Nonaqueous Liquid Electrolytes for Lithium-Based Rechargeable Batteries
Kang Xu
Electrochemistry Branch, Sensor and Electron Devices Directorate, U.S. Army Research Laboratory, Adelphi, Maryland 20783-1197
Received November 3, 2003
Contents
1. Introduction and Scope 4303
1.1. Fundamentals of Battery Electrolytes 4303
1.2. The Attraction of “Lithium” and Its Challenge 4304
1.3. From “Lithium” to “Lithium Ion” 4305
1.4. Scope of This Review 4306
2. Electrolyte Components: History and State of the
Art
4306
2.1. Solvents 4307
2.1.1. Propylene Carbonate (PC) 4308
2.1.2. Ethers 4308
2.1.3. Ethylene Carbonate (EC) 4309
2.1.4. Linear Dialkyl Carbonates 4310
2.2. Lithium Salts 4310
2.2.1. Lithium Perchlorate (LiClO4) 4311
2.2.2. Lithium Hexafluoroarsenate (LiAsF6) 4312
2.2.3. Lithium Tetrafluoroborate (LiBF4) 4312
2.2.4. Lithium Trifluoromethanesulfonate (LiTf) 4312
2.2.5. Lithium Bis(trifluoromethanesulfonyl)imide
(LiIm) and Its Derivatives
4313
2.2.6. Lithium Hexafluorophosphate (LiPF6) 4314
2.3. Brief Summary 4315
3. Liquid Range of Electrolyte Solutions 4315
4. Ion Transport Properties 4318
5. Electrochemical Stability: on Inert Electrodes 4322
5.1. Anion of Lithium Salts 4323
5.2. Solvents 4325
6. Electrochemical Stability: on Active Electrodes 4326
6.1. Passivation on Lithium Anode 4326
6.2. Electrolyte/Carbonaceous Anode Interface:
SEI
4329
6.2.1. Exfoliation and Irreversible Capacities on
a Carbonaceous Anode
4329
6.2.2. Mechanism of SEI Formation 4331
6.2.3. Characterization of Surface Chemistry 4337
6.3. Electrolyte/Cathode Interface 4341
6.3.1. Passivation Film on a Cathode 4342
6.3.2. Characterization of Surface Chemistry 4344
6.3.3. Breakdown of Surface Layer 4346
6.3.4. Passivation of Current Collector 4347
6.4. A Few Words on Surface Characterizations 4351
7. Chemical and Thermal Stability/Safety of
Electrolytes
4352
7.1. Long-Term Stability of Electrolytes at
Elevated Temperatures
4352
7.2. Stability of the SEI or Surface Layer at
Elevated Temperatures
4354
7.3. Thermal Safety of Electrolytes against Abuse 4357
8. Novel Electrolyte Systems 4362
8.1. Problems Facing State-of-the-Art Electrolytes 4362
8.2. Functional Electrolytes: Additives 4363
8.2.1. Bulk Electrolyte: Ion Transport 4364
8.2.2. Anode: SEI Modification 4366
8.2.3. Cathode: Overcharge Protection 4372
8.3. New Electrolyte Components 4378
8.3.1. Nonaqueous Solvents 4378
8.3.2. Lithium Salts 4383
8.4. Novel Electrolytes with a Wide Temperature
Range
4390
8.4.1. Low-Temperature Performance 4390
8.4.2. High-Temperature Performance 4399
8.5. Electrolytes of Low Flammability 4400
8.6. Polymer and Polymer Gel Electrolytes 4405
8.6.1. Solid Polymer Electrolyte 4406
8.6.2. Gel Polymer Electrolyte 4408
9. Concluding Remarks 4410
10. Acknowledgments 4410
11. References and Notes 4411
1. Introduction and Scope
1.1. Fundamentals of Battery Electrolytes
Electrolytes are ubiquitous and indispensable in all
electrochemical devices, and their basic function is
independent of the much diversified chemistries and
applications of these devices. In this sense, the role
of electrolytes in electrolytic cells, capacitors, fuel
cells, or batteries would remain the same: to serve
as the medium for the transfer of charges, which are
in the form of ions, between a pair of electrodes. The
vast majority of the electrolytes are electrolytic
solution-types that consist of salts (also called “elec-
trolyte solutes”) dissolved in solvents, either water
(aqueous) or organic molecules (nonaqueous), and are
in a liquid state in the service-temperature range.
[Although “nonaqueous” has been used overwhelm-
ingly in the literature, “aprotic” would be a more
precise term. Either anhydrous ammonia or ethanol
qualifies as a “nonaqueous solvent” but is unstable
with lithium because of the active protons. Neverthe-
less, this review will conform to the convention and
use “nonaqueous” in place of “aprotic”.]
Because of its physical location in the electrochemi-
cal devices, that is, being sandwiched between posi-
tive and negative electrodes, the electrolyte is in close
interaction with both electrodes; therefore, when new
electrode materials come into use, the need for
4303Chem. Rev. 2004, 104, 4303−4417
10.1021/cr030203g CCC: $48.50 © 2004 American Chemical Society
Published on Web 09/16/2004

compatible electrolytes usually arises. The interfaces
between the electrolyte and the two electrodes often
dictate the performance of the devices. In fact, these
electrified interfaces have been the focus of interest
since the dawn of modern electrochemistry1 and
remain so in the contemporary lithium-based re-
chargeable battery technologies.
In a battery,2 the chemical nature of positive and
negative electrodes (also called cathode and anode,
respectively, although by a more strict definition, this
convention is only correct during discharge) decides
the energy output, while the electrolyte, in most
situations, defines how fast the energy could be
released by controlling the rate of mass flow within
the battery. Conceptually, the electrolyte should
undergo no net chemical changes during the opera-
tion of the battery, and all Faradaic processes are
expected to occur within the electrodes. Therefore,
in an oversimplified expression, an electrolyte could
be viewed as the inert component in the battery, and
it must demonstrate stability against both cathode
and anode surfaces. This electrochemical stability of
the electrolyte, which in actual devices is usually
realized in a kinetic (passivation) rather than ther-
modynamic manner, is of especial importance to
rechargeable battery systems, but it is often chal-
lenged by the strong oxidizing and reducing nature
of the cathode and the anode, respectively. The
severity of this challenge is ever increasing with the
pursuit of new battery systems with higher energy
densities, which drives the exploration of a more
oxidizing cathode and a more reducing anode as
candidate electrode materials, and thus constantly
requests improvements in electrolyte stability. Ad hoc
surface chemistry is often necessary for the kinetic
stability of these new electrolyte/electrode interfaces.
While the potencies of electrode materials are usually
quantified by the redox potential in volts against
some certain reference potential,3 the stability of an
electrolyte can also be quantified by the range in volts
between its oxidative and reductive decomposition
limits, which is known as the “electrochemical win-
dow”. Obviously, the redox potential of both electrode
materials must fall within this electrochemical win-
dow to enable a rechargeable battery operation.
Certainly, electrochemical stability is only one of
the requirements that an electrolyte should meet. A
generalized list of these minimal requirements should
include the following: (1) It should be a good ionic
conductor and electronic insulator, so that ion trans-
port can be facile and self-discharge can be kept to a
minimum. (2) It should have a wide electrochemical
window, so that electrolyte degradation would not
occur within the range of the working potentials of
both the cathode and the anode. (3) It should also be
inert to other cell components such as cell separators,
electrode substrates, and cell packaging materials.
(4) It should be robust against various abuses, such
as electrical, mechanical, or thermal ones. (5) Its
components should be environmentally friendly.
1.2. The Attraction of “Lithium” and Its Challenge
Lithium has long received much attention as a
promising anode material. The interest in this alkali
metal has arisen from the combination of its two
unique properties: (1) it is the most electronegative
metal ( -3.0 V vs SHE), and (2) it is the lightest
metal (0.534 g cm-3).3 The former confers upon it a
negative potential that translates into high cell
voltage when matched with certain cathodes, and the
latter makes it an anode of high specific capacity
(3.86 A h g-1). In the 1950s lithium metal was found
to be stable in a number of nonaqueous solvents
despite its reactivity,4 and this stabilization was
attributed to the formation of a passivation film on
the lithium surface, which prevents it from having a
sustained reaction with electrolytes. Intensified re-
search activities resulted in the commercialization
of a series of lithium-based primary cells in the 1960s
and 1970s, and the electrolyte solvents ranged from
organic (propylene carbonate) to inorganic (thionyl
chloride and sulfur dioxide).5-8
The continued efforts to expand lithium chemistry
into rechargeable technology, however, encountered
severe difficulties in terms of the cycle life and
safety.9,10 Soon it was realized that the source of the
problems was the morphology of the lithium crystals
newly deposited from the electrolytes upon re-
charge.11,12 Needlelike lithium crystals (called “den-
drite”) grow on the anode upon charge and, during
the subsequent discharge, become electrically isolated
from the substrate due to nonuniform dissolution
rates at different sites of the dendrite. The direct
victim of such lithium loss is energy density, because
excessive lithium has to be used in the cell to make
up for the loss.13 But more seriously, a hazard could
Kang Xu was born in Chengdu, China, and received his B.S. degree in
Chemistry from Southwest Normal University in Chongqing, China, in 1985
and M.S. in Polymer Chemistry from Lanzhou Institute of Chemical Physics,
Academy of Sciences, in 1988. After working on polymer electrolyte
materials from 1988 to 1992 at Chengdu Institute of Organic Chemistry,
Academy of Sciences, he went to Arizona State University and received
a Ph.D. degree in Chemistry in 1996 under the tutelage of C. Austen
Angell. From 1997 to 2002, he was awarded the National Research Council
Research Associate Fellowship and the American Society for Engineer
Education Postdoctoral Fellowship, respectively, and he served during
the tenures as a guest researcher at U.S. Army Research Laboratory
with T. Richard Jow as academic advisor. He was employed by the U.S.
Army Research Laboratory in 2002. His research interests concern
materials development for electrochemical energy storage applications,
which include lithium or lithium ion batteries and electrochemical capacitors.
He won R&D Achievement Awards from the Department of the Army in
1999, 2001, and 2002 for his work on electrolyte materials. He authored
over 60 research publications and 11 patents and is a member of the
Electrochemical Society.
4304 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

be caused by such “dead lithium” crystals, which are
electrochemically inactive but chemically hyper-reac-
tive due to their high surface area. When dendrite
growth pierces the separator and results in an
internal short, thermal runaway and explosion ensue.
The work on rechargeable lithium batteries from
the 1970s to 1980s overwhelmingly concentrated on
the electrolyte formulation, in the hope that proper
choices of electrolyte solvents or salts would suppress
or even eliminate the dendritic deposition of lithium.
Among the numerous electrolyte systems investi-
gated, an ether-based solution developed by an Israeli
company seems to have achieved the apex for lithium
metal-based rechargeable batteries.14 A lithium elec-
trode is highly stable in this solution up to 120 °C
because of excellent surface passivation,15 and over
300 depth discharge cycles have been reported.16 A
novel redox mechanism between the salt, LiAsF6, and
solvent, 1,3-dioxolane, shuts down the cell chemistry
at temperatures higher than 125 °C, thus preventing
thermal runaway.16,17 However, dendrite formation,
especially at high charge rates, still results in capac-
ity fade.18 In 1989 incidents of fire due to lithium
rechargeable batteries in the electronic devices, fol-
lowed by the manufacturer recalls, highlighted the
end of general enthusiasm in lithium metal as an
anode.19
1.3. From “Lithium” to “Lithium Ion”
The failure of lithium as an anode due to dendrite
formation prompted the search for a way to circum-
vent the drastic morphological change of the anode
during cell cycling. As a result, “host-guest” chem-
istry was considered. Also known as “intercalation”-
or “insertion”-type electrodes, this concept of revers-
ible chemistry had been applied earlier to cathode
materials for lithium batteries, as represented by the
trail-blazing work of Whittingham20,21 and the sig-
nificant improvements by Goodenough et al. and
others.22,23 Most of the host materials are transition
metal oxides or chalcogenides with stable crystal
lattices, and their layer or tunnel structures provide
the pathways for guest ions such as the lithium ion
to diffuse. By injecting or extracting electrons, the
redox reactions occur on the host lattice while mobile
guest ions intercalate into or deintercalate from the
host matrix to compensate for regional electroneu-
trality. During the whole intercalation/deintercala-
tion cycle, there are no Faradaic changes in the
“guest ion”. If a similar intercalation host could be
found and used as an anode material, then a battery
employing such intercalation cathodes and anodes
would only require the lithium ion to shuttle back
and forth between the electrodes without the pres-
ence of lithium metal. The nickname “rocking-chair
battery” was given to such a device that uses dual
intercalation electrodes,24 the working principle of
which is schematically depicted in Figure 1, using
the example of the state-of-the-art lithium ion chem-
istry.
The concept of rocking-chair lithium batteries was
confirmed experimentally by using lithiated oxides
(Li6Fe2O3, LiWO2) as interaction anodes and other
oxides (WO3, TiS2, V2O5) as cathodes in nonaqueous
electrolytes.25,26 However, the enhanced safety and
extended cycle life were not sufficient to offset the
penalty in energy density caused by the replacement
of lithium metal; hence, these systems were never
commercialized.27,28
A breakthrough was made when Japanese research-
ers exploited an old concept of using carbonaceous
materials as anode intercalation host.29-31 The term
“lithium ion battery” that was introduced by those
researchers eventually prevailed and replaced the
other aliases such as “rocking-chair”,24 “shuttle-
cock”,32 or “swing” batteries.33 In the charged state
of these carbonaceous anodes, lithium exists in its
ionic rather than metallic state, thus eliminating any
possibility of dendrite lithium. The advantage of this
new host is highlighted by the low cost of carbon and
the high lithium ion activity in the intercalation
compound; the latter renders an anode potential close
to that of lithium metal and minimizes the energetic
penalty. In 1990 both Sony34 and Moli35 announced
the commercialization of cells based on petroleum
coke and LiCoO2, though Sony was generally credited
for making this technology a commercial reality. In
the same year Dahn and co-workers published their
seminal report on the principle of lithium intercala-
tion chemistry with graphitic anodes and the effect
of electrolyte solvent in the process.36 In fact, the
conclusions drawn therein constitute the foundation
for the current lithium ion battery industry:
(1) Electrolyte solvents decompose reductively on
the carbonaceous anode, and the decomposition prod-
uct forms a protective film. When the surface of the
anode is covered, the film prevents further decom-
position of the electrolyte components. This film is
an ionic conductor but an electronic insulator.
Figure 1. Schematic description of a “(lithium ion) rock-
ing-chair” cell that employs graphitic carbon as anode and
transition metal oxide as cathode. The undergoing electro-
chemical process is lithium ion deintercalation from the
graphene structure of the anode and simultaneous inter-
calation into the layered structure of the metal oxide
cathode. For the cell, this process is discharge, since the
reaction is spontaneous.
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4305

(2) This reductive decomposition process occurs
only during the first charge and is absent in the
following cycles so that the carbonaceous anode can
be cycled many times in the electrolyte, yielding
stable capacity.
(3) The chemical structure of the electrolyte sol-
vents critically influences the nature of the protective
film, and ethylene carbonate was found to be an
essential component of the solvents that protects the
highly crystalline structure of graphite.
Obviously, the film formed on a carbonaceous
anode plays a critical role in enabling a lithium ion
device to work reversibly. Presuming that the surface
nature of the carbonaceous anode at low potentials
is similar to that of lithium metal in nonaqueous
electrolytes, Dahn and co-workers adopted a model
developed earlier by Peled to describe the passivation
on lithium metal37 and named this surface film on
carbonaceous anodes a “solid electrolyte interface”
(SEI). This term soon became the most frequently
used key word in publications concerning lithium ion
technology in the following decade. Although it will
turn out later that the exact mechanism involved in
the formation is far more complicated and remains
a controversial topic even today, it has been generally
agreed that the electrolyte reduction products are the
main components of an SEI and dictate the chemical
as well as thermal properties of the electrode.
The decade following Dahn’s publication witnessed
an explosive growth in lithium ion technology re-
search, and essentially all aspects of lithium ion
technology were explored with state-of-the-art tech-
niques, while the main excitement revolved around
developing new materials such as carbonaceous
anode and metal oxide cathode materials and the
electrolyte solvents and salts compatible with them.
The result of those intensified efforts was the suc-
cessful commercialization and the rapid thriving of
this youngest battery chemistry. By 2000, the quan-
tity of lithium ion cells manufactured reached 620
million units with a market value of 1 billion
dollars,38 accounting for more than a 90% share of
the rechargeable battery market39 or 63% of total
sales in portable batteries. The employment of new
materials and novel engineering designs has pushed
the cycle life, energy, and power density of this
technology to more than 2000 cycles, 160 W h kg-1,
and 5000 W kg-1, respectively.40 The major driving
force of this market remains the so-called “small
formula batteries” with capacities smaller than 1 A
h; however, industry-size lithium ion cells are in-
creasingly being used in space, military, and other
special applications, especially as traction power
sources for electric or hybrid electric vehicle (EV/
HEV) applications.40
1.4. Scope of This Review
Since the inception of lithium ion technology, there
have been several reviews summarizing the knowl-
edge accumulated about this new technology from
various perspectives, with the latest being in 2003.41-48
Because electrolytes interact closely with both cath-
ode and anode materials during the operation, their
effect on cell performance has been discussed in
almost every one of these reviews. On the other hand,
attention has always been focused on electrode
materials, especially the anodes, and electrolytes as
an important component of the cell have not been
comprehensively treated in any dedicated reviews.
This review intends to fill this deficit by summariz-
ing the progress made during the last 10 years in
the research and development of electrolytes for
lithium-based batteries. Since lithium ion chemistry
is by far the only successfully commercialized re-
chargeable lithium-based technology, emphasis will
be placed on the electrolytes developed for this
system. Liquid electrolytes will take the central
stage, and the scope of the review will include their
ionics, phase diagrams, interfaces with cathode and
anode materials, long-term chemical stability in the
device, thermal properties and performance at ex-
treme temperatures, and safety characterizations.
Whenever an interdisciplinary topic involving both
electrolyte and other cell components is encountered
(i.e., electrolyte/electrode interface and passivation
of electrodes), emphasis will be placed on the role and
effect of electrolyte components.
For the convenience of this discussion, a somewhat
arbitrary demarcation was drawn between “state-of-
the-art” (SOA) and “novel” electrolyte systems, with
the former referring to the ones currently used in
commercialized lithium ion cells and the latter to the
ones improved over the SOA systems but still under
development. It should be pointed out that the exact
electrolyte compositions in commercialized devices
are usually proprietary knowledge, but publications
from the affiliated researchers normally disclose
sufficient information to reveal the skeletal electro-
lyte components employed. The distinction made in
this review concerning the previously mentioned
demarcation is based on such open literature.
This review will focus on the literature published
from 1990 to the middle of 2003. Meanwhile, a
certain amount of attention will also be allocated to
the electrolytes for lithium batteries to avoid omitting
the important progress made in these closely related
fields. When selecting references, efforts were made
to ensure academic quality as well as ready public
accessibility. For this reason, patents, various techni-
cal reports, and conference/workshop presentations/
abstracts were avoided to the extent possible. There
were exceptions, though, when there was no alterna-
tive reference source. Finally, although comprehen-
sive coverage was attempted, it is essentially impos-
sible to cover every aspect in an exhaustive manner.
The choice of the references and the organization of
the content reflect the personal view of the author
only.
2. Electrolyte Components: History and State of
the Art
Most compositions of lithium electrolytes are based
on solutions of one or more lithium salts in mixtures
of two or more solvents, and single-solvent formula-
tions are very rare, if there are any. The rationale
behind this mixed solvent formulation is that the
diverse and often contradicting requirements of bat-
4306 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

tery applications can hardly be met by any individual
compound, for example, high fluidity versus high
dielectric constant; therefore, solvents of very differ-
ent physical and chemical natures are often used
together to perform various functions simultaneously.
A mixture of salts, on the other hand, is usually not
used, because anion choice is usually limited, and
performance advantages or improvements are not
readily demonstrated.
Solid polymer and gel polymer electrolytes could
be viewed as the special variation of the solution-type
electrolyte. In the former, the solvents are polar
macromolecules that dissolve salts, while, in the
latter, only a small portion of high polymer is
employed as the mechanical matrix, which is either
soaked with or swollen by essentially the same liquid
electrolytes. One exception exists: molten salt (ionic
liquid) electrolytes where no solvent is present and
the dissociation of opposite ions is solely achieved by
the thermal disintegration of the salt lattice (melt-
ing). Polymer electrolyte will be reviewed in section
8 (“Novel Electrolyte Systems”), although lithium ion
technology based on gel polymer electrolytes has in
fact entered the market and accounted for 4% of
lithium ion cells manufactured in 2000.38 On the
other hand, ionic liquid electrolytes will be omitted,
due to both the limited literature concerning this
topic and the fact that the application of ionic liquid
electrolytes in lithium ion devices remains dubious.
Since most of the ionic liquid systems are still in a
supercooled state at ambient temperature, it is
unlikely that the metastable liquid state could be
maintained in an actual electrochemical device,
wherein electrode materials would serve as effective
nucleation sites for crystallization.
2.1. Solvents
In accordance with the basic requirements for
electrolytes, an ideal electrolyte solvent should meet
the following minimal criteria: (1) It should be able
to dissolve salts to sufficient concentration. In other
words, it should have a high dielectric constant ().
(2) It should be fluid (low viscosity Ł), so that facile
ion transport can occur. (3) It should remain inert to
all cell components, especially the charged surfaces
of the cathode and the anode, during cell operation.
(4) It should remain liquid in a wide temperature
range. In other words, its melting point (Tm) should
be low and its boiling point (Tb) high. (5) It should
also be safe (high flash point Tf), nontoxic, and
economical.
For lithium-based batteries, the active nature of
the strongly reducing anodes (lithium metal or the
highly lithiated carbon) and the strongly oxidizing
cathodes (transition metal based oxides) rules out the
use of any solvents that have active protons despite
their excellent power in solvating salts, because the
reduction of such protons and/or the oxidation of the
corresponding anions generally occurs within 2.0-
4.0 V versus Li,49 while the charged potentials of the
anode and the cathode in the current rechargeable
lithium devices average 0.0-0.2 V and 3.0-4.5 V,
respectively. On the other hand, the nonaqueous
compounds that qualify as electrolyte solvents must
be able to dissolve sufficient amounts of lithium salt;
therefore, only those with polar groups such as
carbonyl (CdO), nitrile (CtN), sulfonyl (SdO), and
ether-linkage (-O-) merit consideration.
Since the inception of nonaqueous electrolytes, a
wide spectrum of polar solvents has been investi-
gated, and the majority of them fall into either one
of the following families: organic esters and ethers.
The most commonly used solvents from these fami-
lies, along with their physical properties, are listed
in Tables 1 and 2, respectively,50 where the melting
temperature of diethyl carbonate (DEC) deserves
special attention because a significant correction has
been made recently.50e
Table 1. Organic Carbonates and Esters as Electrolyte Solvents
a The mp of DEC recorded in various literature sources (books, papers, commercial catalogs) has been -43 °C, which was
corrected by a very recent measurement (ref 50e). This widespread error of 30° seems to stem from a single source in 1921, which
was then registered by Beilstein Handbuch and escaped detection for approximately eight decades.
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4307

