorganic layer without contaminants are also interesting, so future work could concentrate on a purer electrolyte and still get relevant results, it is in no way something to be abandoned.

When it comes to the modelling, it can be concluded that much more care needs to be put into starting with a correct structure of the modelled compounds, as some of the vibrations making up the Raman spectrum varies very much depending on the structure and the environment of the molecule. As an example, the effect regarding the very high intensities of the hydrogen vibrations at 3000 cm−1 _{could come from there being no surrounding molecules to dampen the vibrations,} which has not been added due to the limited time-span of this project. Another example of this is the Li-oxalate simulation, which hit a minimum which did not correspond to the structure of the real compound, due to its starting structure. More modelled spectra of compounds in different chemical environments could help with this (and help with the assignment of likely combinational modes), and a systematic approach to modelling all of the possible structures present in the SEI, in many different environments, is not out of reach when taking into account the ever increasing computational power available. Another aspect of this problem when it comes to modelling specifi-cally ionic crystalline structures is that it would be convenient if more structures were available as structure files or at least as coordinates for the whole cell. As of now, many of the crystal structures consisting of molecules are available as molecules arranged according to space groups, and while this is a good tool to keep the data compact and also helps visualising the compounds as molecules in a certain arrangement, interpreting the more complicated of these space groups can be a challenge without the appropriate knowledge. Initiatives like Materials Project [25] have started providing full 3D structures for several compounds with smaller unit cells, which aids greatly in the modelling of these compounds.

Regarding the modelling again, more care can to be put into the choice of functional and other calculation parameters, to explore what parameters are best for the system. This is as no time was put into this, and that the effect regarding the hydrogen vibrations could possibly also come from the choice of functional and calculation parameters. This effect however could also come from the choice of laser, as the choice of laser affected the intensity of these vibrations in EC very much. The effect of the laser wavelength is another area that can be explored. It should be possible to calculate the effect it will have, and a more thorough investigation can also be performed, using several compounds and many different lasers. This could also be extended specifically to SERS, as it would be interesting to see if this effect is present at all in SERS, and if it behaves differently due to the way the LSPR is dependent on the laser wavelength.

Also needed in future work is methods for labeling the vibrations of the calculated spectra. The benefit of this is twofold. First, it is labour intensive to label all of the vibrations one by one, and

all though off course not as labour intensive as labeling vibrations without using simulated spectra, it would still be of great use if it were to be automated. Second, the labels put on the vibrations are very dependent on the person labeling them. Experience, biases and other factors come into play when labeling, which will cause biases and inconsistencies in the conclusions drawn about the spectra. One interesting approach would be to teach an AI to handle the labeling of the vibrations. The training material would have to be very well produced, as to avoid the biases talked about earlier, but if implemented it could revolutionise the way spectra are interpreted. Starting with letting the AI just identify the vibrations and plotting them like in figure 8, but eventually moving on to letting it identify vibrations from the spectroelectrochemical spectra. The vibra-tions could then be matched to a large database of molecules and possible candidates could be found. However, the by far largest conclusion to be drawn from this work is that even if modelled Raman spectra and SERS data are very powerful tools, they are not enough to draw any conclusions about the reactions happening on the gold substrate. This speaks very much to the complexity of the SEI, and makes clear why it is still so poorly understood. Suggestions for the future would be using methods which can observe gas output from the reactions to narrow down the reactions taking place, and thus narrow the substances to be simulated. Other spectral techniques could also be used in tandem with SERS. As the IR active vibrations are also given from the DFT calculations that give the Raman vibrations, IR and RAMAN could be used together, as it would give an even more distinct fingerprint for all the molecules involved. I also think that it is here the reaction calculations of the thermodynamics would be more useful. If new and unexpected information is gained using these spectroelectrochemical methods, this information could be confirmed further using these kinds of thermodynamic calculations. As applied in this thesis they were not of much help (partly as they were performed after many compounds had been modelled), other than to show that more care needs to be put in to the calculation of the entropy. Had they been performed earlier, it would have been a possibility to model these reactions in more detail, and possibly also explore the kinetics of the reactions to get a better understanding of the intermediate products. These thermodynamic calculations also need to be validated theoretically, and also rigorously tested against known reactions, to see if they are correct.

7

Acknowledgements

At first I want to thank my supervisors Nataliia and Peter for allowing me to perform this project, helping me, teaching me and giving me feedback on the thesis, despite the trying times. I would like to thank Henrik and all the others at the lab for their extensive help with the lab work, and Ageo and Jolla for always helping me out with modelling related problems. I would also like to thank both the battery group and the modelling group for being open and making me feel welcome.