An interesting observation should be made con-
cerning the dependence of the physical properties on
molecular cyclicity, since it will have a significant
effect on the formulation of electrolytes for lithium
ion cells. While all of the ethers, cyclic or acyclic,
demonstrate similar moderate dielectric constants
(2-7) and low viscosities (0.3-0.6 cP), cyclic and
acyclic esters behave like two entirely different kinds
of compounds in terms of dielectric constant and
viscosity; that is, all cyclic esters are uniformly polar
( ) 40-90) and rather viscous (Ł ) 1.7-2.0 cP), and
all acyclic esters are weakly polar ( ) 3-6) and fluid
(Ł ) 0.4-0.7 cP). The origin for the effect of molecular
cyclicity on the dielectric constant has been at-
tributed to the intramolecular strain of the cyclic
structures that favors the conformation of better
alignment of molecular dipoles, while the more flex-
ible and open structure of linear carbonates results
in the mutual cancellation of these dipoles.
2.1.1. Propylene Carbonate (PC)
Among these solvents, cyclic diesters of carbonic
acid have undoubtedly attracted the main research
attention throughout the entire history of lithium
batteries, especially in the past decade, when their
role in forming an SEI on carbonaceous anodes
became recognized. However, the early interest in
these compounds arose solely from their high dielec-
tric constant and, hence, their ability to dissolve a
wide variety of lithium salts. In 1958, it was observed
that lithium could be electrodeposited from a solution
of LiClO4 in PC,51 and PC became the immediate
focus of investigation.4,9,10 Its wide liquid range
(defined by the difference between Tm and Tb, Table
1), high dielectric constant, and static stability with
lithium made it a preferred solvent, and considerable
efforts were made to purify it52 when a less-than-ideal
stripping/plating efficiency (e 85%) for lithium was
observed during cycling. The capacity of lithium cells
with PC electrolytes also fades accordingly.4,9,10,54
However, it was soon realized that this poor cycling
efficiency was more intrinsic to the electrolyte solvent
than contaminant-originated, and the reaction be-
tween the PC and the newly deposited lithium
particles was thought to be the cause.9 More recent
studies employing spectroscopic means have con-
firmed the PC reduction on a newly formed lithium
surface, and a one-electron reduction process has
been proposed (see Scheme 1).55
The overall capacity fading of lithium cells using
PC-based electrolytes, however, is a more complicated
mechanism, although Scheme 1 plays a part in it.
The static stability of PC against a lithium surface
had been attributed earlier to the existence of a
protective layer,37,56 which consists of the decomposi-
tion products shown in Scheme 1, and prevents the
sustained reaction of PC with lithium.55 On the other
hand, it was the dynamic reactivity that results in
the lithium loss during cycling,9 for which the main
cause is related to the nonuniform morphological
change of the lithium surface rather than chemical
corrosion. Figure 2 schematically shows this nonuni-
formity of the lithium surface during the cycling
process, where uneven growth of the electrodeposited
lithium crystals results in dendrites that in subse-
quent discharge processes (lithium dissolution) pro-
duce lithium particles that are electrically isolated
from the lithium anode. Microscopic studies have
confirmed the existence of dendrites (Figure 3) and
have attributed their formation to the presence of a
passivation film.37,57 Serious safety hazards are often
caused by the generation of both dendrites and
isolated lithium crystals.58,59 The former creates
internal shorts, and the latter is chemically active
with the electrolyte solvents due to their huge surface
areas.
2.1.2. Ethers
In view of the poor cycling efficiency and the
potential hazards associated with PC, people turned
to ethers for improved lithium morphology. In the
1980s, ethers were widely preferred by researchers
as an alternative candidate, because of their low
Table 2. Organic Ethers as Electrolyte Solvents
Scheme 1. Reduction of PC on a Lithium Surface:
One-Electron Process
4308 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

viscosity and resultant high ionic conductivity, but,
most of all, the better lithium morphology during
cycling.60 The cycling efficiency of lithium was re-
ported to be 88% in tetrahydrofuran (THF),60-62 an
average of 96% in 2-methyltetrahydrofuran (2-Me-
THF),62-65 97% in polymethoxy ethers66 and dimethoxy
propane,67 and 98% in diethyl ether,62,68 although the
safety concern over the high vapor pressure of diethyl
ether renders it an impractical candidate. The forma-
tion of dendritic lithium seemed to be sufficiently
suppressed in these solvents even at high charge
rates.62 However, efforts to incorporate ether-based
electrolytes in lithium cells were still troubled by the
poor capacity retention,68-70 and prolonged cycling
(>100 cycles) of the cells still produced dendrite
deposition,71 which terminated the cells by creating
shorts,72 despite the improved lithium morphology
observed in the short term.
In addition to the problem with the lithium anode,
a new factor contributing to the capacity fade sur-
faced as the oxidative decomposition of ether-based
compounds on the cathode surface.61,67,73 Electro-
chemical studies on these ether-based electrolytes
placed the potential for the oxidative breakdown of
the ether functionality at relatively low potentials.
On a platinum surface, for example, THF was found
to be oxidized at 4.0 V vs Li, while PC remained
stable up to 5.0 V.74 In an actual cell, ethers are more
readily decomposed at even lower potentials, because
of the highly catalytic surface of cathode materials.76
With the application of more potent 4.0 V cathode
materials (LixMO2, M ) Mn, Ni, or Co) in lithium or
lithium ion cells, the possibility of using ether
compounds as solvents or cosolvents diminished.75,77
During the 1990s, various ethers were gradually
phased out in most of the electrolyte systems under
investigation. The failure of ether-based electrolytes
served as a perfect example to illustrate that, in a
battery, the electrolyte (solvents and salts) must cope
with challenges from both the anode and the cathode.
On the other hand, the advantage of organic esters,
especially cyclic alkyl carbonates, was rediscovered
because of their excellent stability against oxidation
on cathode surfaces.74,78,79
2.1.3. Ethylene Carbonate (EC)
The interest in alkyl carbonates was renewed with
the emergence of the lithium ion “shuttle” concept.24-31
Gone with the lithium anode is the difficult issue of
lithium morphology; subsequently, the higher anodic
stability of PC made it once again a promising
candidate. In the first generation of the commercial
lithium ion cells, a PC-based electrolyte was used by
Sony along with LixCoO2 as the cathode and petro-
leum coke as the anode.31 However, the real renais-
sance for using alkyl carbonates as lithium electrolyte
solvents was not brought about by PC but, quite
unexpectedly, by its high melting cousin EC.
Compared with PC, EC has comparable viscosity
and slightly higher dielectric constant, which are
favorable merits for a solvent candidate (Table 1). In
fact, its dielectric constant is even higher than that
of the the most common electrolyte solvent on the
earth: water (  79).3 However, because of its high
melting point (36 °C), it was never favored as an
ambient-temperature electrolyte solvent in the early
days of lithium battery research: the liquid range of
the electrolytes based on it would be too restricted.
Its higher melting point than those of other members
of the carbonate family (Table 1) is believed to arise
from its high molecular symmetry, which renders it
a better stabilized crystalline lattice.80, 81
EC was considered as an electrolyte cosolvent for
the first time by Elliot in 1964, who noted that, due
to the high dielectric constant and low viscosity of
EC, the addition of it to electrolyte solutions would
favor ion conductivity.82 The findings did not attract
particular attention from the battery community
Figure 2. (a) Schematic description for the growth of
dendrite crystals on a Li surface. The film consisting of
decomposition products as shown in Scheme 1 prevents the
growth of large granular crystals but rather promotes the
formation of treelike dendrites. (b) Schematic description
for the formation of isolated lithium particles from Li
dendrites. The uneven dissolution of the dendrites leaves
lithium crystals detached from the lithium substrate. The
isolated lithium crystals become electrochemically “dead”
but chemically reactive due to their high surface area.
Figure 3. Micrograph of a single dendrite lithium grown
in PC. (Reproduced with permission from ref 57 (Figure
6a). Copyright 1989 The Electrochemical Society.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4309

until the early 1970s, when Scrosati and Pistoia
exploited it to advantages for lithium battery elec-
trolytes. They reported that, owing to the suppression
of the melting point by the presence of the solute, a
room-temperature melt would form, and extra sup-
pression could be obtained when a small percentage
(9%) of PC was added.83 Further investigation found
that electrolytes based on EC as compared with PC
demonstrated improvements, not only in bulk ion
conductivity but also in interfacial properties such
as lower polarization on various cathode surfaces.84
Following these reports, EC began to appear as an
electrolyte cosolvent in a number of new electrolyte
systems under investigation, many of which still
contained ethers.72,73,85-88 However, the first com-
mercialized rechargeable lithium battery used an
ether-free composition, an EC/PC mixture, as the
electrolyte solvent.89,90 Despite the melting-point
suppression by the solute and other cosolvents, the
higher liquidus temperatures of the electrolyte due
to EC remained a factor limiting the low-temperature
applications of the lithium cell.
The unique position of EC as a lithium battery
electrolyte was established in 1990 when Dahn and
co-workers reported the fundamental difference be-
tween EC and PC in their effects on the reversibility
of lithium ion intercalation/deintercalation with gra-
phitic anodes.36 Despite the seemingly minute dif-
ference in molecular structure between the two, EC
was found to form an effective protective film (SEI)
on a graphitic anode that prevented any sustained
electrolyte decomposition on the anode, while this
protection could not be realized with PC and the
graphene structure eventually disintegrated in a
process termed “exfoliation” because of PC cointer-
calation. The reason for the effectiveness of the SEI
has incited a lot of research interest in the past
decade but remains an unsolved mystery, although
it is generally believed that EC undergoes a reduction
process on the carbonaceous anode via a similar path
to that shown in Scheme 1. Because of the important
role this SEI plays in lithium ion chemistry, the
research efforts on this topic will be reviewed in a
dedicated section (section 6).
2.1.4. Linear Carbonates
After Sony successfully marketed the first genera-
tion lithium ion cells, numerous competitors emerged
and a pursuit for higher energy density started. With
the energetic advantage of highly crystalline carbon
(graphitic) over disordered carbon being recognized,
EC became the core and indispensable component of
the electrolyte formulation.
During the early 1990s, efforts were made to
expand the limited liquid range of EC-based electro-
lytes by using different cosolvents, including PC,91,92
THF and 2-Me-THF,73,85-88 diethoxyethane (DEE),93,94
and dimethoxyethane (DME).45,95-97 None of these
cosolvents performed satisfactorily though, because
the presence of PC usually caused a large irreversible
capacity in the initial cycle of the lithium ion cell,36,91,92
while the ethers were found to be unstable against
the oxidation catalyzed by the surface of the charged
cathode.93,94 Thus, it was generally realized that, for
a lithium ion cell that employs graphite as an anode
and 4.0 V metal oxide (LiMO2, M ) Co, Ni) as a
cathode, the electrolyte must have oxidative stability
up to 5 V vs Li.95-97
In 1994 a formulation that successfully met such
a standard was first described in open literature by
Tarascon and Guyomard, who used a linear carbon-
ate, dimethyl carbonate (DMC), as a cosolvent with
EC.98,99 As it has been pointed out, linear carbonates
differ from their cyclic cousins by their low boiling
points, low viscosity, and low dielectric constant.
They can form homogeneous mixtures with EC at any
ratio, and the resultant mixed electrolytes benefit not
only from the melting-temperature suppression of EC
but also from the low viscosity (higher ion conductiv-
ity) of DMC. But what surprises researchers is the
wide electrochemical stability window of this mixture
electrolyte: it remains stable on a spinel cathode
surface up to 5.0 V. Considering that these linear
carbonates, in the absence of EC, are readily liable
to oxidation on cathode surfaces at 4.0 V vs Li,76
the origin for the above improvement in the electro-
chemical window remains unclear, because the an-
odic stabilities of the ether-based electrolytes were
hardly raised by their mixing with EC93,94 or PC.95,96
It seems that a synergistic effect is achieved when
EC and DMC (or other linear carbonates) are mixed
because the merits of each individual solvent are
imparted on to the resultant mixture: high anodic
stability of EC on cathode surfaces, high solvation
power of EC toward lithium salts, and low viscosity
of DMC to promote ion transport.
This new formulation of electrolytes based on a
mixture of EC with a linear carbonate set the main
theme for the state-of-the-art lithium ion electrolytes
and was quickly adopted by the researchers and
manufacturers.97,100-103 Other linear carbonates were
also explored, including DEC,104-106 ethylmethyl
carbonate (EMC),107 and propylmethyl carbonate
(PMC),108,109 and no significant differences were found
between them and DMC in terms of electrochemical
characteristics. The direct impact of this electrolyte
innovation is that the first generation carbonaceous
anode petroleum coke was soon replaced by graphitic
anode materials in essentially all of the lithium ion
cells manufactured after 1993. At present, the elec-
trolyte solvents used in the over one billion lithium
ion cells manufactured each year are almost exclu-
sively based on the mixture of EC with one or more
of these linear carbonates, although each individual
manufacture may have its own proprietary electro-
lyte formulation.
2.2. Lithium Salts
An ideal electrolyte solute for ambient rechargeable
lithium batteries should meet the following minimal
requirements: (1) It should be able to completely
dissolve and dissociate in the nonaqueous media, and
the solvated ions (especially lithium cation) should
be able to move in the media with high mobility. (2)
The anion should be stable against oxidative decom-
position at the cathode. (3) The anion should be inert
to electrolyte solvents. (4) Both the anion and the
cation should remain inert toward the other cell
components such as separator, electrode substrate,
4310 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

and cell packaging materials. (5) The anion should
be nontoxic and remain stable against thermally
induced reactions with electrolyte solvents and other
cell components.
The available choice of lithium salts for electrolyte
application is rather limited when compared to the
wide spectrum of aprotic organic compounds that
could make possible electrolyte solvents. This differ-
ence could be more clearly reflected in a comprehen-
sive report summarizing nonaqueous electrolytes
developed for rechargeable lithium cells, in which
Dahn and co-workers described over 150 electrolyte
solvent compositions that were formulated based on
27 basic solvents but only 5 lithium salts.50b
Because of the small ionic radius of lithium ion,
most simple salts of lithium fail to meet the minimum
solubility requirement in low dielectric media. Ex-
amples are halides, LiX (where X ) Cl and F), or the
oxides Li2O. Although solubility in nonaqueous sol-
vents would increase if the anion is replaced by a so-
called “soft Lewis base” such as Br-, I-, S2-, or
carboxylates (R-CO2-), the improvement is usually
realized at the expense of the anodic stability of the
salt because these anions are readily oxidized on the
charged surfaces of cathode materials at <4.0 V vs
Li.
Most of the lithium salts that are qualified for the
minimal solubility standard are based on complex
anions that are composed of a simple anion core
stabilized by a Lewis acid agent. For example, the
anion of lithium hexafluorophosphate (LiPF6) could
be viewed as F- complexed by the Lewis acid PF5.
Such anions, also known as anions of superacids,
have a structure in which the formal negative charge
is well distributed by the strongly electron-withdraw-
ing Lewis acid ligands, and the corresponding com-
plex salts are usually lower melting and better
soluble in low dielectric media than their parent
salts.
The requirement for chemical inertness further
excluded a family of lithium salts that have been
widely used in primary lithium batteries: LiAlX4
(X ) halides).110 Since the Lewis acidities of the AlX3
are so strong, their complexation with the moderate
bases such as Cl- does not fully neutralize their
activity, and as a result, they would attack most of
the nonaqueous solvents, especially ethers. The AlX4-
anions also cause severe corrosion to other cell
components such as the separators, usually made of
polypropylene, and the insulating sealant, as well as
the metallic packaging materials. On the other hand,
anions based on milder Lewis acids can remain stable
with organic solvents under normal conditions (e.g.,
ambient temperature) and have been preferentially
investigated by researchers. These salts include
lithium perchlorate (LiClO4) and various lithium
borates, arsenates, phosphates, and antimonates,
LiMXn (where M ) B or As, P, and Sb and n ) 4 or
6, respectively). Table 3 lists some examples of these
salts along with some basic physical properties,50c
including ion conductivity data at room temperature
in PC or EC/DMC (1:1), respectively.50d,111-116 A brief
summary of a few selected lithium salts of signifi-
cance during the development of lithium cell electro-
lytes is given below.
2.2.1. Lithium Perchlorate (LiClO4)
LiClO4 has been a popular electrolyte solute owing
to its satisfactory solubility and high conductivity
(9.0 mS cm-1 in EC/DMC at 20 °C) as well as its
high anodic stability (up to 5.1 V on a spinel cathode
surface in EC/DMC).99 Recent studies found that SEI
films formed in LiClO4 electrolytes, on both lithium
and carbonaceous anode surfaces, are of lower im-
pedance than those formed in LiPF6 or lithium
tetrafluoroborate (LiBF4) electrolytes, because the HF
is absent in the former.104,117 It is believed that HF,
generated as the hydrolysis product of LiPF6 and
LiBF4 by trace moisture in the electrolyte solvents,
reacts with either alkyl carbonate or Li2CO3 and
forms the highly resistive LiF.117,118 Compared with
other lithium salts, LiClO4 also has the merits of
being relatively less hygroscopic and is stable to
ambient moisture.
However, the high oxidation state of chlorine (VII)
in perchlorate makes it a strong oxidant, which
readily reacts with most organic species in violent
ways under certain conditions such as high temper-
Table 3. Lithium Salts as Electrolyte Solutes
a Reference 111. b Reference 146. c Reference 114. d Reference 115. e Reference 116.
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4311

ature and high current charge.58,119 Actually, back in
the 1970s it had already been realized that LiClO4
was impractical as an electrolyte solute for industry
purposes;119 nevertheless, it is still frequently used
as a salt of convenience in various laboratory tests
because it is easy to handle and economical.43,79,91,94
2.2.2. Lithium Hexafluoroarsenate (LiAsF6)
While researchers focused on the morphology of
lithium cycling in nonaqueous electrolytes during the
late 1970s, it was found that the salt plays an
important role beside the solvents, and in general,
LiAsF6 was a superior salt to LiClO4 as an electrolyte
solute for lithium batteries.54 For a long period, the
combination of LiAsF6 with various ethers became
the most popular system under investigation.60,62,68,69,78
On average, lithium cycling efficiencies could reach
>95% in these systems,54,62,66 although long-term
cycling in these electrolytes still promoted the growth
of lithium dendrites.71 Chemical deterioration was
also detected, as indicated by the discoloration of the
LiAsF6/2-Me-THF solution with time, and a reaction
between LiAsF6 and the solvent was suspected.65,68
A mechanism was proposed based on the Lewis
acidity of As(V), which cleaves the ether linkage and
produces a series of gaseous and polymeric products
(see Scheme 2).68
The cathodic stability of the AsF6- anion was
studied on a glassy carbon surface, and a reduction
process was found at 1.15 V vs Li:120
The above process was observed only in the initial
cycles. Nevertheless, any electrochemical reduction
of As(V) would raise concern about the safety of using
LiAsF6 in a commercial battery, because, while
arsenate in its high oxidation state (V) is not par-
ticularly toxic, As(III) and As(0) species are.68,120-122
From the electrochemical point of view, however, the
above reduction could be a benefit, especially for
lithium ion cells, since an SEI formed on an anode
at >1.0 V vs lithium would be very stable during the
operation of a lithium ion cell according to a semi-
empirical rule,106 which will be discussed in more
detail in section 6.
Very similar to the case of LiClO4, an SEI formed
from LiAsF6-based electrolytes, either on a lithium
or carbonaceous anode, mainly consists of alkyl
carbonates or Li2CO3 rather than LiF, as one would
expect from the behavior of its close structural
brothers LiPF6 or LiBF4.104,117,118 This can be at-
tributed to the much less labile As-F bond that is
resistive to hydrolysis.120
The anodic stability of the AsF6- anion proved to
be high. In proper solvents, such as esters rather
than ethers, the electrolyte based on this salt can
remain stable up to 4.5 V on various cathode sur-
faces.78,99 The combination of cathodic and anodic
stability would have made LiAsF6 a very promising
candidate salt for both lithium and lithium ion
batteries had the toxicity not been a source of
concern. Instead, it was never used in any com-
mercialized cells but is still frequently used in
laboratory tests even today.14-18,123-125
2.2.3. Lithium Tetrafluoroborate (LiBF4)
Like LiAsF6, LiBF4 is a salt based on an inorganic
superacid anion and has moderate ion conductivity
in nonaqueous solvents (Table 3). It was out of favor
in the early days of lithium battery research because
the ether-based electrolytes containing it were found
to result in poor lithium cycling efficiencies, which
decayed rapidly with cycle number.60,126-128 The
reactivity of LiBF4 with lithium was suspected as
discoloration occurred with time or heating.128
The researchers who initiated the study on LiBF4
mentioned the multiple advantages of LiBF4 as
compared with other salts (e.g., less toxicity than
LiAsF6 and higher safety than LiClO4),128 but its
moderate ion conductivity has been a major obstacle
to its application. More recent studies on its ionics
and limiting properties in various nonaqueous sys-
tems established that, among the most common
anions encountered, BF4- has the highest mobility,
but its dissociation constant is significantly smaller
than those of LiAsF6 and LiPF6.111,129 The unfavorable
balance of these two properties results in the moder-
ate ion conductivity.
Electrochemically, the BF4- anion was found to be
stable against oxidation on a glassy carbon (GC)
surface up to 3.6 V vs a standard calomel electrode
(SCE), which translates into 5.0 V vs lithium.130,131
When a distinction is made, this stability limit is
somehow lower than those of AsF6- and PF6- anions;
however, caution must be exercised here, as these
data were measured on GC with quaternary am-
monium as supporting electrolyte, instead of on a
surface of cathode materials. This could result in
substantial difference.76
The use of LiBF4 in lithium-based cells has been
rare because of its inferior ion conductivity until
recently, when the thermal instability of LiPF6 and
the moisture sensitivity became recognized. Attempts
to replace LiPF6 in lithium ion cells have been made,
and the cells based on LiBF4 electrolytes showed
improved performance, not only at elevated temper-
atures up to 50 °C132 but, surprisingly, also at low
temperatures as well.132-135 These observations could
bring this salt back to research favor.
2.2.4. Lithium Trifluoromethanesulfonate (LiTf)
Another family of lithium salts is based on the
conjugate bases of the organic superacids, where acid
strength is increased because of the stabilization of
anions by the strongly electron-withdrawing groups,
usually perfluorinated alkyls. In these anions, the
Scheme 2. Reaction between LiAsF6 and Ether
Solvents
AsF6
- + 2e h AsF3 + 3F
-
4312 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

conducting LiTf, the hazardous LiClO4, the thermally
unstable LiBF4 and LiPF6, and the toxic LiAsF6 in
lithium battery applications.122 Extensive studies of
this salt were carried out to determine its ionics in
nonaqueous solutions111,116,129,130,131,139 and its applica-
tions in lithium or lithium ion cells.45,96,144,145
LiIm proved to be safe, thermally stable, and highly
conducting: it melts at 236 °C without decomposition
(a rarity among lithium salts) and does not decom-
pose until 360 °C.146 Its ion conductivity in THF is
an order of magnitude higher than that of LiTf,
although lower than those of LiAsF6 and LiPF6,111,116,122
and when no Al is used, a lithium ion cell based on
a LiNiO2 cathode and a petroleum coke anode can
yield up to 1000 deep discharge cycles with LiIm in
an EC/DMC solution.45
Ionics studies by Webber and Ue revealed that this
salt dissociates very well even in low dielectric
solvents, although its large anion size usually results
in a higher solution viscosity than those of other salts
in a given solvent system. Thus, its good ion conduc-
tivity should be the result of a compromise between
a high degree of dissociation and low mobility.122,129
In this sense, LiIm favors solvents with a low
dielectric constant. Electrochemical stability tests
were carried out on a GC electrode, and Im- was
found to be stable against oxidation in EC/DMC up
to 2.5 V vs a Ag+/Ag reference, which translates to
5.0 V vs Li, an oxidation limit lower than those for
LiBF4 and LiPF6,130,131 but still high enough to be
practical. The morphology of cycling lithium in LiIm-
based electrolytes is apparently superior to that in
other salt-based electrolytes.139
Despite all of these merits, the application of LiIm
in lithium ion cells never materialized because it
caused severe Al corrosion in electrolytes based on
it.141 In situ surface studies using EQCM established
a reaction between the Im- anion and the Al sub-
strate in which Al(Im)3 is produced and adsorbed on
the Al surface.147 Undoubtedly, this corrosion of a key
component of the cell by Im- greatly restricts the
possible application of LiIm, because the role of Al
as a cathode substrate in the lithium-based battery
industry is hard to replace, due to its light weight,
resistance to oxidation at high potential, excellent
processability, and low cost.
Efforts were made to reduce the reactivity of Im-
toward Al. Other salts that could passivate Al were
used as additives with LiIm, and the results were
encouraging.147 Also, structural modification of the
imide anion was made by extending the perfluori-
nated alkyl chain and was found to be effective,
although at the price of lower ion conductivity.141,148
Although LiIm has never been used in any com-
mercial lithium ion devices, it remains an interesting
salt to be investigated, especially for the polymer-
based electrolytes.47
2.2.6. Lithium Hexafluorophosphate (LiPF6)
Among the numerous salts vying for lithium/
lithium ion batteries, LiPF6 was the obvious winner
and was eventually commercialized. The success of
LiPF6 was not achieved by any single outstanding
property but, rather, by the combination of a series
of well-balanced properties with concomitant com-
promises and restrictions. For example, in the com-
monly used carbonate solvent mixtures it has a lower
conductivity than LiAsF6 (Table 3),99,111,116 a lower
dissociation constant than LiIm,129 a lower ionic
mobility than LiBF4,129 a lower thermal stability than
most of the other salts,149 a lower anodic stability
than LiAsF6 and LiSbF6,130,131 and a lower chemical
stability toward ambient moisture than LiClO4, LiIm,
and LiTf. However, none of these other salts could
meet all these multifaceted requirements simulta-
neously as well as LiPF6 does.
LiPF6 was proposed as an electrolyte solute for
lithium-based batteries in the late 1960s,149 and soon
its chemical and thermal instabilities were known.150
Even at room temperature, an equilibrium exists:
The generation of the gaseous product, PF5, drives
the equilibrium to the right, and this process is
favored by elevated temperatures. In the presence of
nonaqueous solvents, the strong Lewis acid PF5 tends
to initiate a series of reactions, such as ring-opening
polymerization or cleavage of ether linkages (Schemes
2 and 4).68,151,152
On the other hand, the P-F bond is rather labile
toward hydrolysis by even trace amounts of moisture
in nonaqueous solvents, producing a series of cor-
rosive products (Scheme 5). Thermal gravimetric
analysis (TGA) reveals that, in a dry state, LiPF6
loses 50% of its weight at >200 °C141 but that, in
nonaqueous solutions, the deterioration occurs at
substantially lower temperatures, for example, as low
as 70 °C.
The sensitivity of LiPF6 toward ambient moisture,
solvents, and high temperature not only restricts its
range of applications, especially in nonaqueous bat-
teries, but also causes tremendous difficulty in its
preparation and purification.150 Before the 1990s,
most of the commercially available LiPF6 had high
amounts of LiF and HF, and the purity issue became
part of the reason the potential of LiPF6 was not fully
realized until recently. The manufacture of high-
purity LiPF6 (HF < 10 ppm) in industrial scale was
achieved by Japanese companies in the late 1980s,
finally leading to the commercialization of lith-
ium ion technology and the ensuing extensive re-
search.86,120,122
In nonaqueous solvents based on mixed alkyl
carbonates, LiPF6 remains one of the most conducting
salts. For example, in EC/DMC (1:1) the conductivity
is 10.7 mS cm-1, only fractionally lower than that of
Scheme 4. Decomposition of Carbonates by PF5
Scheme 5. Hydrolysis of LiPF6 Salts by Moisture
LiPF6(s) h LiF(s) + PF5(g)
4314 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

Even if LiPF6 is replaced by more thermally stable
salts, the thermal stability of passivation films on
both the anode and the cathode would still keep the
high-temperature limits lower than 90 °C, as do the
thermal stability of the separator (<90 °C for polypro-
pylene), the chemical stability of the insulating
coatings/sealants used in the cell packaging, and the
polymeric binder agents used in both cathode and
anode composites.
The lower boundary of the liquid range, on the
other hand, does usually serve as the low-tempera-
ture limit for the electrolytes. The mp of LiPF6/EC/
DMC was determined as a function of solvent com-
position by Tarascon and Guyomard, who concluded
that LiPF6/EC/DMC could be used in the solvent
compositions between 3:7 and 8:2 at temperatures
down to -25 °C.99 The mp-dependence on solvent
composition that was reported in the work, however,
does not seem to be rational, since it shows a long
plateau at  -10 °C in 30-90% DMC after an initial
drop from 20 °C at 20% DMC,99 whereas a typical
simple eutectic feature would be expected instead.
When a closer comparison is made between the
results by Tarascon and Guyomard and the more
recent studies by Ding et al.,50e,159,160 it becomes
obvious that in the former work no distinction was
made between the liquidus and solidus temperatures;
that is, the mp’s at 20% and 90% DMC are liquidus
temperatures, while the rest (the plateau section)
seem to be solidus temperatures as measured by Ding
et al. This confusion could have arisen from the fact
that, for the intermediate compositions between 30%
and 80% DMC, the liquidus transition usually ap-
pears as a broad peak of relatively negligible thermal
effect that could evade notice, while the solidus
transition usually appears as a sharp and conspicu-
ous peak.
Systematic construction of carbonate mixture phase
diagrams was performed by Ding et al.50e,159,160 for a
series of binary carbonate solvent systems that are
frequently used as lithium ion electrolyte solvents.
It was found that all of the binary combinations
between EC, PC, DMC, EMC, and DEC yield the
same basic feature of simple eutectic-type phase
diagrams, characterized by the V-shaped liquidus
lines intersecting at the eutectic composition with a
horizontal solidus line, although the details of each
individual diagram vary greatly depending on their
mp and cyclicity. Figure 4 shows a collection of such
phase diagrams mapped for some EC-based binary
solvent systems.
In these phase diagrams, the liquidus line repre-
sents the temperature at which one of the compo-
nents crystallizes, while, below the solidus line, the
whole system solidifies. Between the solidus and
liquidus lines are the regions where solid and liquid
coexist. Since there is no solid phase above the
liquidus lines and the liquid is thermodynamically
stable, Ding et al. suggested that the liquidus tem-
peratures should be adopted as the lower boundary
of the liquid phase, instead of the solidus tempera-
tures.159,160 The patterns of these phase diagrams are
typical of binary systems in which the two compo-
nents are mutually soluble in their liquid states but
insoluble in their solid states; therefore, solid solu-
tions do not form below solidus temperatures, and
the binary system exists in the form of a heteroge-
neous mixture of both solids.
Typically, the liquidus lines of a binary system
curve down and intersect with the solidus line at the
eutectic point, where a liquid coexists with the solid
phases of both components. In this sense, the mixture
of two solvents should have an expanded liquid range
with a lower melting temperature than that of either
solvent individually. As Figure 4 shows, the most
popular solvent combination used for lithium ion
technology, LiPF6/EC/DMC, has liquidus lines below
the mp of either EC or DMC, and the eutectic point
lies at -7.6 °C with molar fractions of 0.30 EC and
0.70 DMC. This composition corresponds to volume
fractions of 0.24 EC and 0.76 DMC or weight frac-
tions of 0.28 EC and 0.71 DMC. Due to the high mp
of both EC (36 °C) and DMC (4.6 °C), this low-
temperature limit is rather high and needs improve-
ment if applications in cold environments are to be
considered.
A rather counterintuitive conclusion that can be
extracted from these phase diagrams, however, is
that simply introducing a low-melting component
does not necessarily extend the liquid range, as
evidenced by the replacement of DMC by the lower
melting EMC (-53 °C) in an EC-based binary solvent
system (Figure 4). The mismatch between the EC and
EMC mp’s creates a liquidus line that approaches the
mp of EC for most of the compositions; therefore, the
liquid range toward the low-temperature end actually
shrinks as compared with that of the EC/DMC binary
system. The replacement of DMC by another low-
melting linear carbonate, DEC (-74 °C), produces a
similar effect.23d It was proposed later that this poor
compatibility between EC and linear carbonates
Figure 4. Liquid-solid phase diagrams of EC/DMC, EC/
EMC, and PC/EC. (Reproduced with permission from ref
159 (Figure 9). Copyright 2000 The Electrochemical Soci-
ety.)
4316 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

due to the amount of experimental work required.
Consequently, the possibility of using computer
modeling to circumvent the laborious experimental
mapping was considered.167 Because a phase diagram
is merely a graphic representation of thermodynam-
ics through minimizations in the free energy of the
system under certain constraints, free-energy model-
ing of individual phases in a multicomponent system
should represent the phase boundary through phase
equilibrium calculations. Since the electrolyte sys-
tems are usually used at constant pressure condi-
tions, the Gibbs free energy is the quantity minimized
in calculating phase equilibria, and the computa-
tional approach commonly known as the CALPHAD
method was used by Liu. The initial attempt seemed
to be successful because phase diagrams of unary
systems for neat DMC, EC, and PC, and of binary
systems of EC/DMC, PC/DMC, and EC/PC were
reproduced with satisfactory accuracy when com-
pared with the work of Ding et al. Furthermore, by
combining the three binary systems, a ternary phase
diagram for EC/PC/DMC was predicted as shown in
Figure 5. As one would expect, the eutectic composi-
tion was heavily PC-rich because PC was the lowest
melting of the three components. Although the eu-
tectic composition as predicted in the ternary phase
diagram (0.10 EC/0.15 DMC/0.75 PC) was impracti-
cal for cell applications because of PC’s destructive
effect on most carbonaceous anodes, the effectiveness
of PC, when used in smaller amounts (<30%), in
improving low-temperature performance had been
confirmed.168 Undoubtedly, continued efforts to pre-
dict the thermal properties of ternary or higher order
systems should be encouraged for prospective elec-
trolytes, as they would provide useful knowledge for
estimating their operating temperature range.
4. Ion Transport Properties
The ability to conduct ions is the basic function of
electrolytes, which would determine how fast the
energy stored in electrodes can be delivered. In liquid
electrolytes, the transport of ions is realized via a
two-step process: (1) the solvation and dissociation
of ionic compounds (usually crystalline salts) by polar
solvent molecules and (2) the migration of these
solvated ions through the solvent media.
During the solvation, the stability of the salt crystal
lattice is energetically compensated by the coordina-
tion of solvent dipoles with the naked ions (especially
cations); therefore, these ions should always migrate
with a “solvation sheath” around them, which con-
sists of a certain number of oriented solvent mol-
ecules. According to the results obtained from various
modeling approaches including ab initio quantum
mechanics, the small ionic radius of lithium usually
allows no more than four solvent molecules in its
solvation sheath.129,169-171 Using a new mass spec-
trum (MS) technique, a more recent determination
of the coordination number (CN) for lithium ions
seems to support these computational results: among
the various solvated lithium ion species, the peaks
corresponding to Li(solv)23+ are the most abundant
no matter whether the solvent is EC, PC, or ç-buty-
rolactone (çBL).172,173 This latter experimental ob-
servation also revealed the stability of the solvation
sheath, which obviously remains intact even during
the electrospraying under vacuum and the subse-
quent ionization process. Therefore, there should be
sufficient confidence in the general belief that the
composition of solvated lithium ion remains un-
changed during its migration in an electrolyte solu-
tion.1 Considering that both cation and anion could
be coordinated by solvents, ion conduction actually
consists of the oriented movement of ion/solvent
complexes of both charges.
Ionic conductivity ó, which quantifies the ion
conduction ability, reflects the influence of these two
aspects, that is, solvation/dissociation and the sub-
sequent migration, in terms of the free ion number
ni and the ionic mobility íi,:
where Zi is the valence order of ionic species i, and e
is the unit charge of electrons. For a single salt
solution, the cations and anions are the only two
charged species present.174
Ion conductivity has essentially become the quan-
tity used as the field-trial standard for any prospec-
tive electrolytes, because it can be easily measured
with simple instrumentation, and the results are
highly accurate and reproducible. The methodology
and the fundamental principles involved with the
measurement have been summarized in a detailed
review.175 On the other hand, no reliable method has
been available so far for the exact determination of
ion mobility (or a related property, diffusivity Di) and
ionization degree, especially in electrolyte solutions
in the concentration ranges of practical interest.1
The lack of ionic mobility data causes a serious
inconvenience when the ion conduction ability of an
electrolyte is evaluated, because the measured con-
ductivity is the result of the overall migration of both
anions and cations, while for lithium batteries only
the portion of the current that was carried by the
lithium cation matters. This portion of the current
from lithium ion movement, which determines the
Figure 5. Calculated ternary phase diagram for EC/PC/
DMC as expressed in the form of a composition triangle
plane. The dotted lines represent the isotherms with 10 K
intervals with 300 K marked. (Reproduced with permission
from ref 167 (Figure 12). Copyright 2003 The Electrochemi-
cal Society.)
ó ) ∑
i
niíiZie (1)
4318 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

rate at which the battery operates, is quantified by
the lithium ion transference number (tLi):
There have been numerous efforts aimed at esti-
mating the lithium ion transference numbers in
nonaqueous solutions, and the data obtained via
different approaches vary appreciably. Nevertheless,
it is generally accepted that, in much diluted non-
aqueous solutions, lithium ion transference numbers
range from 0.20 to 0.40, depending on the properties
of salts and solvents.1,111,129,138,176-178 In other words,
the anions are much more mobile than the lithium
ions in nonaqueous electrolytes. The small cation
current portion in nonaqueous electrolytes is believed
to be caused by the high surface charge density on
the cations (especially lithium ion) due to their small
ionic radii, so they are much more favorably solvated
and must move at slower speed with the solvation
sheath, while high populations of anions could re-
main relatively “naked”. Solvation enthalpy calcula-
tions for cations and anions support this argu-
ment:169,176,178 in typical carbonate solvents, the former
range between 20 and 50 kcal mol-1 while the latter
are below 10 kcal mol-1.
A lithium ion transference number significantly
less than 1 is certainly an undesired property,
because the resultant overwhelming anion movement
and enrichment near electrode surfaces would cause
concentration polarization during battery operation,
especially when the local viscosity is high (such as
in polymer electrolytes), and extra impedance to the
ion transport would occur as a consequence at the
interfaces. Fortunately, in liquid electrolytes, this
polarization factor is not seriously pronounced.
The unavailability of data on dissociation degree
and mobility has thus made ion conductivity an
alternative metric that has been universally adopted
by the battery research and development community
to evaluate the transport ability of electrolytes.
However, it should always be remembered that such
a metric of convenience is based on an unstated
assumption; that is, the increase in the overall
conductivity should originate, at least partially, from
the improvement in the cation conductivity. Quali-
tatively, this assumption holds true, since a correla-
tion does usually exist between ion conductivity and
power performance in batteries, although quantita-
tively the distribution of this increase between anions
and cations is unknown.
The efforts to improve ion conductivity have re-
volved around eq 1, that is, aiming at increasing
either the salt dissociation degree (ni) or the ionic
mobility (íi). Since these two factors are decided
simultaneously by the physicochemical natures of the
salt and solvents, different approaches involving
either of these electrolyte components have been
adopted.
For lithium electrolytes, the only variable in salt
structure is the anion. In a given nonaqueous solvent
system, the dissociation of a lithium salt would be
facilitated if the anion is well stabilized by electron-
withdrawing functionalities. Successful examples of
such anions include PF6- or Im-, whose lithium salts
dissociate readily as compared to those based on the
parental anions (LiF and lithium alkylamide, respec-
tively).
On the other hand, the mobility of an ion is known
to vary inversely with its solvation radius ri according
to the Stokes-Einstein relation:1
where Ł is the viscosity of the media. With the cation
species fixed, this approach seems to be of little use
to increase cation mobility. However, a larger anion
with lower anion mobility shows the application of
this approach in another way, which results in a
higher cation transference number, although the
overall conductivity could decrease because of the
reduction in the anion contribution. This effect was
observed in imide and its derivatives.138 The extrem-
ity of this approach was represented by the salts with
oligomeric or polymeric anions, where t+ approaches
1.0 but the overall ion conductivities suffer
drastically.136,137,179-182 Hence, the approach of using
large anions to enhance tLi is not widely pursued in
liquid electrolytes.
So far, very few attempts at improving ion conduc-
tivity have been realized via the salt approach,
because the choice of anions suitable for lithium
electrolyte solute is limited. Instead, solvent composi-
tion tailoring has been the main tool for manipulating
electrolyte ion conductivity due to the availability of
a vast number of candidate solvents. Considerable
knowledge has been accumulated on the correlation
between solvent properties and ion conductivity,1 and
the most important are the two bulk properties of
the solvents, dielectric constant  and viscosity Ł,
which determine the charge carrier number (ni) and
ion mobility (íi), respectively.
In order for a solvated ion to migrate under an
electric field, it must be prevented from forming close
ion pairs with its counterions by the solvating
solvent. The effectiveness of the solvent molecule in
shielding the interionic Coulombic attraction is closely
related with its dielectric constant. The critical
distance for the ion pair formation q is given by eq 4
according to Bjerrum’s treatment, with the hypoth-
esis that ion-pair formation occurs if the interionic
distance is smaller than q:1,183
where z, 0, k, and T are the valence orders of ions,
the dielectric constant of vacuum, Boltzmann’s con-
stant, and temperature, respectively. Apparently, in
a solvent with a higher dielectric constant, ions would
have a higher probability of staying free at a given
salt concentration and ion association would be less
likely to occur. Most of these solvents are of high
boiling temperature and high viscosity (Table 1).
tLi )
íLi
∑
i
íi
(2)
íi )
1
6ðŁri
(3)
q )
jzizjje
2
8ð0kT
(4)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4319

Further analysis of related studies seems to argue
that the success of the high-/low-Ł combination
might not be due to the simple additive effect of these
two properties but, rather, a synergistic action of
these two variables through a mechanism that in-
volves the solvent’s preference for cations in its
solvation sheath. During the dissolution process of a
certain salt lattice in mixed solvents, the solvation
of the ions by a solvent molecule of a higher dielectric
constant would be energetically favored over that by
a molecule of a lower dielectric constant. Conse-
quently, it would be reasonable to expect that, after
equilibrium is established in the solution, the ions
would have solvation sheaths that are mainly com-
posed of the high- solvents. Modeling results using
molecular quantum mechanics support this hypoth-
esis by showing that the less favored solvent mol-
ecules (low ) actually could be readily replaced by
the favored ones.197 Thus, in the EC/EMC system, the
solvation shell should be predominantly composed of
EC.171 The most forceful evidence of this comes from
the experimental observation, where a new MS
technique employing low-energy ionization was ap-
plied to electrolytes based on a series of binary
compositions, including EC/DEC, EC/DMC, PC/DEC,
PC/DMC, çBL/DEC, and çBL/DMC. In each of these
systems, the overwhelmingly abundant species de-
tected were the solvation complexes of lithium with
the cyclic solvents (EC, PC, or çBL), with coordina-
tion numbers between 2 and 3.173
This selective solvation of lithium ions by high-
solvent molecules would exclude the solvents of low-Ł
from the solvation sheath and leave the latter as free,
noncoordinating solvent molecules. As a result, the
media in which the solvated ions migrate are mainly
composed of these free solvent molecules, which
impart their low-Ł to benefit the movement of the
solvated ions. In this way, a synergistic participation
from both high- and low-Ł solvents contributes to
the optimization of ion conduction.
The implication of such a picture of the solution
structure on the microscopic level not only concerns
ion transport but also further relates to the electro-
chemical stability of the electrolytes in lithium ion
cells, because these solvent molecules in the solvation
sheath, such as EC or PC, migrate with the ions to
electrode surfaces and are probably more involved in
the oxidative or reductive processes than the non-
coordinating, low-Ł solvent molecules, such as the
linear carbonates. This could have a profound impact
on the chemical nature of the electrolyte/electrode
interfaces (section 6).
Knowing how ion conduction is determined by the
interplay between the dielectric constant and viscos-
ity, the dependence of ion conductivity on different
variables that are of practical interest can be ex-
plained consistently. Extensive studies have been
carried out on the effects of salt concentration, solvent
composition, and temperature on ion conductivity in
different electrolyte systems,113,195,196,198,199 among
which the most representative is the meticulous work
by Ding et al. on a series of binary systems pertinent
to the state-of-the-art lithium ion cells.195,196 Figure
7 shows the LiPF6/EC/DMC system as an example
of surface plots based on the close fit of experimental
data to a fourth degree trivariant polynomial func-
tion:
where ó, m, x, and T are ion conductivity in mS cm-1,
salt concentration in mol kg-1, the mole fraction of
EC, and temperature in °C, respectively. Figure 7
summarizes the changes in ion conductivity with
these variables and exhibits a general trend which
has been observed repeatedly for various electrolyte
systems. This trend allows the tailoring of salt
concentration and solvent composition to maximize
ion conductivity at a given temperature for practical
interests:
(1) Salt Concentration (m). At low salt concen-
trations (<1.0 m), the number of free ions increases
with salt concentration; consequently, ion conductiv-
ity also increases until it peaks at a higher concen-
tration. After this conductivity maximum, any in-
crease in salt concentration results in higher ion
aggregation and higher viscosity of the solution,
which reduces both the free-ion number and the ionic
mobility simultaneously. The location of this maxi-
mum conductivity on the salt concentration axes
(mmax) is decided by the dielectric constant of the
solvents as well as temperature. Generally speaking,
a higher dielectric constant would shift the occur-
rence of ion pairing to higher salt concentrations,
while a higher temperature reduces the solution
viscosity. The common result of both scenarios is the
shift of mmax to higher salt concentrations.
(2) Solvent Composition (xEC). At a given tem-
perature, the solvent composition determines the
outcome of the interplay between dielectric constant
and viscosity; hence, a similar relation between ó and
xEC as shown for the PC/DME system in Figure 6a
should be expected, as is indeed the case. However,
temperature and salt concentration have such a
pronounced effect on this dependence of conductivity
on solvent composition that sometimes this relation
will appear as monotonic in the given range of solvent
compositions.
For example, at a given salt concentration of 1.6
m, solvents with a higher xEC are favored at high
temperatures (>50 °C) because the influence of
viscosity is less pronounced and ó increases mono-
tonically with xEC. At low temperatures (<10 °C), this
relation is reversed because of the predominate role
of viscosity. At intermediate temperatures between
20 and 40 °C, ó peaks versus xEC, indicating that at
neither high nor low xEC is the compromise between
 and Ł able to optimize ion conduction. Similarly,
salt concentration also affects the dependence of
conductivity on solvent composition and produces the
various shapes in ó-xEC relations shown in Figure
7, including single maximum curves and monotonic
increases or decreases at different salt concentrations
and temperatures.
(3) Temperature (T). With other variables being
the same, ion conductivity increases with tempera-
ture monotonically until at very high temperatures
the dielectric constant outweighs the viscosity in
affecting ion conduction. Such high temperatures,
ó ) f(m,x,T) (8)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4321

in the outlaying layer of the SEI, as indicated by its
signature binding energy of 289.0 eV at the C 1s
region, its abundance rapidly decreases with sput-
tering time. On the other hand, O 1s spectra clearly
reveal the increasing abundance of Li2O species. LiF
exists throughout the SEI and is relatively indepen-
dent of sputtering, a result of the sensitivity of BF4-
anion toward trace moisture in electrolytes. Kana-
mura et al. proposed two possible paths for the
formation of LiF: (1) simple acid-base reaction
between HF and alkyl carbonate or Li2CO3, or (2)
direct reduction of BF4- anion by lithium.222,223 Ad-
ditional reactions between the solvents and the
lithium electrode also seem possible after the initial
formation of the SEI, since the abundance of organic
species increases in the inner layer with storage time
increases, according to C 1s spectra. These organic
species might be some polymeric products that re-
sulted from PC or other carbonates instead of alkyl
carbonates alone, as evidenced by the C 1s signal
around 286 eV and by an earlier XPS work.224 This
latter process is attributed to the permeation of
solvent through the SEI and its subsequent reaction
with lithium. The resulting polymeric films, most
likely polyether moieties, are embedded with LiF
crystals. Figure 9 schematically shows the lithium
surface structure and these subsequent reactions of
the SEI.222
Aside from voltammetric techniques, ac impedance
is also a powerful tool widely used to study the
interfacial properties of lithium in nonaqueous elec-
trolytes. It is one of the few in situ techniques and
therefore is often used in combination with voltam-
metry, known as electrochemical impedance spec-
troscopy (EIS). As an example, Figure 10 shows the
impedance response of a symmetrical cell, lithiumj1.0
M LiX in PC/ECjlithium, drawn in the Nyquist plot,
where LiX is LiPF6 or LiClO4. Typically, two semi-
circles would be observed for such cells at high and
medium frequencies, if the time constants for each
component are sufficiently separated, along with a
spike at the low-frequency end.175 It is generally
accepted that the semicircle at medium frequency
corresponds to the ionic migration process in the SEI
and the one at lower frequency to the charge-transfer
process on lithium, whereas the intercept at the high
frequency end with the real axis represents bulk
electrolyte resistance. Examination of the interfacial
resistance in various electrolyte solutions reveals that
the SEI on lithium grows with time of exposure to
electrolytes, and the chemical nature of both solvent
and salt anion seems to relate closely to the semi-
circle for the interfacial film.86,225-227 For example, the
resistance of the SEI that formed in the LiPF6-based
electrolyte is smaller than the one formed in the
LiClO4-based electrolyte, and the presence of EC also
renders a more conductive SEI on lithium.86,225-227
An empirical rule (with frequent exceptions though)
might be stated here concerning the resistance of the
Figure 9. Schematic illustrations of the surface film
formed on lithium in nonaqueous electrolytes based on
LiBF4 solutions and the subsequent reactions. (Reproduced
with permission from ref 222 (Figure 12). Copyright 1995
The Electrochemical Society.)
Figure 10. Impedance complex plane (Nyquist plots) of
lithium electrode in (A) 1.0 M LiPF6/EC/PC and (B) 1.0 M
LiClO4/EC/PC at initial time (0.0 h) and after 24 h. Re and
Im stand for the real and imaginary parts of the impedance
measured, respectively. Frequency was indicated in the
figure for selected data points. Note that the first semi-
circle corresponds to SEI impedance. (Reproduced with
permission from ref 86 (Figure 2). Copyright 1992 The
Electrochemical Society.)
4328 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

SEI: an electrolyte with higher bulk ion conductivity
usually results in an SEI of lower impedance, either
on a lithium or carbonaceous electrode, as will be
discuss in later sections.
As a mass sensor with nanogram sensitivity, a
quartz crystal microbalance (QCM) was used in
combination with voltammetry by Naoi et al. to
monitor the change occurring on the lithium surface
during SEI formation.228 Like EIS, it is one of the
few analytical tools that can reveal in situ informa-
tion on the interfacial process, which includes the
change in mass as well as the surface morphology of
the lithium electrode. It was found that, during the
cycling process, the already-formed SEI was repeat-
edly destroyed and rebuilt, as indicated by the
consistent mass increase with cycle numbers. Among
the various combinations of salts and solvents tested,
the LiPF6-based electrolyte seems to have the most
rapid reaction kinetics with lithium, since the lithium
electrode in it was observed to gain net mass even
during the stripping step, suggesting that the reac-
tion kinetics between the fresh lithium surface and
the electrolyte is fast enough to compensate for the
mass loss caused by the lithium dissolution. On the
other hand, the net mass accumulated on lithium is
much higher in LiClO4- and LiTf-based electrolytes
than in the LiPF6-based electrolyte. The conclusion
from the above two observations seems to point to a
more efficient and protective SEI formed by the
LiPF6-based electrolyte. The author ascribed this
result to the trace amount of LiF in the SEI, which
is absent in LiClO4- and LiTf-based electrolytes.118
The measurement of lithium surface roughness also
reveals LiPF6 as the favored salt in electrolytes
because it forms a smoother and more uniform SEI,
thus minimizing the probability of dendrite growth
on a relative scale as compared with the cases of the
other salts studied.
The significance of an SEI on lithium stability
should be evaluated from two different angles: one
is the static stability that relates to standing storage,
and the other is the dynamic stability that relates to
reversibility. It is the SEI formation on the lithium
surface that leaves lithium statically stable in a
nonaqueous electrolyte; conversely, the SEI also
renders a nonuniform surface morphology for depos-
ited lithium, so that the current density across the
surface is unevenly distributed during the lithium
stripping/deposition, with the direct consequence
being dendrite growth.
The roughness of the SEI depends heavily on the
chemical nature of the electrolyte. For example, it
was argued that an SEI consisting of LiF/Li2O would
provide a much more uniform current distribution,223
while in numerous earlier works it was also observed
that trace moisture has a positive effect on the
lithium cycling efficiency in nonaqueous electrolytes
by assisting in forming a compact and uniform
SEI.57,229 Nevertheless, the dendrite issue is a major
challenge to lithium metal-based chemistry that still
remains unresolved. The prospects for this battery
technology, still attractive because of its high energy
density as compared with the state-of-the-art lithium
ion technology, rely on the discovery of a new
electrolyte system that can suppress or even elimi-
nate lithium dendrite formation.
6.2. Electrolyte/Carbonaceous Anode Interface:
SEI
6.2.1. Exfoliation and Irreversible Capacities on a
Carbonaceous Anode
It has been known since the mid 1950s that
graphite can form intercalation compounds with
lithium ions, which are accommodated in the
interstitial region between the planar graphene
sheets.230-232,233 The most lithium-enriched intercala-
tion compound of this family has a stoichiometry of
LiC6, and its chemical reactivity is very similar to
that of lithium metal. There have been a number of
different chemical approaches to the preparation of
these compounds, for example, by direct reactions of
graphite with molten lithium at 350 °C,232 with
lithium vapor at >400 °C,234 or with lithium powder
under high pressure,235,236 and so forth.
On the other hand, the electrochemical synthesis
of these lithium graphite intercalation compounds
(Li-GIC) has been proven difficult. In earlier work,
it had been found that the most commonly used
electrolyte solvent, PC, decomposed reductively on
the graphite electrode at a potential of 0.80 V, and
the irreversible process led to the physical disinte-
gration of graphite.237 The occurrence of this irrevers-
ible reduction apparently prohibits any possibility of
the lithium ion intercalating into graphite, which
should happen at a much lower (and therefore more
reductive) potential. The destruction of graphite by
PC was repeatedly observed in different electrolytes
based on PC, and this disintegrating process of the
graphite structure was named “exfoliation”.238-243
Besenhard et al. proposed that the exfoliation was
caused by the cointercalation of PC molecules with
lithium ions into the interplanar structure of the
graphite and the subsequent decomposition there-
in.239-242 As a result, the multilayer structure of
graphite, which is only held together by weak van
de Waals forces, falls apart because of the strain
introduced by the gaseous products, believed to be
mostly propylene.234-237,243
Realizing that the solvent must be the key to the
exfoliation, later researchers explored different polar
organic molecules such as dimethyl sulfoxide (DMSO)
and DME as candidates to replace PC, in the hope
that they would not cointercalate or decompose; but
most of these efforts failed to endorse the usefulness
of Li-GIC as a negative electrode to replace lith-
ium.239-241 In the 1980s, the only successful example
of electrochemical intercalation of lithium into graph-
ite was reported by Yazami and Touzain in 1983 with
a polymer electrolyte based on poly(ethylene oxide)
(PEO).244 As it is essentially impossible for the
macromolecular solvent PEO to cointercalate, this
electrolyte supported the reversible lithium ion in-
tercalation into and deintercalation from natural
graphite. Using electrochemical titration techniques,
the potential of the stage I and II Li-GIC was
determined to be between 0.50 and 0.20 V vs Li, thus
confirming the conceptual feasibility that Li-GIC
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4329

ers on the cycling graphite failed to offer any evidence
that a substantial change in the interlayer distance
occurred around the cointercalation potential.96,124,252,255
In these experiments, the maximal shift in 2ı of the
(002) diffraction peak during graphite cycling, which
can reflect the size of the guest species, only corre-
sponds to an expansion of 0.35 Å in the c-axis; hence,
bare lithium ions seem to be the only species that
can be accommodated.252-254 It should be pointed out
that the existence of ternary GICs is beyond question,
and the doubt here is about whether they could be
electrochemically formed under a similar situation
in the forming of lithium ion cells.256-260 So far, the
chemical or electrochemical synthesis of ternary GICs
has failed to produce any compositions based on
carbonate solvents, despite the success of Ogumi and
co-workers with various solvents such as ether or
alkylsulfoxide,257-260 casting more doubt on the Be-
senhard model.
In defense of the Besenhard model, Chung et al.
argued that the lack of XRD evidence for ternary
GICs might be simply due to either their rapid
decomposition or the localized presence of them near
the graphite edges.255 Since XRD probing of the
material is based on the averaged diffraction re-
sponse of the sample lattice, these wider spacings of
the ternary GICs might not be detected as an
averaged bulk property.
In situ EQCM studies of graphite in various
electrolytes also challenge the formation of ternary
GICs with the real-time monitoring of the graphite
electrode mass increase during cathodic polariza-
tion.101 It was found that, between 0.8 and 0.5 V,
where such GICs are supposed to be stable, the mass
change per quantity of electricity (¢m/¢Q) was 27-
35 g F-1, corresponding well to Li2CO3 that has a ¢m/
¢Q of 36.9 g F-1. If solvents such as EC (¢m/¢Q )
88.07 g F-1), PC (¢m/¢Q ) 102.1 g F-1), or the
solvated lithium ion [Li+(PC)n] (¢m/¢Q > 300 g F-1
assuming a coordination number of 3) cointercalate
into the graphene structure, the corresponding mass
gain on the graphite anode, which is too conspicuous
to miss, should have been well recorded by the quartz
crystal sensor.
The thermodynamic stability of a ternary GIC is
also questionable. Obviously, between a bare lithium
ion and one solvated by molecular dipoles, the
intercalation of the former between two giant
graphene anions is far more favored thermodynami-
cally than that of the latter. The fully lithiated GIC
LiC6, for example, does not solvate in nonaqueous
electrolyte solvents, and the tendency of lithium to
prefer binary (i.e., without solvent cointercalation)
instead of ternary GICs has also been noticed in the
solution syntheses.257
Despite the concerns raised by XRD, EQCM, and
thermodynamics, the Besenhard model still received
extensive support from various experimental obser-
vations as summarized below and soon became the
prevalent model used by researchers in the lithium
ion battery community.
It had been discovered earlier that when electrolyte
solvents decompose reductively on graphite, one of
the products is gaseous propylene.237 Dey et al.
proposed a surface mechanism involving a two-
electron process as shown in Scheme 8. Arakawa and
Yamaki quantitatively analyzed the gas volume
generated during the electrochemical decomposition
of PC on a graphite electrode and found a mismatch
between the Coulombic quantity and the equivalents
of propylene gas generated, with an efficiency be-
tween 50% and 70%, depending on current density.243
Apparently this result conflicts with Scheme 8, and
other reaction processes must also exist simulta-
neously. Using a kinetic treatment, they suggested
a mechanism (see Scheme 9) where a ternary GIC is
the intermediate, which underwent two parallel but
competitive paths to form either the gaseous product
propylene and Li2CO3 or lithiated binary GICs. Using
this mechanism, Arakawa and Yamaki successfully
explained the relation between gas volume rate and
time.243
Following a similar approach, Shu et al. used an
EC/PC mixture instead of neat PC as electrolyte
solvent, and their analysis of propylene gas volume
corroborates the observations of Arakawa and Ya-
maki.261 Furthermore, because EC was present in
their electrolyte, the reversible lithium intercalation
could occur after a long plateau at 0.8 V (represent-
ing PC decomposition), therefore a correlation be-
tween the gas volume and this irreversible process
was able to be established, as shown in Figure 13.
Considering Aurbach’s spectroscopic observations (to
be discussed later), a modified mechanism (see
Scheme 10) was proposed by Shu et al., wherein a
competition exists between the surface reaction lead-
ing to radical anions and the formation of ternary
Figure 13. Correlation of gas evolution on a graphite
electrode in 1.0 M LiClO4/PC/EC (50:50) with the irrevers-
ible process at 0.80 V during the first discharge. Note the
level off of gas volume as soon as reversible lithium ion
intercalation starts. (Reproduced with permission from ref
261 (Figure 2). Copyright 1993 The Electrochemical Soci-
ety).
Scheme 9. Electrochemical Reduction of PC on
Graphite: Ternary GIC Mechanism
Scheme 10. Electrochemical Reduction of PC on
Graphite: Modified Ternary GIC Mechanism
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4333

GICs, via the one-electron process in both cases.
According to Shu et al., these intermediate species
underwent further single-electron reduction and
produce Li2CO3 and propylene gas, while alkyl car-
bonates are generated via radical termination as
shown in Schemes 1 and 755,214 to become the major
ingredients in the surface film.
In separate research, Matsumura et al. performed
quantitative analyses of lithium content in graphite
electrodes with plasma spectroscopy and correlated
the results with the quantities of electricity associ-
ated with the reversible and irreversible processes.262
With XPS, they found that after the graphite is
delithiated, there is a certain amount of lithium
remaining in the graphite that cannot be removed
electrochemically. Depth profiles established by sput-
tering the graphite sample with O2+ confirmed that
these lithium species are distributed rather evenly
in the bulk of the graphite. The author ascribed the
existence of these electrochemically nonremovable
lithium ions as the result of lithium reaction with
active sites on the carbon surface.262 However, there
is also the possibility that these lithium signals are
from the lithium-containing ingredients in the three-
dimensional SEI that exists in the graphite matrix.
Kim and Park investigated the mechanism of
lithium ion intercalation in graphite anodes employ-
ing solid-state NMR.263 Their results perhaps offer
the most direct evidence in support of Besenhard’s
GIC model. By adding strongly coordinating additives
for lithium ions, 12-crown-4 and 18-crown-6 ethers,
into the electrolyte solution, they were able to observe
an obvious Knight shift in the 7Li signals of the
graphite powder that was caused by the coordination.
Separate 13C NMR tests conducted on the same
graphite sample also identified the signals of crown
ether as well as carbonate (more likely its decomposi-
tion product) in the graphite powder following lithia-
tion. Assuming that the adsorbed additives and
solvents on the graphite surface have all been
thoroughly removed during the washing procedure
that preceded NMR measurements, the above obser-
vation should be considered as the first confirmation
that solvent molecules indeed are found in the bulk
of the graphite, and their cointercalation with lithium
ions during the lithiation process would most likely
be the path. However, since the author did not
present any blank test to prove the effectiveness of
the washing procedure, the possibility of surface
contamination due to the remnant solvent molecules
being trapped in the porous structure of the graphite
electrode could not be completely excluded. Never-
theless, solid-state NMR proved to be an effective tool
in studying the bulk structure of graphite anodes,
and more efforts on SEI mechanisms should be done
with this technique.
In situ Raman spectra studies performed on graph-
ite anodes also seem to reveal a cointercalation
occurrence that leads to exfoliation. Huang and Frech
used solutions of LiClO4 in EC/EMC and EC/DME
as electrolytes and monitored the E2g2 band at 1580
cm-1 in the Raman spectra of the graphite that was
cycled between 2.0 and 0.07 V.264 Reversible lithium
intercalation and deintercalation was indicated by
the corresponding shift of this band in the EC/DMC-
based electrolyte. But in the presence of DME, the
graphite surface structure was detected to be ir-
reversibly altered in the range between 0.9 and 0.5
V, as indicated by a shoulder on the E2g2 band. Since
no lithium ion intercalation is supposed to occur in
this potential range, the authors attributed the
Raman spectral changes to the extensive DME coint-
ercalation. Interestingly enough, a DME-based ter-
nary GIC was indeed electrochemically obtained and
identified by Abe and Ogumi and co-workers with
XRD.257-260 As a matter of fact, the results from this
Raman study support the Besenhard model but also
cast doubt upon it simultaneously because no such
irreversible E2g2 band shift had been observed in the
EC/DMC electrolytes, although obviously the SEI
was formed in that case too.
Various microscopic means were also applied to
study the SEI formation process, but the reproduc-
ibility of the results is highly dependent on the
condition under which the observations were made
and the pretreatment history of the samples. Even
for the same observation, the interpretations could
vary from author to author. For example, with a
scanning tunnel microscope (STM), Inaba et al.
observed the formation of some “blisters” on the
graphite surface during its cathodic polarization and
described them as the swelling of the graphene layer
due to solvent cointercalation;265 however, Farrington
and co-workers, after observing the same phenom-
enon with an atomic force microscope (AFM), ascribed
these island structures to the depositions of the
decomposition products from the solvent.266 One
common phenomenon that was observed by all of
these microscopic experiments is the stepwise forma-
tion of the surface species,265-269 which appear first
near the edge sites of the highly ordered graphite
surface at potentials as high as 1.6 V and then grow
and cover the whole electrode at potentials below 0.80
V, as shown in Figure 14. Since the intrinsic reduc-
tion potentials of the related solvents are much lower
Figure 14. Schematic diagram summarizing the stepwise
formation of the SEI on a graphite surface. (Reproduced
with permission from ref 266 (Figure 10). Copyright 1997
The Electrochemical Society.)
4334 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

have similar stereo hindrances and would exert
similar strains into the graphene structure and cause
exfoliation, while t-BC may experience severe geo-
metric constraints for cointercalation. Hence, it can
exhibit a reaction behavior on a graphite surface that
is qualitatively different from that of PC.
Again, caution must be taken when using the
graphic representation as shown in Figure 17 be-
cause such oversimplification might be misleading
when a complex process such as the formation of an
SEI on graphite is handled. For example, the graphic
representation in Figure 17 suggests that cointerca-
lation of solvent occurs without the assistance of
lithium ions, while, in actual cointercalation, the
solvents that coordinated with lithium ions in the
solvation sheath would preferentially cointercalate.
With the supermolecular structure of the solvation
sheath in consideration, it would be more difficult to
predict the effect of diastereomers on cointercalation
by the analogue of Figure 17.
On the other hand, with an average solvation
number of four, it would be hard to imagine that
cointercalation would occur without breaking the
solvation structure of the lithium ion, considering the
required expansion in the graphite structure to
accommodate such a gigantic guest. Most probably,
at the edge sites of the graphite where cointercalation
occurs, the solvated lithium ion is progressively
stripped of its solvation sheath, and the free solvents
would then insert into the graphite interior in a close
way, as shown by Figure 17.
To summarize, various models have been proposed
to depict the formation of an SEI on a graphite anode,
based on the common knowledge that the reductive
decomposition of electrolyte components leads to the
formation of a protective film on the anode. However,
these models differ in the mechanism by which the
SEI is formed, especially concerning the issue of
whether a ternary GIC is formed before the reductive
decomposition occurs. Although each of these models
can elegantly account for certain experimental ob-
servations, the Besenhard model that evolves around
solvent cointercalation seems to be supported by the
most experimental evidences, despite the fact that
the electrochemical formation of a key species of this
model, a metastable ternary GIC intermediate, has
not been experimentally confirmed. Nevertheless, it
is generally agreed nowadays that to a certain extent
the solvent cointercalation does occur and is at least
a part of the process related to the formation of the
SEI. The complete clarification of the above contro-
versy relies on obtaining more experimental evidence
at a microscopic level from further studies.
6.2.3. Characterization of Surface Chemistry
Relative to the controversy associated with the
mechanism of SEI formation, there is less uncer-
tainty in the knowledge about the chemical composi-
tion of the SEI, due mainly to the exhaustive surface
spectroscopic studies carried out by Aurbach and co-
workers on carbonaceous anodes in various nonaque-
ous electrolytes, adopting both in situ and ex situ
approaches.104,108,123,124,249,250 Table 6 lists the chemical
compounds as identified by these spectroscopic means
and the proposed chemical reactions leading to those
species.27-284 As it has been pointed out, the solvents,
especially the cyclic carbonates, play a more impor-
tant role in the surface chemistry of the anode than
the salt anions.178
Compared with the surface chemistry of nonactive
electrodes212-214 or lithium electrode,55,117,118,209 simi-
lar chemical species were identified despite the
differences in the electrode surfaces. A major modi-
fication of the previously accepted two-electron re-
ductive pathway as suggested by Dey and Sullivan237
was proposed by Aurbach and co-workers based on
the identification of lithium alkyl carbonate by FT-
IR.108,124,249 They suggested that the surface reductive
process for most carbonate molecules proceeds via a
single-electron path leading to the intermediate, as
shown in Schemes 1 and 7, and that Li2CO3 and
alkylenes were formed through either the continued
reduction of this intermediate or the secondary
reaction between it and trace moisture in the system.
Specifically, the following structures were assigned
to the decomposition products from EC and PC,
which supposedly constitute the main composition of
the SEI layer:102,104,117
The predecessor of alkyl carbonate, a radical anion,
has been experimentally observed by ESR, the life
span of which depends on the carbonate structure
and ranges from minutes to hours but is independent
Figure 17. Schematic drawing of the GIC-exfoliation
model. Differentiation of the stereo difference among EC,
PC, and related carbonates by graphene structure.
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4337

of the salt species used.178 Moreover, this one-electron
reduction mechanism does seem to be strongly fa-
vored by the fact that the intermediate does not
convert to alkylenes in 100% yield, reminiscent of the
work by and Arakawa and Yamaki243 and Shu et
al.261 who have already reported that the generation
of gaseous products does not match the electric
quantity injected into the anode.
A more quantitative confirmation of alkyl carbon-
ate formation came from transmission electron mi-
croscopy (TEM), by which electron diffraction (ED),
electron energy loss spectra (EELS), and imaging
were conducted on disordered and graphitic carbon-
aceous anodes cycled in LiClO4/EC, respectively.280
Although the ED does not identify any crystalline
phase other than the hexagonal structured graphite
and Li2CO3 on the surface (probably suggesting that
alkyl carbonates are noncrystalline), the atomic
concentration ratio of oxygen and carbon (O/C) as
determined by EELS ranges between 1 and 1.5,
unequivocally indicating that the proposed alkyl
carbonate formed from EC reduction. Interestingly,
this O/C ratio was also found to vary with the
potential to which the carbonaceous anodes were
cathodically polarized: above 0.90 V an O/C ratio )
3.0 was obtained for the presence of Li2CO3, while
below 0.80 V an O/C ratio ) 1.0-1.5 was obtained.
Thus, the authors proposed that the SEI was formed
in a two-step process: (1) formation of Li2CO3 occurs
at potentials between 1.0 and 0.80 V, and (2) forma-
tion of alkyl carbonates is favored at lower potentials
below 0.80 V.
In retrospect, probably a more reasonable explana-
tion could hereby be proposed concerning this poten-
tial-dependent reductive decomposition. Combining
the observations from Shu et al. about the correlation
of gas products with a 0.80 V plateau,261 Farrington
and co-workers about the deposition of the film onto
the surface,266 Aurbach and co-workers about the
competition between single- and two-electron reduc-
tive paths,108,124 and Kanamura et al. about the
multilayered structure of the SEI,222,223 one can
conclude that, at potentials above 0.80 V, where the
film has not completely covered the basal surface of
graphite, the two-electron process as proposed in
Scheme 8 is likely the predominant process because
of the good electronic conductivity of the graphite
surface, leading to Li2CO3 and ethylene; below 0.80
V the single-electron process prevails because of the
much slower electron hopping kinetics, leaving the
surface with alkyl carbonate depositions and render-
ing negligible gas evolution simultaneously. There-
fore, the potential-dependence of the decomposition
compounds as observed in a TEM study by Naji et
al.280 is actually the potential-dependence of the
competition between one- and two-electron paths, as
shown in Schemes 1 and 7.
In addition to FT-IR, XPS experiments performed
on a graphite anode that had been cycled in various
carbonate-based electrolytes also identified an alkyl
carbonate species. Bar-Tow et al. characterized the
surface of a highly oriented pyrolytic graphite (HOPG)
that had been cycled in LiAsF6/EC/DEC and found
the C 1s signal located at 289 eV,281 which had been
previously observed on a lithium surface and identi-
fied as alkyl carbonates by Kanamura et al.222,223,282,283
after referencing with the C 1s signal of Li2CO3 at
290.5 eV. Surface sputtering with Ar+ reduces the
abundance of this species rapidly, suggesting that
this species might only be stable on the top layer of
the SEI.
Besides lithium alkyl carbonates, XPS also identi-
fied a wide variety of decomposition products from
other electrolyte components, including polyether
moieties as well as the lower valence As species. The
depth profile of the SEI established by prolonged
sputtering by Ar+ reveals the multilayered structure
of the SEI, as shown in Figure 18a, in which the
organic species such as alkyl carbonate and Li2CO3
are present in predominant percentages on the solu-
tion side of the SEI while simple inorganic species
such as Li2O or As(0) are more stable on the graphite
side of the SEI probably because of the more complete
reduction facilitated by faster electron-tunneling
kinetics.
Differences in chemical composition were also
observed in the SEIs formed on basal planes and edge
sites. The former were more enriched with organic
species and the latter with inorganic species, espe-
cially with the decomposition products that obviously
originated from salt anions. The authors thus con-
cluded that, on the basal plane, the major contribu-
tion to SEI formation is from solvent reductive
Figure 18. Depth profile of various chemical elements in
the SEI formed on HOPG: (a) basal plane and (b) edge
section. (Reproduced with permission from ref 281 (Figure
4). Copyright 1999 The Electrochemical Society.)
4338 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

in the MS.286 No sign of CO or CO2 had been found,
contradicting Schemes 1, 7, and 8. The production of
H2 was ascribed to the reduction of trace moisture
contamination; thus, the SEI formation process
seemed to start below 0.80 V. However, since the
MS can only detect gaseous species, another surface
Table 6. Reductive Decompositions on Carbonaceous Anodes
A. Electrolyte Solvents
B. Electrolyte Salts and Other Components
4340 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

probable reason might be the stability of these
cathode hosts against solvent cointercalation and
exfoliation due to the layered structure being held
together by Coulombic interactions between op-
positely charged slabs composed of metal cations and
oxide anions.288 Nevertheless, irreversible decomposi-
tions, although oxidative in nature, do occur on
cathode surfaces, and the decomposition products
very likely form a passivation film that prevents any
sustained solvent decomposition. In this sense, the
interface between electrolyte and cathode should
possess the same physicochemical fundaments of the
SEI on anodes, that is, electronic insulator and
lithium ion conductor. A number of authors have
used the term “SEI” as well to describe the passiva-
tion of cathode surfaces in lithium/lithium ion cells;
however, by the currently accepted convention the
term is usually reserved for carbonaceous anodes.
Instead, a more general reference of “passivation
film” or “surface layer” has been used for electrolyte/
cathode interfaces.
6.3.1. Passivation Film on a Cathode
The detection of protective films on cathode sur-
faces has not been as straightforward as that of the
SEI on anodes, partially because a native surface
film, mainly composed of Li2CO3, already exists on
all transition metal oxide cathode materials based on
manganese, cobalt, and nickel.289-294 This surface
component could arise from the precursors used to
synthesize these metal oxides or, more likely, from
the reaction between the metal oxides and the CO2
in the atmosphere during the processing of these
strongly oxidizing materials.290 Upon contact with
electrolytes, this native film is usually eroded by the
acidic electrolyte salts currently employed by the
lithium ion industry, and to make things more
complicated, the active materials are usually involved
in the subsequent oxidation of the electrolyte solvents
on the exposed cathode surface. Proving the existence
of a surface film on the cathodes has been difficult,
and spectroscopic identification, which has been
proven to be an effective tool in studying the SEI on
anodes, often yields ambiguous results.
Goodenough and co-workers were perhaps the
earliest authors to suggest that a film exists on the
cathode/electrolyte interface.295 In an attempt to
simulate the ac impedance responses of an intercala-
tion-type cathode in liquid electrolytes, they discov-
ered that Li1-xCoO2 in PC cannot be described as a
simple intracompound lithium ion diffusion; instead,
a more complex electrochemical process, including
the formation of a surface layer on the electrode due
to the oxidation of the electrolyte, must be considered.
They proposed an equivalent circuit, shown in Figure
20, for the lithium intercalation process into such
cathodes, wherein Re, Rsl and Csl, Rct and Cct, and Zw
represent the bulk resistance of electrolytes, the
resistance and capacitance of the surface layer, the
resistance and capacitance of the charge-transfer
process, and the Warburg impedance, respectively.
In such a model, lithium migration has to go through
the surface layer because of its serial nature in order
for the charge-transfer and further diffusion within
the solid to occur. This equivalent circuit and its
numerous variations have become a universally
adopted model that simulates the behavior of both
anode and cathode in nonaqueous electrolytes.
In formulating new electrolyte compositions that
can withstand the high potentials of the cathode
materials, Guyomard and Tarascon also realized that
oxidative decomposition of electrolyte components
occurred on cathode surfaces, and passivation of the
surface prevented the bulk electrolytes from further
decompositions.93,98 Using LiPF6/EC/DMC and the
manganese spinel LiMn2O4, they systematically ex-
plored the origin of the oxidative decomposition by
quantitatively analyzing the irreversible capacity
associated with electrolyte oxidation and established
a correlation between it and the surface area of the
composite cathode. Thus, they concluded that the
oxidative decomposition of electrolytes on cathodes
is surface-catalyzed, and very likely the cathode
surface would be covered evenly with a surface layer
that is composed of the decomposition products.
Almost during the entire 1990s, the main interest
of the lithium ion research community was focused
on electrolyte/anode interfaces while its cathode
counterpart was overlooked until various lithium ion
systems, especially those based on manganese spinel
cathodes, were found to suffer power loss and capac-
ity fade upon prolonged cycling or storage at elevated
temperatures.296-302 Preliminary diagnostic studies
pointed to electrolyte/cathode interfaces as the source
of the degradation, and intensive research has been
carried out to address this issue since the late 1990s.
Aurbach et al. studied the interfacial behavior of
various cathode materials in LiAsF6/EC/DMC using
EIS and found that, for all of these cathodes, that is,
LiNiO2, LiCoO2, and LixMn2O4, the impedance spec-
tra obtained reflected several processes in series.290
In other words, the overall lithium ion intercalation
in and deintercalation from a variety of LiMO2 bulk
materials included the inevitable step of lithium ion
migration through a certain surface layer in a man-
ner very similar to that of the reversible lithium ion
Figure 20. Equivalent circuit based on surface layer
formation on cathode materials (a, top) and the electrolyte/
cathode interface (b, bottom). (Reconstructed based on ref
295.)
4342 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

with the results of Zhang et al.303,304 and Croce et
al.,305 where the change of surface-layer resistance
with potential is obvious.
Some authors pointed out that the surface layer
on the metal oxide cathodes might not be as stable
as that on the carbonaceous anode, and the cell power
loss and capacity fade are mainly caused by degrada-
tion of this electrolyte/cathode interface.293 Their
argument was supported by the increase in imped-
ance with cycling, which very likely suggests a
continued growth of surface layers on the cathode.292
There was also one report revealing the complete
absence of surface layers on cathodes after prolonged
cycling, probably indicating that the electrolyte com-
ponents or impurities react with the surface layer
and leave the cathode surface with the IR-insensitive
LiF.298
6.3.2. Characterization of Surface Chemistry
The overwhelming majority of the studies on the
oxidative decompositions of solvents and salt
anions were carried out on nonactive elec-
trodes.74,81,120,130,131,204,206,207,306-308 On the basis of the
detection of ring-opening reactions by in situ FT-IR,
a mechanism involving a single-electron process
producing a radical cation was proposed for carbonate
solvents (Scheme 14).307 Subsequent decomposition
of the intermediates leads to gaseous as well as solid
products, which form a solid film on the electrode.
The existence of such radical species during the
oxidative decompositions of electrolyte components
has been confirmed for a wide variety of electrolytes
by ESR, and the surface of a charged cathode (LiCoO2
at 4.3 V) is identified as the source of its generation,
because a parallel blank test with electrolytes in the
absence of a cathode produces no radical species.309
The authors depict a picture of delocalized radical
cation structures that are coordinated by the neigh-
boring solvent molecules, whose half-life is on the
order of minutes depending on the electrolytes. Thus,
it is reasonable to believe that Scheme 14 represents
the initial oxidative cleavage of the carbonate sol-
vents.
The systematic surface characterization was con-
ducted by Aurbach and co-workers using FT-IR
spectroscopy for LiCoO2, LiNiO2, and LiMn2O4 spinel
cathodes.290 After cycling in electrolyte LiAsF6/EC/
DMC between 3.0 and 4.4 V, all of the cathode
surfaces were found to be covered with a wealth of
new chemical species, while the signals correspond-
ing to the native Li2CO3 diminished. Preliminary
interpretation of these spectra led the authors to
conclude that the species very much resemble the
lithium alkyl carbonates observed on various anode
surfaces that have been formed, as supported by the
signals around 1650 cm-1. With further study, they
confirmed that the electrolyte-related species were
formed on the cathode surface upon storage or cycling
but emphasized that, because of the complicated
nature of the surface oxidation processes, it is impos-
sible to obtain an unambiguous picture of the surface
chemistry.292 Since the native Li2CO3 does not change
when in contact with pure (salt-free) solvents, the salt
anions obviously played a crucial role in forming
these new surface species. Cycling seems to facilitate
the surface chemistry, as Figure 21 shows, since a
variety of absorptions corresponding to CsH, CdO,
and CsO bonds were identified. The carbonyl func-
tionality around 1800-1700 cm-1 was believed to be
polycarbonate species, a possible source of which
could be the oxidation of EC or the ring-opening
polymerization of EC catalyzed by a nucleophilic
mechanism (Scheme 15). The nucleophilic initiator
RO-, as evidenced by the absorptions near 1100 cm-1,
could be generated on the surface of the cathode
because of the possible reactions between the active
mass and the solvents (Scheme 16). Meanwhile, the
signals around 1650 cm-1 (asymmetric CdO), 1350-
1300 cm-1 (symmetric CdO), 1100 cm-1 (CsO), and
850 cm-1 (OCO2-) still suggest species with similar
structure to alkyl carbonates. Since the mechanism
Scheme 14. Possible Electrochemical Oxidation
Path for PC
Figure 21. FTIR spectra (diffuse reflectance mode) mea-
sured from a pristine LiNiO2 composite cathode and with
a cathode after galvanostatic cycling in Li salt/EC/DMC
solutions. (Reproduced with permission from ref 292 (Fig-
ure 1). Copyright 2000 The Electrochemical Society.)
Scheme 15. Possible Mechanism for the
Formation of Polycarbonates
Scheme 16. Surface Nucleophilic Reactions
between LiNiO2 and Electrolyte Solvents
4344 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

misleading, and the authentic anodic stability limits
of any electrolyte component in an actual lithium ion
device can only be evaluated on the surface of the
electrode that is used in the device.
Unfortunately, this approach for electrochemical
stability determination has not been widely adopted.
The few exceptions include the seminal electrolyte
work by Guyomard and Tarascon.98,99 In the formula-
tion of new electrolytes for lithium ion technology,
the spinel composite electrode was used as the
standard working surface in all of the voltammetric
measurements. The oxidative decomposition limits
of the new electrolytes thus determined are sum-
marized in Table 7 along with a handful of stability
data that were determined in a similar approach for
other electrolyte systems.
Figure 24 shows the result of the voltammetry for
EC/DMC- and EC/DEE-based electrolytes. The typi-
cal two-step process between 3.9 and 4.3 V corre-
sponds to the delithiation of the spinel structure,
which is reversible. The oxidative decomposition of
DEE occurs below 4.4 V almost immediately after the
delithiation is completed. This proximity to the
potential of the cathode electrochemistry leaves little
tolerance for overcharge. On the other hand, the EC/
DMC-based electrolyte only shows negligible back-
ground current before 5.1 V. After 4.3 V, the spinel
lattice has actually turned into ì-MnO2. On the
surface of the ì-MnO2, the low current of the EC/
DMC exhibits the extreme stability of this mixed
carbonate formulation. As already pointed out in
earlier sections, the replacement of ether compounds
by linear carbonate results in a significant expansion
of the oxidative potential limits by 0.80 V.
The anodic limit for the electrochemical stability
of these carbonate mixtures has been determined to
be around 5.5 V in numerous studies.314-317 Thus,
new electrolyte formulations are needed for any
applications requiring >5.0 V potentials. For most
of the state-of-the-art cathode materials based on the
oxides of Ni, Mn, and Co, however, these carbonate
mixtures can provide a sufficiently wide electrochemi-
cal stability window such that the reversible lithium
ion chemistry with an upper potential limit of 4.30
V is practical.
When the poor anodic stability of DMC or EMC
alone on a similar cathode surface is considered,76 the
role of EC in stabilizing the solvent system becomes
obvious. A conclusion that could be extracted from
these studies is that the existence of EC not only
renders the electrolyte system with superior cathodic
stability by forming an effective SEI on the carbon-
aceous anode but also acts as a key component in
forming a surface layer on the cathode surface that
is of high breakdown potential. It is for its unique
abilities at both electrodes that EC has become an
indispensable cosolvent for the electrolyte used in
lithium ion cells.
6.3.4. Passivation of Current Collector
Another issue closely related with the anodic
stability of electrolytes is the interaction between
electrolyte components and the commonly used cath-
ode substrate Al in lithium ion cells.
Al and Cu foils are used as current collectors for
cathode and anode materials, respectively, in lithium
Figure 23. Anodic stability of linear dialkyl carbonates
on GC and a composite cathode surface. Electrolytes: 1.0
LiPF6 in DMC or EMC, respectively. (Reproduced with
permission from ref 76 (Figure 2). Copyright 1999 The
Electrochemical Society.)
Table 7. Electrochemical Stability of Electrolyte
Solvents: Active Electrodes
electrolyte
working
electrode Eaa ref
1.0 M LiClO4/PC/DME (1:1) LiV3O8 4.6 307
1.0 M LiAsF6/PC/DME (1:1) LiV3O8 4.7 307
1.0 M LiClO4/EC/DEE (1:1) LiMn2O4 4.55 93
1.0 M LiIm/EC/DEE (1:1) LiMn2O4 4.4 99
1.0 M LiIm/EC/DME (1:1) LiMn2O4 4.35 99
1.0 M LiPF6/EC/DEE (1:1) LiMn2O4 3.8 93
1.0 M LiBF4/EC/DEE (1:1) LiMn2O4 3.4 99
1.0 M LiClO4/PC LiMn2O4 >5.1 99
1.0 M LiBF4/EC/DMC (2:1) LiMn2O4 >5.1 98
1.0 M LiClO4/EC/DMC (2:1) LiMn2O4 >5.1 99
1.0 M LiPF6/EC/DMC (2:1) LiMn2O4 >5.1 44, 98
1.0 M LiClO4/EMS LiMn2O4 5.8 75, 314
1.0 M LiIm/EMS LiMn2O4 5.8 75, 314
1.0 M LiPF6/EMS LiMn2O4 5.8 314
1.0 M LiPF6/EiBS LiMn2O4 5.8 314
1.0 M LiPF6/DMC LiMn2O4 4.0 314
1.0 M LiPF6/EMC LiMn2O4 4.5 314
1.0 M LiPF6/FPMS/DMC LiMn2O4 5.55 314
1.0 M LiPF6/FPMS/EMC LiMn2O4 5.55 314
1.0 M LiPF6/EC/PC/DMC (1:1:2) LiNiVO4 >4.9 314
a Anodic limits, potential referred to Li+/Li.
Figure 24. Voltammogram at 55 °C for the electrolytes
(a) 1.0 M LiClO4/EC/DEE and (b) 1.0 M LiPF6/EC/DMC
on spinel cathode LiMn2O4. (Reproduced with permission
from ref 98 (Figure 1). Copyright 1993 The Electrochemical
Society.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4347

Microscopic examination of the aged Al retrieved
from well-cycled cells found that, even in electrolytes
based on LiPF6 or LiBF4, localized pitting occurred
on the Al (Figure 26a).298,318,319 Closer examination
of these corrosion sites under an SEM revealed that
what had optically appeared to be a “pit” was actually
a “mound” or “nodule” (Figure 26b) where both Al(0)
and Al2O3 existed, according to XPS analysis.321 Due
to the anodic state of the Al during the electric
cyclings in these cells, the presence of Al(0) practi-
cally suggests that the mounds were somehow elec-
trically isolated from the foil. The authors presented
two possible explanations for this situation: (1)
corrosion and its associated reaction products under-
mined the surface of a developed pit and caused the
overlying metal to bulge or break away or (2) a
soluble Al(III) corrosion product was electrodeposited
onto the poorly conductive solid corrosion products
during the cathodic process (discharge) so that reduc-
tion actually occurred. The fact that the open circuit
voltage (OCV) of Al lies between 3.2 and 3.6 V makes
the latter reductive step possible in normal battery
cyclings.
On the other hand, when the electrolyte salts were
LiTf or LiIm, serious corrosions in the form of high-
density pits were visible before any extensive cycling
was conducted,141,322 as shown by Figure 26c, where
Al was merely polarized anodically in a LiIm-based
electrolyte for a much smaller time scale than that
of prolonged cycling in an actual rechargeable cell.
The damage to Al in these cases was on a macroscopic
scale.
Impedance analyses of the Al under corrosion were
conducted via EIS. On the basis of the models
previously established for the corrosion of other
metals in both aqueous and nonaqueous electro-
lytes,325,326 the corrosion process was proposed as a
two-step adsorption/oxidation/desorption process
(Scheme 19).147,156
The impedance response with frequency can be
closely simulated by the equivalent circuit shown in
Figure 27a, where Re, Rct, Cdl, Rad, and Cad represent
the resistance or capacitance for the electrolyte
solution, charge-transfer, double layer, and adsorbed
layer, respectively. An interesting correlation was
found between the passivating ability of various
anions and the resistances of the two impedance
components Rct and Rad, which are high for LiPF6-
and LiBF4-based electrolytes and low for LiTf- or
LiIm-based electrolytes.156 Using the rationale pro-
posed by the authors, the former component (Rct) is
Figure 26. Micrographs of a corroded Al surface in various electrolytes. (a, top left) Micrograph in LiPF6/EC/DMC
electrolyte, after 150 electrical cycles. The light areas are mounds or nodules. (Reproduced with permission from ref 321
(Figure 7). Copyright 1999 The Electrochemical Society.) (b, bottom left) SEI image of the cross-sectional view of the mounds
as shown in part a. (Reproduced with permission from ref 321 (Figure 8a). Copyright 1999 The Electrochemical Society.)
(c, right) Micrograph in LiIm/PEO after 1 h of galvanostatic polarization at 100 íA cm-2. (Reproduced with permission
from ref 322 (Figure 14). Copyright 1999 The Electrochemical Society.)
Scheme 19. Al Corrosion in Nonaqueous
Electrolytes
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4349

at a series of potentials in nonaqueous electrolytes
based on LiIm, LiTf, and a new asymmetric imide,
Li(CF3SO2)(C4F9SO2)N, using both XPS and FT-IR.148
These spectra confirmed the presence of organic
species on the Al surface, and by monitoring the IR
signature absorption for the carbonate functionality
at 1710 cm-1, they determined that these species
appeared at 3.6 V in LiIm- and LiTf-based electro-
lytes; however, when using LiPF6, LiBF4, and LiClO4,
this absorption was not found till 5.2, 4.4, and 4.6
V,327 respectively. Hence, they argued that, while Al
was well passivated by the native Li2CO3 film, such
an oxidation by electrochemical means would not
occur. In other words, the difference in the effect of
the electrolyte components on Al stability is not in
how they decompose to form a new protection layers
such as through the familiar mechanism of SEI
formationsbut in how inert they are to the existing
native layer. This inertness depends on the cathode
potential, and the occurrence of solvent oxidation is
indicative of the breakdown of the native passivation
film. Apparently, Tf- and Im- anion species are
corrosive to the native film, so the breakdown poten-
tial of it is much lower in the electrolytes based on
these salts. Therefore, it would be reasonable to
believe that a reaction occurring between Al2O3 and
electrolyte solutes at high potentials is responsible
for the onset of major corrosion, during which the
coordination ability of Im- and Tf- anions toward
Al3+ acts as the driving force that accelerates the
anodic dissolution of Al substrate.
Large anion size was also found to favor Al stabi-
lization, since the 1710 cm-1 signature did not appear
till a potential higher than 5.0 V was reached for the
new, asymmetric imide. Unlike the interpretation of
Krause et al.,141 Kanamura et al. attributed its
inertness to the well-distributed negative charge and
the lower probability of these anions forming an ion
pair with Al3+.148
The only in situ and semiquantitative study on Al
corrosion was carried out by Yang et al. using EQCM
to monitor the mass change of an Al electrode in
various electrolytes based on LiIm, LiTf, LiPF6,
LiBF4, and LiClO4.147 The most interesting discovery
was that, during the process of corrosion in all of the
electrolytes, the mass change of Al was positive (i.e.,
the Al electrode gained net weight) instead of losing
weight, as one would expect from a simple anodic
dissolution of Al.
Furthermore, normalization of the weight gain
against the corresponding charge quantity offered a
semiquantitative method to identify the possible
species formed, and in the case of LiBF4- and LiPF6-
based electrolytes, ¢m/¢Q ) 26-27 was obtained,
suggesting that essentially identical species were
adsorbed on Al. Compounds with similar ¢m/¢Q
values include AlF3 (¢m/¢Q ) 19) and Al2(CO3)3 (¢m/
¢Q ) 30). On the other hand, LiTf- and LiIm-based
electrolytes are characterized by very high ¢m/¢Q
values. For example, in chronoamperometry stepped
from 2.0 to 4.5 V, an Al electrode in LiIm/PC is
accompanied by a ¢m/¢Q ) 174. It appears that the
initial oxidation could be due to the formation of an
Al compound of large molecular weight such as Al-
(Im)3 (¢m/¢Q ) 280.2) or a reaction intermediate of
Al+ or of a solution species, 141 but these compounds
rapidly desorb from the Al surface, reducing the
numerator in ¢m/¢Q.
On the basis of the EQCM observations, the
authors proposed an adsorption/oxidation/desorption
mechanism for the severe pitting corrosion of Al in
LiIm- and LiTf-based electrolytes, which is schemati-
cally shown in Scheme 19 and Figure 27b.147 Accord-
ing to this mechanism, Al oxidizes to form adsorbed
Al(Im)3 that eventually desorbs from the surface
because these species are soluble in the electrolyte
solvents. It is the desorption of these oxidized prod-
ucts that leaves the otherwise smooth Al surface with
pits. The possibility also exists that, before desorption
occurs, the adsorbed species undergoes further oxida-
tion; however, since the oxidation of Im- is insignifi-
cant below 4.5 V according to studies carried out on
nonactive electrodes similar to Al,81,130,131,206 it seems
unlikely that further oxidation of the adsorbed Al-
(Im)3 would occur.
On the other hand, the oxidized products in LiPF4
and LiBF4 electrolytes could consist of less soluble
species such as AlF3 and Al2(CO3)3, as suggested by
¢m/¢Q ) 26-27, indicating negligible desorption
from the Al surface and therefore less corrosion of
the Al substrate.
6.4. A Few Words on Surface Characterizations
Beyond any doubt, the electrode/electrolyte inter-
faces constitute the foundations for the state-of-the-
art lithium ion chemistry and naturally have become
the most active research topic during the past decade.
However, the characterization of the key attributes
of the corresponding surface chemistries proved
rather difficult, and significant controversy has been
generated. The elusive nature of these interfaces is
believed to arise from the sensitivity of the major
chemical compounds that originated from the decom-
position of electrolyte components.
Hence, the presence of trace impurities, which
either pre-exist in pristine electrode and bulk elec-
trolyte or are introduced during the handling of the
sample, could profoundly affect the spectroscopic
images obtained after or during certain electrochemi-
cal experiments. This complication due to the impu-
rities is especially serious when ex situ analytic
means were employed, with moisture as the main
perpetrator. For cathode/electrolyte interfaces, an
additional complication comes from the structural
degradation of the active mass, especially when over-
delithiation occurs, wherein the decomposition of
electrolyte components is so closely entangled with
the phase transition of the active mass that dif-
ferentiation is impossible. In such cases, caution
should always be exercised when interpreting the
conclusions presented.
On the other hand, when the chemical composition
of the surface layer is discussed, distinction should
be made regarding the conditions under which such
characterizations were carried out and the history
of the electrode surface. For example, an electrode
surface that was subjected to long-term cycling is
certainly different from the surface state of the same
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4351

gradual decomposition of electrolyte solvents could
occur in the bulk of the solution without involving
electrodes. This propagation of radicals inevitably led
to the polymerization of solvents. Using liquid chro-
matography, which is more sensitive to the detection
of polymeric species, reductively induced polymeri-
zation of ester-based solvents was confirmed.
In the gradual degradation mechanism proposed
by those authors, radical anions are first generated
during SEI formation through a single-electron pro-
cess, as proposed by Aurbach et al.,108,124,249 and then
polymerization proceeded gradually from these radi-
cal anions with subsequent and continuous produc-
tion of more active alkyl radicals. In this sense, even
a perfect SEI that insulates any electron transfer
through it cannot prevent the continuous degradation
of electrolyte solvents.
7.2. Stability of the SEI or Surface Layer at
Elevated Temperatures
The presence of a protective SEI or surface layer
prevents those irreversible reactions of electrolytes
on anode/cathode surfaces that are otherwise favored
by thermodynamics. Like the chemical process in the
bulk electrolyte, the reactivity of the surface films
formed in state-of-the-art electrolytes is negligible at
room temperature. However, during long-term stor-
age and cycling, their stability is still under question.
When conducting a differential scanning calorim-
etry (DSC) study on the stability of carbonaceous
anodes in electrolytes, Tarascon and co-workers
found that, before the major reaction between lithi-
ated carbon and fluorinated polymers in the cell,
there was a transition of smaller thermal effect at
120 °C, marked peak (a) in Figure 28.338 They
ascribed this process to the decomposition of SEI into
Li2CO3, based on the previous understanding about
the SEI chemical composition and the thermal stabil-
ity of lithium alkyl carbonates.102,117,249,281 Interest-
ingly, those authors noticed that the above transition
would disappear if the carbonaceous anode was
rinsed in DMC before DSC was performed, while the
other major processes remained (Figure 28). Thus,
they concluded that the components of the SEI layer
are soluble in DMC.338 Because of the similar
physicochemical properties between various linear
carbonate solvents, this dissolution of the SEI might
very likely be universal in all state-of-the-art elec-
trolytes.
This corrosion of the SEI by linear carbonate
solvents would undoubtedly produce adverse effects
on the performance of lithium ion cells. During long-
term cycling, the damaged SEI has to be repaired
constantly by the same electrochemical reactions that
occurred in the initial formation process, which
consumes the limited lithium ion source in the cell
and increases the impedance at the electrode/
electrolyte interface.
A more detailed study on the thermal stability of
the SEI on lithiated and delithiated graphitic anodes
was carried out by Amatucci and co-workers.339 By
storing the cells at 70 °C for 4 days and then
resuming cycling at room temperature, they pre-
sented direct evidence that the SEI was damaged at
the elevated storage temperature by successfully
detecting the reappearance of the irreversible process
at 0.75 V, which has been previously determined to
be the formation of the SEI by reductive decomposi-
tion of the electrolyte. Furthermore, a correlation was
also established between the capacity loss after such
storages and the rebuilding of the SEI, with the latter
being proportional to such factors as the storage
temperature, the storage time, and the surface area
of the anode. An Arrhenius behavior was actually
observed for the dependence of capacity loss on the
storage temperature, with an activation energy of
39.8 kJ mol-1.339
As expected, the state-of-charge (SOC) of the anode
also influences this rebuilding process, since the
capacity loss due to the storage is much higher for
the fully lithiated carbon anode than for the fully
delithiated one. This fact was explained by the
authors in terms of the reactivity of the carbon
surfaces, which are partially exposed because of SEI
corrosion, toward electrolyte components. Thus, in-
tercalated lithium ions continuously diffuse from the
interior of the graphitic structure through the im-
perfect SEI coverage and participate in the reaction
with electrolyte solvents to re-form the SEI. Under
stationary storage, an equilibrium would be reached
between the SEI dissolution rate and the lithium ion
diffusion rate. However, the net effect would be the
irreversible consumption of electrolyte solvents as
well as lithium ions.
The long-term deterioration of the SEI at elevated
temperatures also was observed by McLarnon and
co-workers, who conducted postmortem analyses on
the electrodes from consumer lithium ion cells that
were stored at various temperatures between 40 and
70 °C for weeks.298 Using the IR absorption at 838
cm-1 as the indicator of the SEI component, they
quantitatively correlated the stability of the SEI on
a carbonaceous anode against storage temperature
and SOC. In an extreme situation where the cell was
stored at 70 °C with 9% swing of SOC, no remaining
SEI was found on the anode surface.
Figure 28. DSC trace of the reactions occurring between
a fully lithiated graphitic anode and electrolyte. Anode
surfaces both rinsed with DMC and unrinsed were studied.
(Reproduced with permission from ref 338 (Figure 1).
Copyright 1998 The Electrochemical Society.)
4354 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

a critical role in defining the surface state of the
anode. A highly speculative mechanism was proposed
in accordance with this hypothesis in which the LiF
formed at 60 °C from a LiBF4-based electrolyte blocks
the intercalation edge sites of the graphitic anode
more effectively than the LiPF4-based counter-
parts.277
The work by Ross and co-workers on the thermal
stability of DEC on a lithiated graphitic anode further
challenged the proposal that Li2CO3 was formed
during the decomposition of the SEI.284 By the well-
controlled exposure of the fully lithiated graphite
LiC6 to DEC vapor in a vacuum and gradually
heating the electrode, they used XPS to monitor the
transformation of the SEI as the function of temper-
ature. From 0 to 110 °C, the major surface species
were believed to be oxalate anions as well as alkox-
ides, while, from 120 to 140 °C, obvious chemical
reactions were observed. On the basis of the C/O
stoichiometric change, it seemed that organic anions
such as oxalate anions were converted into simple
inorganic anions. The authors thus proposed that, in
the exothermic process that was observed by Taras-
con and co-workers in the DSC experiment,338 the SEI
most probably decomposed into simple inorganic
species such as Li2O and CO, instead of Li2CO3.284
On a cathode surface, the instability of electrolytes
at elevated temperatures is also well studied, and it
has been generally established that the continuous
decomposition and rebuilding of surface layers are
responsible for the fade in capacity and loss in
power.296,297,343-346
Using thermogravimetric analysis (TGA), Du Pas-
quier et al. studied the thermal breakdown of the
surface layer on a spinel cathode that was aged in
LiPF6/EC/DMC at 100 °C for 8 h,301 and they detected
the presence of two distinct weight loss processes at
140 and 500 °C. While assigning the latter to the
transformation of spinel into Mn2O3, they believed
that the former process corresponded to the burning
of the organic coating on the cathode surface and the
concomitant release of CO2, similar to the thermal
transition that had been observed on the graphitic
anode in the same temperature range.338 Naturally,
at room temperature, the above destructive process
might only proceed with negligible kinetics.
For a spinel manganese cathode, the decomposition
of the SEI is often entangled with the instability of
the cathode structure in electrolytes at elevated
temperatures, and the participation of electrolyte
components in the Mn2+ dissolution or the dispro-
portionation of the spinel is already widely ac-
cepted.293,301,343 The acidic nature of the electrolyte,
originated from the moisture-sensitivity of PF6- or
BF4- anions, is obviously the main cause, since a
close correlation has been established by Du Pasquier
et al. between the HF content and the Mn2+ content
in the electrolyte301 and LiPF6 has been identified as
the salt species that causes the most severe Mn2+
dissolution in state-of-the-art electrolytes301,343 (Fig-
ure 31), apparently due to its higher tendency to
hydrolyze. Elevated temperatures promote this dis-
solution process in a pronounced way, as Figure 32
shows. The continuous trend in Mn2+ concentration
implies that the dissolution process is kinetically
governed, and there is no threshold temperature at
which this process is switched on. However, accelera-
tion does occur above 40 °C.301 XRD analyses on the
recovered spinel powder stored at 100 °C revealed
the formation of a new phase, which the authors
identified to be a protonated ì-MnO2 that is partially
inert with respect to electrochemistry. To account for
the Mn2+ concentration and the formation of a
protonated ì-MnO2 phase, the authors believed that
Figure 30. Capacity loss due to storage at elevated
temperatures for Li/graphite half-cells. All cells were
precycled at room temperature (cycles 1-3) prior to storage
at indicated temperatures for 1 week, followed by continued
cycling at room temperature. The electrolytes used were
EC/DMC (2:1) and (a) 1.0 M LiPF6 or (b) 1.0 M LiBF4. The
cells were stored in delithiated states. (Reproduced with
permission from ref 277 (Figure 2). Copyright 2001 The
Electrochemical Society.)
Figure 31. Effect of lithium salt on Mn2+ dissolution from
Li1.05Mn1.95O4 stored in EC/DMC (2:1) electrolytes at 55 °C.
The Mn2+ concentrations in the electrolytes were deter-
mined by atomic absorption spectroscopy. (Reproduced with
permission from ref 301 (Figure 4). Copyright 1999 The
Electrochemical Society.)
4356 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

conductivity and thermal reactivity could be best
balanced.370
A recent publication discussing the salt effect on
the thermal safety of a LiCoO2 cathode was entitled
with an intriguing but seemingly paradoxical ques-
tion: “Can an electrolyte for lithium ion batteries be
too stable?”, which revealed the relative importance
of the interfacial stability arising from the passiva-
tion efficiency against that of the “intrinsic stability”
with respect to the safety of cathode materials in
lithium ion cells.371 Using an improved DSC tech-
nique that enabled the direct analysis of electrolytes
containing volatile components, the authors investi-
gated the thermal stability of the LiCoO2 cathode in
the presence of the electrolytes based on LiPF6, LiIm,
and LiBeti. The latter two salts, which were based
on an imide anion stabilized by two perfluorinated
sulfonyl groups, had been developed in the early
1990s to replace the thermally unstable LiPF6 and
have been reported to be thermally stable till >300
°C.146
Surprisingly, when the cathode material, LiCoO2,
was in the presence of these “thermally stable” salts,
LiIm and LiMe, much higher reactivity was detected
than that in the presence of LiPF6, as indicated by
the total absence of any combustion suppression on
SHR that had been observed with LiPF6.371 DSC
results of LiCoO2 in the presence of LiIm- or LiBeti-
based electrolytes confirmed the above observation,
which showed the onset thermal decomposition of
LiCoO2 to be at 280 °C, whereas in LiPF6-based
electrolytes the same thermal event was much sup-
pressed in terms of heat evolution as the concentra-
tion of LiPF6 increased. In other words, the presence
of LiIm and LiBeti did not introduce any increase in
the thermal stability of the electrode, while LiPF6,
although believed to be thermally unstable, efficiently
suppressed the thermal decomposition of the cathode.
The authors ascribed the above “abnormality” to
the passivation of the cathode surface by the reaction
products of the electrolyte solvent and salt. Since
LiPF6 readily decomposes organic solvents such as
EC through a ring-opening mechanism at relatively
low temperatures, the decomposition products, which
were believed to consist of a wide variety of com-
pounds, including a PEO-like polymer shown by
Schemes 4, 12, and 15, deposited on the cathode
surface and formed a protective layer between the
highly oxidizing cathode materials and the bulk
electrolyte solvents. According to the hypothesis of
the authors, this polymeric coating on the cathode
particles strongly delayed the thermal combustion of
the solvents by hindering the release of oxygen,
resulting in a more controlled thermal decomposition
of LiCoO2. On the other hand, similar polymer species
were not formed from LiIm- or LiBeti-based electro-
lytes because of the stability of these salts, and the
thermal combustion of electrolyte solvents in the
presence of charged cathode materials proceeded
unhindered, releasing heat that is sufficient to trigger
thermal runaway.
Another well-studied salt, LiBF4, was also believed
to be more thermally and chemically stable than
LiPF6 in terms of its higher tolerance against trace
moisture and lower tendency to react with cyclic
carbonates in a manner similar to that shown by
Schemes 4, 12, and 15.132,133 However, a similar
paradoxical conclusion was drawn about the thermal
safety of LiCoO2 in the presence of the electrolyte
based on it, whose inability to produce polymeric
coatings on the cathode is held responsible.371 The
above results from Dahn and co-workers suggested
that the salts that were traditionally thought to be
thermally stable, such as LiBF4, LiIm, and LiBeti,
should not be used in lithium ion cells if the thermal
safety is the top concern in the application environ-
ment, which includes large size cells working under
high-rate discharge, at elevated temperatures or at
the risk of mechanic abuses and so forth, and that
LiPF6 remains the electrolyte solute of choice in
terms of thermal safety.
8. Novel Electrolyte Systems
8.1. Problems Facing State-of-the-Art Electrolytes
Summarizing the materials reviewed in sections
2-7, one can immediately conclude that the current
state-of-the-art electrolyte systems for lithium ion
batteries are far from perfect and that, at least in
the following four aspects, there is still room for
possible improvement. Therefore, research and de-
velopment efforts are continued in an attempt to
reformulate new electrolyte systems or to modify the
current state-of-the-art electrolyte systems.
(1) Irreversible Capacity. Because an SEI and
surface film form on both the anode and cathode, a
certain amount of electrolyte is permanently con-
sumed. As has been shown in section 6, this irrevers-
ible process of SEI or surface layer formation is
accompanied by the quantitative loss of lithium ions,
which are immobilized in the form of insoluble salts
such as Li2O or lithium alkyl carbonate.262 Since most
lithium ion cells are built as cathode-limited in order
to avoid the occurrence of lithium metal deposition
on a carbonaceous anode at the end of charging, this
consumption of the limited lithium ion source during
the initial cycles results in permanent capacity loss
of the cell. Eventually the cell energy density as well
as the corresponding cost is compromised because of
the irreversible capacities during the initial cycles.
The extent of the irreversible capacity depends on
both the anode material and the electrolyte composi-
tion. Empirical knowledge indicates that the PC
presence, which is well-known for its tendency to
cause the exfoliation of the graphene structures, is
especially apt to induce such irreversible capacities.
On the other hand, reformulation of the electrolyte
may lead to significant reduction in the irreversible
capacity for given electrode materials.
(2) Temperature Limits. The two indispensable
components of the present lithium ion electrolyte
systems are LiPF6 as salt and EC as solvent. Unfor-
tunately, these two components also impart their
sensitivity to extreme temperatures to the lithium
ion technology, thus imposing temperature limits to
the operation of lithium ion cells. In a somewhat
oversimplified account, one can hold EC responsible
4362 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

for the lower, and LiPF6 for the higher, temperature
instabilities.
Thus, at temperatures lower than the liquidus
temperature (usually above -20 °C for most electro-
lyte compositions),50e,159,160 EC precipitates and dras-
tically reduces the conductivity of lithium ions both
in the bulk electrolyte and through the interfacial
films in the system. During discharge, this increase
of cell impedance at low temperature leads to lower
capacity utilization, which is normally recoverable
when the temperature rises. However, permanent
damage occurs if the cell is being charged at low
temperatures because lithium deposition occurs,
caused by the high interfacial impedance, and results
in irreversible loss of lithium ions. An even worse
possibility is the safety hazard if the lithium deposi-
tion continues to accumulate on the carbonaceous
surface.
At temperatures higher than 60 °C, various de-
compositions occur among the electrolyte compo-
nents, electrode materials, and SEI or surface layers,
while LiPF6 acts as a major initiator or catalyst for
most of these reactions.152,310,332,333 The damage caused
by high-temperature operation is permanent. Be-
cause gaseous products accumulate, a safety hazard
is also likely. Therefore, the specified temperature
range for the normal operation of most commercial
lithium ion cells is -20 °C to +50 °C. While sufficient
for most consumer purposes, the above range severely
restricts the applications of lithium ion technology
for special areas such as military, space, and vehicle
traction uses.
(3) Safety and Hazards. The linear carbonate
solvents are highly flammable with flash points
usually below 30 °C. When the lithium ion cell is
subject to various abuses, thermal runaway occurs
and causes safety hazards. Although electrode ma-
terials and their state-of-charge play a more impor-
tant role in deciding the consequences of the hazard,
the flammable electrolyte solvents are most certainly
responsible for the fire when a lithium ion cell vents.
The seriousness of the hazard is proportional to the
size of the cell, so flame-retarded or nonflammable
lithium ion electrolytes are of special interest for
vehicle traction batteries.
(4) Better Ion Transport. In most nonaqueous
electrolytes, the ion conductivity is much lower as
compared with aqueous solutions, and the part of the
current that is carried by the lithium ions is always
less than half.176 Although it has been found that,
for actual cell operation, the impedances at the
interfaces of anode/electrolyte and cathode/electrolyte
weigh far more than the bulk ion conductivity does,
there is a semiempirical rule with very few excep-
tions: the higher the bulk ion conductivity of an
electrolyte is, the more conductive the SEI or surface
films formed in this electrolyte can be. On the other
hand, the improvement in lithium ion transference
number is certainly welcome, although its signifi-
cance in liquid electrolytes might not be as high as
in polymer electrolytes.
Since the inception of lithium ion technology, there
has been a great deal of research aimed at improving
the state-of-the-art electrolyte systems via various
approaches, including the development of new elec-
trolyte solvents and salts and the application of
functional additives. This section is dedicated to cover
these efforts, most of which, although, have not been
adopted in actual lithium ion electrolytes.
8.2. Functional Electrolytes: Additives
Instead of entirely replacing the major components
of the current state-of-the-art electrolyte systems that
cause problems, an efficient and economical alterna-
tive is to modify certain targeted functions of the
electrolytes by incorporating a new component at
small concentrations, known as an additive, so that
its potential impact on the existing electrolyte can
be minimized. In this way, the bulk properties of the
electrolyte system can be maintained with the al-
ready proven merits such as cost and environmental
concerns barely changed, since the presence of the
new component in the bulk is negligible. On the other
hand, the additive could significantly change the
targeted property. This is especially pronounced in
terms of interfacial properties because these additives
are usually preferentially involved in interfacial
redox processes before the main components of the
bulk electrolyte are.
The additive approach has been used for lithium
batteries to improve the surface morphology of a
lithium electrode so that dendrite growth can be
avoided.10, 372-375 Since the concept of the SEI was
proposed by Peled et al.,37 the emphasis has been
placed on the reductive decomposition of these addi-
tives and the effect of the decomposition products on
the physicochemical properties of the SEI. Obviously,
when carbonaceous materials replaced lithium metal
as anodes, the same line of thought led to the
attempts at controlling the SEI chemistry by means
of using various additives.376 During the past decade,
this approach has been thoroughly studied, the focus
of which has been placed at the SEI on the anodes,
although additives targeting other cell components
have also been developed. However, because of direct
commercial interest, most of the work on additives
has never been published in technical journals,
especially the work that has eventually been accepted
for use in commercial lithium ion cells. As an
alternative, patent disclosures and conference ab-
stracts did reveal scattered information about this
aspect, although fundamental insight is usually
unavailable in these forms of literature. In recent
years, electrolytes containing additives have been
named “functional electrolytes” by some research-
ers.377
According to the functions targeted, the numerous
chemicals tested as electrolyte additives can be
tentatively divided into the following three distinct
categories: (1) those used for improving the ion
conduction properties in the bulk electrolytes; (2)
those used for SEI chemistry modifications; and (3)
those used for preventing overcharging of the cells.
Since the additives designed for the last purpose are
usually compounds with oxidation potentials close to
the operating potential of the cathode materials, the
coverage of additive studies has included essentially
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4363

every major component of the lithium ion cell that
interacts with the electrolytes.
8.2.1. Bulk Electrolyte: Ion Transport
The ability of crown ether to coordinate with
lithium ion has long been recognized, and in terms
of the cavity size, 12-crown-4 and 15-crown-5 have
been identified as the most efficient ligands for
lithium ion.378 The idea of using these cyclic polyether
compounds to promote the solvation of lithium salts
in nonaqueous electrolytes was actively pursued
when rechargeable cells based on lithium metal
electrodes were still the commercial objective. It has
been found that the presence of both 12-crown-4 and
15-crown-5 can effectively improve the solubility of
the lithium salts and increase the ionic conductivity
in the resultant electrolytes, especially when the
solvents have low dielectric constants.379-383 This
improvement in bulk ion conductivity is also reflected
in the interface properties, as the charge-transfer
resistance on the LiCoO2 cathode is also reported to
be reduced because of the presence of 12-crown-
4.380,382 The electrochemical stability limits are not
obviously influenced by crown ethers, but considering
their ether-like structure, one should be concerned
with their stability on the fully charged cathode
surface in the long term. In polymer-based electro-
lytes, 12-crown-4 was also found to decrease the glass
transition temperature of the system.382
On the other hand, since the increase of ion
conductivity is realized through the coordination of
lithium ions by crown ether molecules, the lithium
transference number is lowered as a result of its
presence. In other words, the addition of crown ethers
in nonaqueous electrolytes actually promotes the
undesired anion transport. The main barrier for the
application of crown ethers in electrolyte, however,
is their toxicity. The environmental concern over the
processing and disposal of any materials containing
these crown ethers makes it impossible for industry
to adopt them in large-scale applications.
To develop an additive that selectively coordinates
with salt anions and frees lithium ion for conduction,
McBreen and co-workers pursued a molecular design
and tailor-synthesis approach that yielded several
families of novel compounds based on nitrogen or
boron centers with strongly electron-withdrawing
substituents.
The first family of the so-called anion receptors was
aza-ethers that were based on cyclic or linear amides,
where the nitrogen core was made electron-deficient
by the perfluoroalkylsulfonyl substituents so that
these amides would preferentially interact with the
electron-rich anions through Coulombic attraction,
contrary to how their unsubstituted counterparts
would act.384,385 Two selected representatives from the
aza-ether family are shown in Table 8. When used
as additives in solutions of various lithium halides
LiX in THF, these novel compounds were found to
increase both the solubility and the ion conductivity
of these solutions. For example, the ion conductivity
of the LiCl/THF solution was 0.0016 mS cm-1, while
the LiCl/THF solution with one of the linear aza-
ethers containing eight perfluoroalkylsulfonyl sub-
stituents (n ) 5 for the linear aza-ether shown in
Table 8. Performance-Enhancing Additives for Bulk Properties
4364 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

Table 8) exhibited an increase of 900 times at 1.4
mS cm-1.384 There was an obvious relationship be-
tween the ability of the aza-ether to coordinate the
lithium cations and the extent by which ion conduc-
tivity was improved, since the latter proportionally
increased with both the number of the electron-
withdrawing substituents in the molecule and the
electron-withdrawing ability of the substituents.384,385
For example, tosylsulfonyl, a weaker electron-with-
drawing group as compared with perfluoroalkylsul-
fonyl, proved to be a much less efficient additive.385
On the basis of these dependences, it seemed that
these aza-ethers did act as anion receptors in the
nonaqueous solutions.
To further confirm that these aza-ethers actually
coordinate with the salt anions, the authors used
near X-ray absorption fine structure spectroscopy
(NEXAFS) to study the coordination symmetry of the
chloride anion in LiCl/THF solutions, in the absence
or presence of aza-ether additives, and detected that
the presence of the additives created a clear split in
the Cl K-edge white line peak, as shown in Figure
37, indicative of the interaction between chloride and
the electron-deficient nitrogen.384 Further XRD stud-
ies conducted on complex crystals grown from the
cosolutions of the cyclic aza-ether and lithium halide
salts supported the above NEXAFS observations with
new Bragg peaks representing a larger d spacing
(15 Å) in the crystal. Thus, the authors concluded
that these new aza-ethers are indeed anion receptors,
whose preference in coordinating ions was an exact
opposite to that of the conventional crown ethers.
With the addition of these molecules, the ion con-
ductivity and the lithium transference number were
enhanced simultaneously, a benefit that has rarely
been achieved thus far in nonaqueous lithium elec-
trolytes.
Unfortunately, these aza-ethers showed limited
solubility in the polar solvents that are typically
preferred in nonaqueous electrolytes, and the elec-
trochemical stability window of the LiCl-based elec-
trolytes is not sufficient at the 4.0 V operation range
required by the current state-of-the-art cathode
materials. They were also found to be unstable with
LiPF6.390 Hence, the significance of these aza-ether
compounds in practical applications is rather limited,
although their synthesis successfully proved that the
concept of the anion receptor is achievable by means
of substituting an appropriate core atom with strong
electron-withdrawing moieties.
In their continued efforts, McBreen and co-workers
selected boron, an electron-deficient atom, as the core
to build a series of new anion receptors using the
same tactics with electron-withdrawing substituents.
These new additives can be classified roughly into
thesethreesubcategories: borate,386-388borane,313,386,387,389-391
and boronate.392 Selected representatives from each
category are also listed in Table 8.
Basically, these boron-based anion receptors are
much more efficient in coordinating anions, perhaps
because of the electron-deficient nature of boron,
since most of them can even effectively dissociate LiF
up to 1.0 M, which is virtually insoluble in most
organic solvents, and yield ion conductivities as high
as 6.8 mS cm-1 in DME.313 Considering that the
electrochemical oxidation potential for F- is 5.9 V,
this new electrolyte does indeed seem attractive in
providing a wide electrochemical stability window.
On a GC electrode, the electrochemical stability
limits were found to be in the range 4.05-5.50 V,
usually set by the oxidation of electrolyte sol-
vents,388,392 while, on various cathode materials,
stable cycling performance was observed with the
upper voltage limit as high as 4.30-4.50 V.313,387,390
Cycling tests at elevated temperature (55 °C) further
showed that the electrolytes based on LiF coordinated
by these additives were stable when compared with
the state-of-the-art electrolytes based on LiPF6.389
Similar stability was also found on the carbonaceous
anode surface, where the authors concluded that the
presence of these anion receptors did not interfere
with the formation of the SEI film, and the dissolu-
tion of the SEI was not detected even after heat
treatment that would dissolve LiF salt.391
Among the three subcategories, boronate com-
pounds seemed to be the most efficient in coordinat-
ing with anions and enhancing lithium ion stability,
although the number of electron-withdrawing sub-
stituents in boronate is only two. The authors thus
inferred that the ability of these anion receptors to
capture an anion depends not only on the electron-
deficiency of the core atom but perhaps also on the
steric hindrance presented by these substituents on
the core. With only two substituents, the core of the
boronates is obviously more exposed and therefore
more easily accessible for an anion. The higher ion
conductivity achieved by boronate additive therefore
comes from the better balance between the electron-
deficiency and steric openness of this compound as
Figure 37. NEXAFS spectra at the K-edge of chloride for
(a) LiCl crystal, (b) 0.2 M LiCl/THF, (c) 0.2 M LiCl/THF +
0.1 M aza-ether that does not have electron-withdrawing
substituents on N, and (d) 0.2 M LiCl/THF + 0.1 M linear
aza-ether with n ) 3 in Table 8. Note the white line peak
split when the electron-withdrawing substituents perfluo-
romethylsulfonyl are present. (Reproduced with permission
from ref 384 (Figure 2). Copyright 1996 The Electrochemi-
cal Society.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4365

cointercalation/decomposition process of PC.251,395,396
Thus, Ein-Eli et al. believed that the SEI formation
was initiated by the predominant decomposition of
SO2.395,396 On the basis of the FT-IR analysis of the
graphitic anode surface, they suggested that, in
addition to the solvent decomposition products, lithium
alkyl carbonate, the SEI also contained reduction
products originated from SO2 such as Li2S and
lithium oxysulfur species. An extra merit of SO2 as
an additive in electrolytes is the increase in ion
conductivity, which is caused by its high dielectric
constant and low viscosity.396 However, the obvious
Table 10. LUMO Energy Level and Reduction Potentials of Solvents and Additives
4368 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

that the additive based on ferrocene had little del-
eterious effect on the stability of the cell components.
These promising results initiated a new round of
research activities based on structural modifications
on the cyclopentadieneyl rings.425-429
In a systematical study, Golovin et al. investigated
a series of metallocene derivatives in terms of their
redox potentials, mass transport properties, and
chemical and electrochemical stabilities in both elec-
trochemical test cells and commercial-size AA re-
chargeable cells.429 Figure 43 shows the complete
voltammetric scan of the ferrocene-containing elec-
trolyte on a GC working electrode, where the peaks
indicated as O2 and O3 represent the oxidation of
cyclopentadieneyl rings and electrolyte solvents,
respectively, while R2 and R3 stand for the reductive
decomposition of electrolyte solvents and the deposi-
tion of lithium from solution. Obviously, on the anode
side, the limit was set by the lithium deposition, as
the reduction of solvents only occurs in the first
charging cycle in a lithium-based cell. The anodic
limit, on the other hand, was imposed by the O2. The
redox potential of the shuttle agent ferrocene was
indicated by a pair of closely located peaks [R1]/[O1]
Table 11. Cathode Surface Layer Additives: Overcharge Protection
CID: current interrupter device activated by internal pressure.
4374 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

at 3.25 V, which occurred reversibly. With various
substitutions on the cyclopentadieneyl rings (Table
11), the redox potential could be adjusted to occur in
the range 3.09-3.55 V. For cathode materials based
on LixMnO2, whose potential at full charge is 3.50 V,
these substituted ferrocenes could make feasible
candidates as redox additives, since their redox
potentials can be adjusted to be between those of the
major oxidative decomposition of electrolytes (>4.0
V) and the cathode full utilization (3.50 V).
The effect of these ferrocene-based additives on
overcharge protection is shown in Figure 44, where
AA cells based on lithium, LixMnO2, and electrolytes
with or without additives were overcharged. In the
absence of these redox shuttles (A), the cell voltage
continues to rise, indicating the occurrence of major
irreversible decompositions within the cell whereas
the presence of shuttle agents (B-E) locks the cell
potential in the vicinity of their redox potentials
indefinitely. The successive cyclings of these over-
charged cells showed that no adverse effect was
caused by these ferrocene additives on cycle life.
Furthermore, the reversibility of the redox reactions
and the long-term stability of them with respect to
other cell components were also tested. The authors
had observed over a hundred “turnovers” of these
ferrocene shuttles in the cell while demonstrating a
reversible shuttle effect, and prolonged cyclings of
these cells showed excellent cycle life, indicating the
good compatibility between ferrocene additives and
the bulk electrolytes.
The practical importance of overcharge protection
by these ferrocene additives was further confirmed
with a battery pack in which cells with mismatched
capacity were intentionally connected in series and
subjected to overcharge.427 As expected, the added
ferrocene acted to prevent the cell with low capacity
from being overcharged as the other cell continued
to charge. At the end of the charge, all cells had
attained essentially the same state-of-charge, and
successive discharge and recharge showed excellent
cycling characteristics, despite the mismatched ca-
pacities in the pack. If no redox additives were used,
such mismatch in capacity would result in poor cycle
life or even hazards.
The ability of these ferrocenes as carriers for
shuttling current through the cell was determined
by charging the cell with a constant current at the
fully charged state (3.5 V). When the interelectrode
spacing was between 25 and 50 ím, the limiting
shuttle current of these ferrocenes was found to be
as high as 2.0 mA cm-2.429
One adverse effect of these additives on cell per-
formance seemed to be related to their blocking of
ionic paths on the surface of cathode materials, as
indicated by the reduced power capabilities in the
presence of ferrocenes. Analysis on the concentration
changes of ferrocene additives in the electrolyte
solutions before and after their exposure to cathode
materials established that an adsorption of ferrocene
species occurred on the cathode surface, 93% of which
would be covered when as low as 0.3 M ferrocenes
were present in the electrolyte solution.429 This
surface deactivation resulted in the loss of both rate
capability and capacity.
Redox shuttles based on aromatic species were also
tested. Halpert et al. reported the use of tetracyano-
ethylene and tetramethylphenylenediamine as shuttle
additives to prevent overcharge in TiS2-based lithium
cells and stated that the concept of these built-in
overcharge prevention mechanisms was feasible.430
Richardson and Ross investigated a series of substi-
tuted aromatic or heterocyclic compounds as redox
shuttle additives (Table 11) for polymer electrolytes
that operated on a Li2Mn4O9 cathode at elevated
temperatures (85 °C).431 The redox potentials of these
compounds ranged between 2.8 and 3.5 V, and like
ferrocene-based additives, they are only suitable for
cathode materials of low voltage.
Accompanying the commercialization of lithium ion
technology, the emergence of 4.0 V class cathode
materials based on spinel, LiCoO2, and LiNiO2 pre-
sents a more stringent requirement for the selection
Figure 43. Cyclic voltammograms of 0.08 M ferrocene in
1.0 M LiAsF6/EC/PC conducted at 20 mV s-1. (Reproduced
with permission from ref 429 (Figure 2). Copyright 1992
The Electrochemical Society.)
Figure 44. Voltage profile of overcharged Li/LixMnO2 AA
cells containing different substituted ferrocenes as redox
additives in 1.0 M LiAsF6/EC/PC: (A) reference; (B) fer-
rocenyl ketone; (C) dimethylaminomethylferrocene; (D)
ferrocene; (E) n-butylferrocene. (Reproduced with permis-
sion from ref 429 (Figure 6). Copyright 1992 The Electro-
chemical Society.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4375

compounds as pyrocarbonates,48b biphenyl,48b,435,436
cyclohexylbenzene,48b and phenyl-tert-butyl carbon-
ate,437 while the latter include biphenyl and other
substituted aromatic compounds.438,439 Related pub-
lications on these additives have been absent except
for patent disclosures and conference presentations.
However, according to Yoshino et al., some of these
additives have been used in commercial lithium ion
cells for several years.48b,440 The major difference
between these additives and the redox shuttles is
that, once activated by the cathode potential, the
gassing or polymerizing of these additives terminates
the cell permanently, while the operation of redox
additives is nondestructive, at least theoretically.
Since the redox additives have a maximal current
limit defined by their diffusion coefficient and con-
centration, the presence of the gassing- and polymer-
izing-type additives would serve as a more reliable
line of defense against catastrophic failure from
overcharging. Therefore, the integration of multiple
additives of different types into a single electrolyte
seems to be a feasible approach on the condition that
the destructive additives should have higher activa-
tion potential.
8.3. New Electrolyte Components
The modification of electrolytes via additives is
attractive to industry as an economical approach;
however, its impact on electrolyte performance is
mainly restricted to tuning interfacial-related prop-
erties because of their small concentration in the
electrolyte, while other challenges for the state-of-
the-art electrolytes such as temperature limits, ion
conductivity, and inflammability are still determined
by the physical properties of the bulk components.
Improvements in these bulk-related properties can
only be realized by replacing the bulk components of
the electrolytes with new solvents and salts, but such
efforts have been met with difficulty, since more often
than not the improvement in the individually tar-
geted properties is achieved at the expense of other
properties that are also of vital importance to the
performance of electrolytes. Such “collateral damage”
undermines the significance of the improvements
achieved and, in some cases, even renders the entire
effort unworthy.
Nevertheless, research activities in this arena
continue to be driven by the potential commercial
interest that might arise from any possible replace-
ment of the state-of-the-art electrolyte components.
Realizing that the probability of success is rather
limited with any radical change of the entire elec-
trolyte system, an increasing number of researchers
on novel electrolytes are adopting the current state-
of-the-art electrolytes as the platforms for innovation
and attempting to approach the targeted improve-
ments without serious sacrifices in the well-estab-
lished merits, which at least should include (1) facile
ion transport as characterized by ion conductivities
above 5 mS cm-1 at room temperature, (2) electro-
chemical stability on both carbonaceous anode and
metal oxide cathode materials in the range 0-5 V,
(3) inertness to other cell parts such as packaging
materials and anode and cathode current collectors,
(4) wettability toward porous separators as well as
electrode materials, (5) relatively low toxicity, and
(6) relatively low cost.
This section reviews these research efforts in the
past decade on developing new solvents and lithium
salts for nonaqueous electrolytes of lithium ion cells,
but the cosolvents or additives developed for non-
flammable electrolytes, most of which are phosphorus
or fluorinated molecules, are not included, since their
presence is intended for improvement in safety rather
than performance. They will be reviewed in section
8.5.
With few exceptions, these new electrolyte solvents
focus on possible improvements in low-temperature
performance, while new salts are intended to offer
higher thermal stability. This divided directions of
pursuit after the targeted improvements is appar-
ently created by the fact that solvent and salt,
respectively, impose the upper and lower tempera-
ture limits of the current state-of-the-art electrolytes.
8.3.1. Nonaqueous Solvents
The state-of-the-art electrolytes use mixtures of
cyclic and acyclic carbonates as solvents, whose
functions are to solvate lithium salts and to facilitate
lithium ion transport, respectively. The key cyclic
solvent, EC, is also responsible for forming a protec-
tive SEI on graphitic anodes and probably a similar
surface layer on metal oxide cathodes. However, this
indispensable solvent simultaneously sets the narrow
range of service temperature for these electrolytes
with its high melting point, while its replacement by
other low-melting solvents such as its structural
analogue PC is often rendered difficult by the re-
quirement for SEI-forming ability. The attempts at
solving this dilemma have been directed at the
structural modification of either EC or PC so that a
balance between low melting point and favorable
interfacial chemistry could be reached.
On the other hand, the linear carbonates used in
the state-of-the art electrolytes, DMC, DEC, and
EMC, serve as diluents to the high-melting and
viscous EC. They have been known to be unsuitable
as single solvents because of their inability to solvate
lithium salts as well as their instability on the
oxidizing surface of cathode materials, while the
gassing of lithium ion cells during long-term cycling
is also believed to arise from them. However, any
possible replacements for linear carbonates to serve
as cosolvents with EC or PC should at least possess
the major prerequisites of lower viscosity and lower
melting point.
Preferably, the new solvents are also expected to
possess better stability or ability in interfacial chem-
istry on both anode and cathode materials so that
the new electrolyte formulation can rely less on EC;
or they are expected to be less inflammable, as a
major shortcoming of the linear carbonates is their
low flash points (Tf) (Table 1). In the search for new
solvents, fluorination has been adopted as a favorable
approach to achieve improvements in these two
aspects because the presence of C-F bonds in organic
molecules is found to affect interfacial chemistry on
carbonaceous anodes in a positive manner,441-443 and
4378 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

many organofluorine compounds act as flame-retar-
dants.444,445 Table 12 summarizes the novel electro-
lyte solvents according to their structural cyclicity.
8.3.1.1. Cyclic Solvents. Halogen substitution on
the carbonate ring of EC and PC is postulated to
serve the dual purpose of lowering melting temper-
ature by breaking the molecular symmetry and
improving the SEI-forming ability. Shu et al. used
chloroethylene carbonate (ClEC) as a cosolvent for
PC and found that an effective and protective SEI
could be formed on a graphitic anode, with the
Coulombic efficiency in the first charging cycle com-
parable to that of the commercial electrolytes for
lithium ion cells.446,447 The potential plateau at 0.80
V, characteristic of the reductive decomposition of PC,
was completely eliminated due to the presence of
ClEC, while a new process was observed at 1.70 V.
When taking the irreversible capacity in the first
cycle as a metric, the optimum concentration of ClEC
was determined to be 30 vol %,448 although in a
ternary solvent system containing EC, its concentra-
tion could be minimized to 5%.447 Further electro-
Table 12. Novel Nonaqueous Solvents and Their Major Properties
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4379

anodically stable solvent has been confirmed by the
various applications that followed the report of Xu
and Angell.316,317,472,473 For example, Seel and Dahn
have successfully used EMS as the nonaqueous
electrolyte for an anion intercalation cell, which
enables the otherwise impossible staging of graphene
layers with PF6- anions at 5.60 V;316 and the elec-
trolyte based on the same sulfone also supported the
complete delithiation of a new cathode material at
5.4 V.317
Unfortunately, EMS cannot form an effective SEI
on graphitic anode materials, thus undermining the
possibility of its use in lithium ion cells. In their
further work, Xu and Angell partially fluorinated the
alkyl of an asymmetric linear sulfone, inspired by the
reports that fluorinated alkyl could improve the SEI
chemistry on a graphitic anode.441-443 The fluorinated
sulfone, 3,3,3-trifluoropropylmethyl sulfone (FPMS),
has a melting point at 56 °C and could only be used
as cosolvent with other diluents such as DMC and
EMC.314 The anodic stability of sulfone compounds
seemed to be maintained, despite the presence of
fluorination and the mixing with linear carbonates,
as evidenced by the oxidative limits at 5.70 V
observed in Figure 49 for the mixed solvents. The
SEI-forming ability, however, was indeed improved
due to the fluorination, as shown by Figure 50,
wherein the electrolytes based on these mixed sol-
vents supported reversible lithium ion intercalation/
deintercalation of the graphite anodes. The cycling
tests in the longer term confirmed that the surface
of a graphitic anode was well protected because close-
to-unity Coulombic efficiency was obtained, and the
authors suggested that FPMS might be a promising
candidate for the lithium ion cell if the manufactur-
ing cost of it could be reduced.
8.3.2. Lithium Salts
The pursuit of new lithium salts has been driven
by the thermal instability of the current state-of-the-
art lithium salts that were based on perfluorinated
anions, and the thermal as well as chemical inertness
has been taken as the main metric to evaluate any
potential candidates to replace those salts, although,
similar to the new solvent efforts, often the improve-
ment in these properties comes at the price of other
properties that are equally important for the opera-
tion of a lithium ion cell.
8.3.2.1. Lithium Tris(trifluoromethanesulfon-
yl)methide (LiMe). Following the development of
LiIm, Dominey invented a new lithium salt based on
a carbanion that is stabilized by three perfluorinated
methanesulfonyl groups.474 Because of the effective
delocalization of the formal charge on the anion, the
salt LiMe (Table 13) could be dissolved in various
nonaqueous media and showed better ion conductivi-
ties than LiIm.146 Its stability at high temperature
is confirmed by TGA studies, which detected no sign
of decomposition before 340 °C, while accordingly the
electrolyte solution based on the salt remained stable
at 100 °C. The electrochemical stability of the salt
was studied in THF solution, and the cyclic voltam-
metry conducted on GC showed major anodic decom-
position process at 4.0 V. Although, on the basis of
the previously published data this decomposition
seemed to be caused by the solvent THF rather than
by the salt anion,64,74,177 the authors did not report
further electrochemical measurements in other more
stable solvents.146 Initially the salt was reported to
be inert toward an Al current collector,116,153 but a
more detailed study later found that corrosion of Al
Figure 49. Effect of cosolvent FPMS on the anodic
stability of the mixed solvents. Also shown for comparison
are the neat linear carbonates. In all cases, 1.0 m LiPF6
solutions were used, and slow scan voltammetry was
conducted at 0.1 mV s-1, with lithium as counter and
reference electrodes and spinel LixMn2O4 as working
electrode. (Reproduced with permission from ref 314 (Fig-
ure 6). Copyright 2002 The Electrochemical Society.)
Figure 50. Galvonostatic cycling of anode (Li/graphite)
and cathode (Li/LixMn2O4) half-cells using 1.0 m LiPF6 in
FPMS/EMC 1:1 and 1:2 mixture solvents, respectively.
i ) 0.001 mA cm-2. (Reproduced with permission from ref
314 (Figure 8). Copyright 2002 The Electrochemical Soci-
ety.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4383

by Barthel et al. in the mid 1990s based on a borate
anion chelated by various aromatic ligands.475-480
Table 13 summarizes the representative members
from this class along with their selected physical and
electrochemical properties. The authors described the
synthesis and chemical or physical characterizations
of these salts in detail but only provided limited
electrochemical data. Generally speaking, this class
of salts is rather stable thermally and decomposes
only at very high temperatures without melting,
although moisture can still decompose them through
hydrolysis.481 The solubilities of most of these salts
in nonaqueous media seemed to be dependent on the
substitution of aromatic ligands, and moderate to
good ion conductivities have been reported, ranging
from 0.6 to 11.1 mS cm-1 depending on the solvents
used.
The stabilities of these salts against oxidation were
studied with cyclic voltammetry on various inert
electrodes, and an interesting correlation was estab-
lished between the number of electron-withdrawing
substituents on aromatic rings and the anodic sta-
bilities (i.e., higher oxidation potentials were found
for the better stabilized borate anions with more
electron-withdrawing groups). This correlation could
be well explained by the order of the HOMO energy
levels obtained by quantum chemical calculations.477,479
Thus, the anodic decomposition potential ranges from
3.6 V for the unsubstituted borate475 to 4.60 V for the
borate with fluorinated and sulfonated aromatic
ligands.479 A similar relationship was reported by
Sasaki et al., who listed the following order of
oxidation potential limits according to the voltam-
metry results obtained on a Pt electrode, and re-
vealed the obvious dependence of the anodic stability
of these anions on the electron density of the aromatic
rings:481
In addition to the above thermodynamic consider-
ation, kinetics also play an important role in deter-
mining the anodic stability of these salts. For ex-
ample, some salts whose decomposition products are
polymeric moieties were found to passivate the
electrode surface effectively.478 Therefore, although
the intrinsic oxidation potentials for these anions
were not as high (4.0 V), they showed stability up
to 4.50 V in subsequent scans. It should be cautioned
here, though, as the passivation was only observed
on an inert electrode surface, whether similar pas-
sivations would occur on an actual cathode surface
and act to extend the potential range of application
for these salts remains to be tested. Al was reported
to be stable in the electrolytes based on at least two
of these salts.479,480
The cycling test for one of the salts was conducted
by Handa et al. in lithium cells using a low-potential
cathode material, V2O5.482 OCV and initial discharg-
ing behavior similar to those of other nonaqueous
electrolytes were reported, although no data concern-
ing extended performance were given. The key prop-
erty that would decide whether these salts could be
used in lithium ion technology (i.e., the ability of
forming a protective SEI on the surface of graphitic
materials) has not been reported for any of these
salts.
8.3.2.3. Lithium Borates with Nonaromatic
Ligands. The presence of aromatic ligands in
Barthel’s salts was believed to be responsible for the
high melting points and basicity of the borate anions,
which in turn translate into moderate or poor solu-
bilities and ion conductivities as well as low anodic
stabilities. To avoid use of these bulky aromatic
substituents, Xu and Angell synthesized a series of
borate anions that are chelated by various alkyl-
based bidentate ligands, which serve as electron-
withdrawing moieties by the presence of fluorine or
carbonyl functionalities.113,483,484 Table 13 lists the
selected members of this aromatic-free borate family.
Compared with their aromatic counterparts, these
novel salts showed much higher ion conductivity and
anodic stability, while maintaining comparable ther-
mal stability. Detailed studies of ionics indicated that
these salts could well dissociate in the media of
moderate dielectric constants and yield ion conduc-
tivities slightly lower than those for state-of-the-art
electrolyte solvents.485,486 As an example, in EC/DMC
solutions of lithium bis(oxalato)borate (LiBOB) and
LiPF6, the ion conductivities are 7.5 and 10 mS cm-1,
respectively. For at least one of these salts, the
lithium ion transport number seemed to be higher
than 0.50 because of the large anions size.485,486 The
dependence of ion conductivity on salt concentration
is also different from the familiar bell-shaped de-
pendences observed for LiPF6- or LiBF4-based solu-
tions: the isothermal ion conductivity of these lithium
borate solutions remains almost independent of salt
concentration in the range 0.5-1.0 M, which could
be advantageous for practical applications.
Among these new borates, particular attention
should be paid to a salt based on oxalato ligands,
which has aroused intense interest recently in the
lithium ion research and development community.
This salt was invented by Lischka et al.487 and
independently synthesized and investigated by Xu
and Angell, who also gave it the popular name
LiBOB. Following these extensive physical charac-
terizations, a rather comprehensive electrochemical
evaluation was conducted on this salt by Xu et
al.,155,324,488,489 who found that the solutions of LiBOB
in mixed carbonate solvents met the complete set of
stringent requirements for electrolyte solute intended
for lithium ion cell applications: (1) it is anodically
stable on the surface of composite cathode materials
up to 4.3 V, (2) it can form a protective SEI on the
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4385

surface of graphitic anode materials that supports
reversible lithium ion intercalation/deintercalation,
and (3) it stabilizes an Al current-collector at high
potentials up to 6.0 V. Figures 51 and 52 present a
brief summary of these qualifications, by all of which
LiBOB showed performances comparable with or
even superior to those of LiBF4, LiClO4, and LiPF6.
Especially, compared with the industry standard,
LiPF6, this salt also offers the additional advantages
of being thermally stable, being economical in terms
of manufacturing cost, being more environmentally
friendly by decomposing into less corrosive products
in the presence of moisture. The combination of these
merits has not been found in other novel salts thus
far investigated and has certainly made LiBOB a
hopeful contender for lithium ion technology applica-
tions. The stable performance of LiBOB-based elec-
trolytes was confirmed by the extended cycling of the
full lithium ion cells using such electrolytes, where
no capacity fading was detected during the operation
of 200 cycles.155
An unexpected but certainly welcomed discovery
on LiBOB is its peculiar cathodic chemistry.324 As has
been pointed out in the preceding sections, the
cornerstone of lithium ion technology is the formation
of a protective SEI on the carbonaceous materials,
and the conventional wisdom in this technology is
that PC cannot be used in combination with graphitic
anode materials because of its strong tendency to
cointercalate and exfoliate the graphene structure.
Prior to LiBOB, the presence of electrolyte solutes
alone has never been able to challenge this wisdom,
as Figure 53 shows. As an exception, however, LiBOB
in neat PC solution successfully enables the revers-
ible lithium ion intercalation/deintercalation on vari-
ous graphitic anode materials, with capacity utiliza-
tion and Coulombic efficiencies comparable with
those of the state-of-the-art electrolytes. Considering
that similar stabilization of graphene structure in PC
was only able to be achieved by molecular additives
before, the authors postulated that the BOB anion
must have participated in the SEI formation during
the initial lithiation process of the graphite, most
likely through a single-electron reductive process.324
The durability of such a protective SEI was put to a
stringent test by cycling a full lithium ion cell based
on LiBOB/PC or other EC-free formulations as elec-
Figure 51. Cathodic and anodic stability of LiBOB-based
electrolytes on metal oxide cathode and graphitic anode
materials: Slow scan cyclic voltammetry of these electrode
materials in LiBOB/EC/EMC electrolyte. The scan number
and Coulombic efficiency (CE) for each scan are indicated
in the graph. (Reproduced with permission from ref 155
(Figure 2). Copyright 2002 The Electrochemical Society.)
Figure 52. Passivation of Al substrate in LiBOB-based
electrolytes: Time-decaying current observed on an Al
electrode at various potentials containing 1.0 M LiBOB in
EC/EMC. Inset: the dependence of steady-state current
density (at t ) 103 s) on applied potential as obtained on
an Al electrode in electrolytes based on various salts in the
same mixed solvent. (Reproduced with permission from ref
155 (Figure 1). Copyright 2002 The Electrochemical Soci-
ety.)
Figure 53. Stabilization of graphite in PC by LiBOB.
Voltage profiles of lithium/graphite half-cells containing 1.0
m lithium salts in neat PC as electrolytes. Only for LiBOB/
PC was the complete lithiation/delithiation cycle achieved.
(Reproduced with permission from ref 324 (Figure 1).
Copyright 2002 The Electrochemical Society.)
4386 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

More stringent electrochemical characterizations
were carried by Aurbach and co-workers, who com-
paratively investigated the interfacial properties of
the electrolytes based on LiFAP on anode and cath-
ode materials against the benchmark salts LiPF6 and
LiBeti through various instrumental means, includ-
ing voltammetry, EIS, FT-IR, and XPS.499,500 They
found that, while on both the graphitic anode and
the spinel cathode the LiFAP-based electrolyte showed
higher capacity utilization and better capacity reten-
tion, slower kinetics for lithium ion intercalation/
deintercalation were found in both cases. The cause
for the slow kinetics seemed to be related to the
resistive SEI and surface films on each electrode,
which was unexpected because there should have
been little LiF present in the SEI, due to the stability
of LiFAP, and it was concluded that LiF was mainly
responsible for the high-impedance interfaces on
either the anode or the cathode with LiPF6 as the
electrolyte solute. FT-IR and XPS indicated that
LiFAP-based electrolytes generated higher amounts
of lithium alkyl carbonate on the inert electrode (Pt
and Au) surfaces that were cathodically polarized to
simulate the surface chemistry on a cycled graphitic
anode, but these spectroscopic means failed to iden-
tify any direct involvement of FAP anion in SEI
formation. Nevertheless, it is apparent that in LiFAP-
based electrolytes the surface species are mainly
reduction products from solvents, while in LiPF6-
based electrolytes the anion is apparently more
involved in the reductive decompositions, as evi-
denced by the XPS C 1s and F 1s spectra.499 The
higher stability of LiFAP is shown clearly by Figure
56, wherein two sets of CV were measured for LiFAP-
and LiPF6-based electrolytes after stabilization of the
electrode and after 1 week of cyclings, respectively.
In all the voltammetric profiles, four distinct stages
of lithium intercalation are visible, but a very sig-
nificant difference exists between LiFAP and LiPF6:
nearly identical current responses for LiFAP but an
obvious gap for LiPF6 in the anodic section were
observed. In other words, there were long-term
secondary processes in the LiPF6-based electrolyte
that affected the electrode kinetics, while the gra-
phitic anode in the LiFAP-based electrolyte was
under better protection from such undesirable pro-
cesses.
Therefore, the authors concluded that, although
direct identification was not available through spec-
troscopic means, the FAP anion must have partici-
pated in the formation of surface layers, which served
as protection against sustained decompositions on
one hand but were also responsible for the high
impedance across the interfaces on the other. These
robust surface films might exist on both anode and
cathode surfaces and consist mainly of lithium alkyl
carbonates because of the low level of HF in the
solution.
The thermal stability of LiFAP was also studied
by Aurbach and co-workers in EC/DEC/DMC solution
using ARC. As compared with LiPF6, LiFAP delayed
the onset thermal decomposition of the electrolyte by
10 °C; however, the self-heating became much more
severe once the reactions started.500 The above ARC
was conducted in the absence of electrode materials.
In brief summary, LiFAP as a potential replace-
ment for LiPF6 would result in a stable performance
in lithium ion cells with possibly increased thermal
Figure 55. Presence of fluoroalkyls enhancing the anion
stability against moisture. Consumption of H2O and the
generation of HF in 1.0 M salt solutions of EC/DMC with
added H2O at 500 ppm (for LiPF6) and 1000 ppm (for
LiFAP), respectively: triangle, H2O concentration; circle,
HF concentration. (Reproduced with permission from ref
498 (Figure 2). Copyright 2001 Elsevier.)
Figure 56. Slow scan (10 íV s-1) voltammetry on a
graphite working electrode: (a) 1.0 M LiFAP in EC/DEC/
DMC; (b) LiPF6 in EC/DEC/DMC. Solid line: pristine
graphite. Dashed line: after 1 week of cycling. (Reproduced
with permission from ref 499a (Figure 4). Copyright 2003
The Electrochemical Society.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4389

higher molecular weight, the SEI could be formed
with more desirable attributes. Combining the ob-
servations of Ein-Eli et al.,102 Smart et al.,461,462 and
Herreyre et al.,406 it could be tentatively concluded
that longer alkyl chains in the carboxylic acid section
of the esters play a critical role in determining the
cathodic stability of this component on a graphite
anode. The tests in AA-size full lithium ion cells were
only reported for EA- and MA-based quaternary
electrolytes, and Figure 59 shows the discharging
profiles of these cells at -40 °C. Despite their
negative effect on anode capacity utilization at room
temperature, MA and EA still improved the capacity
significantly.
On the other hand, the presence of these esters in
the electrolyte solutions raised concern over the long-
term performance at room temperatures, because EIS
studies indicated that the resistance associated with
the SEI film increased at a much higher rate for
ester-based electrolytes as compared with the com-
positions that were merely based on carbonates. The
authors attributed this rising cell impedance to the
reactivity of these esters toward the electrode active
material, which resulted in the continued growth of
the SEI film in the long term and suggested that
alkyl esters, especially those of acetic acid, might not
be appropriate cosolvents for low-temperature ap-
plication electrolytes.461
The work of Herreyre et al., however, took a much
more optimistic tone on the use of linear esters EA
and MB.406 In LiCoO2/graphite cells, the ternary
electrolyte compositions such as 1.0 M LiPF6 in EC/
DMC/EA, EC/DMC/MB, and PC/EC/MB were re-
ported to be able to deliver as much as 88-95% of
the rated capacity at -30 °C with a C/2 rate or 81-
87% at -40 °C, while at room temperature the
capacity fading rate (0.05% per cycle for MB and
0.09% per cycle for EA) and capacity retention with
high-temperature storage (60 °C, 14 days at full
state-of-charge) of these cells were comparable with
those of the state-of-the-art electrolytes (0.05% per
cycle). Thus, the authors concluded that these new
electrolytes could be used as cosolvents for low-
temperature electrolytes. Considering that Smart et
al. and Herreyre et al. used entirely different cells
as their testing vehicles, the above discrepancy might
not be too incomprehensible, since similar discrep-
ancies had been encountered when the trends ob-
served in anode half-cells did not correlate well with
that of prototype cells with a different cell design.462
It actually indicated the complexity in the operation
of lithium ion chemistry, during which various factors
including chemistry as well as engineering exert their
influences.
Besides the Coulombic capacity, Herreyre et al.
also pointed out that the depression in cell voltage
at low temperatures could be mitigated by EA and
MB as cosolvents because of the reduced resistance
in both bulk electrolyte and electrolyte/electrode
interfaces. Moreover, they found that this cell voltage
depression was also related to the salt concentration;
using 1.5 M instead of 1.0 M LiPF6 in the electrolyte
enabled the increase of the cell working range from
-30 to -40 °C, as, in most cases, the cell cycling is
regulated by a preset cutoff potential.
Out of the belief that the alkyl esters are reactive
in a lithium ion cell during the long-term operation,
Smart et al. proposed a “carbonate-only” guideline
and formulated a series of quaternary compositions
consisting of EC/DEC/DMC/EMC, in which the EC
concentration remained under 25%.515 The solution
of 1.0 M LiPF6 in EC/DEC/DMC/EMC (1:1:1:3) was
reported to be the best composition, whose ion
conductivity at -40 °C was 1.32 mS cm-1, and its
cycling in LiNiCoO2/MCMB lithium ion cells was
comparable to that of the baseline electrolyte 1.0 M
LiPF6 in EC/DMC (3:7) or (1:1) at room temperatures.
At -20 °C, the superiority of these quaternary
electrolytes became pronounced, as compared with
the ternary and binary baselines shown in Figure 60.
More importantly, these electrolytes also allowed
the lithium ion cells based on them to be charged at
low temperatures at reasonable rates, as shown by
Figure 61, which had been impossible for most of the
lithium ion cells because the high impedance of the
Figure 59. Effect of ester-containing quaternary electro-
lytes on the discharge capacity of AA-size lithium ion cells
(0.4-0.5 A h) at -40 °C under the drain rate of 25 mA.
The electrolytes are (1) 1.0 M LiPF6/EC/DEC/DMC
(1:1:1), (2) 0.75 M LiPF6/EC/DEC/DMC/MA (1:1:1:1), (3)
0.75 M LiPF6/EC/DEC/DMC/EA (1:1:1:1), and (4) 0.75 M
LiPF6/EC/DMC/MA (1:1:1). (Reproduced with permission
from ref 462 (Figure 12). Copyright 2002 The Electrochemi-
cal Society.)
Figure 60. Discharge capacities of LiNiCoO2/MCMB
lithium ion cells at -20 °C with different carbonate-based
electrolytes. Cells are charged at room temperature and
discharged using a C/15 rate. (Reproduced with permission
from ref 515 (Figure 4). Copyright 2003 Elsevier.)
4392 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

cells at a discharged state would result in lithium
deposition on a graphite surface due to the high
overpotential.515 Using a three-electrode cell, Smart
et al. monitored the potential of the cathode and
anode of a LiNiCoO2/MCMB lithium ion cell during
its cycling at -20 °C, and they confirmed that,
although the potential of the graphite anode was
indeed driven to negative regions in the charge
process, no lithium deposition occurred, as evidenced
by the absence of a potential plateau at 0.0 V
corresponding to lithium stripping in the following
discharge process, which had been observed previ-
ously in a similar cell using LiPF6/EC/EMC (1:3) as
the electrolyte.165 They attributed this to the facile
kinetics of lithium ions in the bulk electrolyte as well
as in the SEI film on the graphite anode. The cycling
tests of SAFT prototype lithium ion cells carried out
in a wide temperature range between -70 and 40 °C
confirmed that the above quaternary composition is
an excellent low-temperature electrolyte. As illus-
trated in Figure 62, when the cell is continuously
cycled at -20 °C using a C/10 charge rate and a C/5
discharge rate, stable capacity retention can be
obtained with 80% of the rated capacity delivered.
Using lower rates (C/15 charge and C/10 discharge),
70% of the rated capacity can be accessed even when
cycled at -40 °C.
Fluorinated carbonates were also used by Smart
et al. as low-temperature cosolvents (Table 12), in the
hope that better low-temperature performances could
be imparted by their lower melting points and favor-
able effects on SEI chemistry.466 Cycling tests with
anode half-cells showed that, compared with the
ternary composition with nonfluorinated carbonates,
these fluorinated solvents showed comparable and
slightly better capacity utilizations at room temper-
ature or -20 °C, if the cells were charged at room
temperature; however, pronounced differences in
discharge (delithiation) capacity could be observed if
the cells were charged (lithiated) at -20 °C, where
one of these solvents, ethyl-2,2,2-trifluoroethyl car-
bonate (ETFEC), allowed the cell to deliver far
superior capacity, as Figure 63 shows. Only 50% of
the capacity deliverable at room temperature was
achieved in the best case, however. EIS and polariza-
tion studies on the interface between the electrolyte
and the graphitic anode showed that these fluori-
nated cosolvents in general created less resistive
surface films and more rapid charge-transfer kinetics
when compared with nonfluorinated carbonates, and
the difference increased with decreasing tempera-
ture. Full lithium ion cells based on LiNixCo1-xO2 and
MCMB were also assembled with one of these novel
solvents, and good reversibility was observed, al-
though no data on subzero temperature tests were
reported.
According to the authors, additional merits of such
fluorinated carbonates would include their lower
flammability and higher stability against storage at
elevated temperatures as compared with their non-
fluorinated counterparts; therefore, the incorporation
of these novel solvents into the current commercial
Figure 61. Discharge capacities of LiNiCoO2/MCMB
lithium ion cells at -40 °C with different carbonate-based
electrolytes. Cells are charged at -40 °C and discharged
using a C/4 rate. (Reproduced with permission from ref 515
(Figure 6). Copyright 2003 Elsevier.)
Figure 62. Cycle life performance of SAFT DD-size
lithium ion cells containing 1.0 M LiPF6/EC/DEC/DMC/
EMC (1:1:1:3) at various temperatures. Cutoff voltages for
low-temperature charge were indicated in the graph.
(Reproduced with permission from ref 515 (Figure 11).
Copyright 2003 Elsevier.)
Figure 63. Delithiation capacity of an MCMB anode at
-20 °C in various electrolytes following charge (lithiation)
at -20 °C. The drain rate is 50 mA (C/12). (Reproduced
with permission from ref 466 (Figure 4). Copyright 2003
Elsevier.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4393

drawn by Smart et al., who had identified RSEI as the
limiting resistance in the lithium cell operation at
low temperatures.462 In fact, Smart et al. had already
realized the effect of Rct on low-temperature behavior,
since they also used an effective exchange current
density to represent the lithium ion intercalation
kinetics. This latter quantity should reflect the
overall effect of RSEI and Rct.462,508
Thus, by comparing the impedance spectra of the
lithium ion cells, Zhang et al. showed that the
resistance corresponding to the charge-transfer pro-
cess (Rct) dominates the total cell resistance at low
temperatures. For example, at -30 °C (Figure 66),
the semicircle and the interception corresponding to
Rf and Rb have to be magnified in the inset to be
visible. Considering the huge difference between the
magnitudes of Rb or Rf and Rct, the authors concluded
that the poor low-temperature performance of lithium
ion cells is caused by the slow kinetics of charge-
transfer at these temperatures. The decoupling of the
bulk resistance (Rb, related to the ion conductivity
in the bulk electrolytes) from the charge-transfer
process and film resistances (Rct and Rf, related to
the ion transport in the surface film and the Faradaic
process, respectively) can also be clearly seen in
Figure 66 (at -30 °C): that is, although the cell based
on LiBF4 has higher Rb (the inset), its Rct is much
lower than that of the cell with LiPF6.134 Similar
observations were also made with EC/DMC/DEC
formulations.135 It should be noted that, since full
lithium ion cells were used in these EIS studies, Rct
and Rf mentioned above should reflect the combina-
tion of these corresponding processes on both anode/
electrolyte and cathode/electrolyte interfaces.
In another work unrelated to the low-temperature
electrolytes, Mohamedi et al. characterized the spray-
deposited thin film of spinel cathode material by
means of EIS and studied the correlation between
electrolyte composition and the impedance compo-
nents. Among the three lithium salts investigated,
the lowest Rct and Rf were obtained in a LiBF4-based
electrolyte.517 This observation indirectly corrobo-
rated the conclusions of Zhang et al. and seemed to
indicate that LiBF4 possessed certain qualities of an
electrolyte solute for low-temperature-oriented ap-
plications. In combination with the advances ob-
tained in the solvents approach, further improvement
in the performance at extreme temperatures might
be a probable perspective.
8.4.1.3. Limiting Factors for Low-Tempera-
ture Operation. One controversial topic that has
raised wide attention relates to the limiting factors
of the low temperature of lithium ion cells. The
researchers not only debated about whether the
anode or cathode controls the overall low-tempera-
ture performance of a full lithium ion cell but also
disagree upon the rate-determining steps that govern
the low-temperature kinetics of lithium ion intercala-
tion at the graphitic anode.
Due to the contributions from Smart et al.,462 the
emphasis of low-temperature study on electrolytes
has been placed on the anode side, and the SEI film
on the graphitic surface has generally been recog-
nized as the most resistive component in the journey
of the lithium ions during the cell operation, which
must travel across the electrolyte and intercalate into
or deintercalate from the bulk graphite structure.
This hypothesis was mainly established on the basis
of two observations made in the EIS studies of the
lithium/graphite half-cells: (1) the resistance corre-
sponding to the surface film component (RSEI) far
outweighs the component representing electrolyte
bulk resistance, and (2) the steep temperature-
dependence of RSEI in the low-temperature ranges
below -20 °C matches the rapid deterioration of the
half-cell performance, while the ion conductivities do
not suffer any dramatic drop in this range.164,168
These two phenomena have been repeatedly observed
in various electrolyte systems in which novel cosol-
vents were added to depress the liquidus temperature
and to improve solution transport properties,
and complementary evidence was also obtained
from various electrochemical polarization tech-
niques.466,511,512
The above hypothesis was questioned by Huang et
al., who suggested that the critical factor that limits
the anode capacity accessible at the low temperatures
is the kinetics of the lithium ion in the bulk carbon-
aceous anode instead of the surface film.513 Their
argument was based on the universally observed
asymmetric behavior of graphitic anodes toward
lithiation and delithiation at low temperatures; that
is, while the fully charged graphite can release the
intercalated lithium ions at temperatures below -20
°C with relative readiness, the attempt to lithiate a
fully discharged graphite anode at the same temper-
ature is severely hampered by the high resistance at
the surface. Considering the electrolyte nature of the
SEI, which serves as an electronic insulator but an
ionic conductor, the authors pointed out that it should
neither behave like a diode that only impedes lithium
insertion but allows extraction nor vary in its resis-
tance depending on the state-of-charge of the graph-
ite. Therefore, the above asymmetric behavior should
not be attributed to the SEI, whose resistance should
Figure 66. Nyquist plots of the impedance spectra as
measured for the fully charged lithium ion cells at -30 °C
in which the inset shows the magnified view of the high-
frequency part. Electrolytes are 1.0 m LiPF6 (hollow) and
LiBF4 (solid) in PC/EC/EMC (1:1:3). Note that the semi-
circles in the inset are almost invisible in the scale of the
whole spectra. (Reproduced with permission from ref 134
(Figure 4). Copyright 2002 Elsevier.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4395

be independent of the lithiation degree in the graph-
ite, but rather to the lithium ion diffusion within the
carbonaceous anode structure, which has direction-
ality due to the concentration polarization. Thus, the
diffusion coefficient of lithium ion within the carbon-
aceous anode (DLi) would always be higher for lithium
deintercalation but lower for intercalation, and the
gap between the two processes becomes more promi-
nent as temperature goes down.513
To further support their argument that low-tem-
perature lithium ion intercalation is governed by
lithium diffusion instead of the SEI, the authors
investigated the effect of carbon particle size on its
specific capacity by choosing two carbonaceous an-
odes of the same capacity at room temperature but
different particle sizes (6 and 25 ím). They found
that, when cycling at temperatures below -20 °C, the
anode with fine particles outperformed the one with
coarse particles in the charge capacity, as Figure 67
shows. The authors believed that this particle size
effect was consistent with their contention that
lithium diffusion is the rate-determining step that
limits the low-temperature performance of carbon-
aceous anodes, because the large diameter of the 25
ím particle would require longer lithium diffusion
lengths and, consequently, a lower lithiation degree
would be realized in this case, with the assumption
that DLi is the same for both anode materials.
Combining the above evidences, Huang et al.
further concluded that the improvement in the low-
temperature performances of lithium ion cells would
eventually rely on the effort to develop anode materi-
als of high lithium diffusion coefficients instead of
the electrolytes and SEI that were less resistant.
The conclusion of Huang et al. was supported by
Lin et al., who used a three-electrode cell design to
monitor the voltage profile of both the anode and
cathode in a full lithium ion cell during cycling at
low temperatures and found that these cyclings
resulted in the deposition of metallic lithium on the
graphite surface and the subsequent permanent
capacity loss when an unusual cycling regulation was
applied to deal with the reduced capacity accessible
at low temperatures.165 These authors attributed the
lithium deposition to the excessive concentration
polarization on the graphite surface.
However, more recent work by Wang et al.511 and
Zhang et al.512 seemed to present convincing evidence
challenging the suggestion by Huang et al. that
electrolyte and the SEI resistances do not affect the
low-temperature performance of graphitic anodes.
The former authors used the combination of the
galvanostatic intermittent titration (GITT) technique
and EIS to analyze the individual impedance com-
ponents that simulate graphite in the quaternary
electrolyte composition LiPF6/EC/PC/DMC/EMC
(4:1:3:2). Since the reaction resistance as measured
by GITT is the sum of the electrolyte bulk resistance,
RSEI, Rct, stage transformation, and the resistance
that corresponds to lithium diffusion in graphite, it
is possible to distinguish the impact of their temper-
ature-dependences on the overall electrode kinetics.
They found that, during the low-temperature cy-
clings, the diffusional resistances were similar during
charge and discharge, thus contradicting the hypoth-
esis by Huang et al. that the concentration-depen-
dence of lithium diffusion would result in higher
concentration polarization to lithiation than to delithi-
ation. Furthermore, in a quantitative manner, they
also found that the magnitude of RSEI makes it the
dominating component at the low temperatures when
compared with the other impedance components and,
hence, concluded that at -30 °C the limiting factor
for the lithium intercalation is RSEI.
The latter authors used anode and cathode sym-
metrical cells in EIS analysis in order to simplify the
complication that often arises from asymmetrical
half-cells so that the contributions from anode/
electrolyte and cathode/electrolyte interfaces could be
isolated, and consequently, the temperature-depend-
ences of these components could be established. This
is an extension of their earlier work, in which the
overall impedances of full lithium ion cells were
studied and Rct was identified as the controlling
factor.134,135 As Figure 68 shows, for each of the two
interfaces, Rct dominates the overall impedance in the
symmetrical cells as in a full lithium ion cell, indicat-
ing that, even at room temperature, the electrodic
reaction kinetics at both the cathode and anode
surfaces dictate the overall lithium ion chemistry. At
lower temperature, this determining role of Rct
becomes more pronounced, as Figure 69c shows, in
which “relative resistance”, defined as the ratio of a
certain resistance at a specific temperature to that
at 20 °C, is used to compare the temperature-
dependences of bulk resistance (Rb), surface layer
resistance (Rsl), and Rct. For the convenience of
comparison, the temperature-dependence of the ion
conductivity measured for the bulk electrolyte is also
included in Figure 69 as a benchmark. Apparently,
both Rb and Rsl vary with temperature at a similar
pace to what ion conductivity adopts, as expected, but
a significant deviation was observed in the temper-
ature dependence of Rct below -10 °C. Thus, one
Figure 67. Effect of coke particle size on the charge
capacity at various temperatures. (Reproduced with per-
mission from ref 513 (Figure 8). Copyright 2000 The
Electrochemical Society.)
4396 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

could conclude that the rate-determining factor that
limits the capacity utilization of the graphitic anode
as well as the cathode should most likely be Rct.
Perhaps a more important finding by these authors
is the “concentration-dependence” of Rct. As Huang
et al. has pointed out, RSEI should be independent of
state-of-charge,513 and the authors proved that RSEI
remains relatively constant on both charged and
discharged anode and cathode surfaces, as shown by
the insets of Figure 70. However, Rct showed a strong
dependence on the lithiation degree of both the anode
and cathode. Two extreme situations were simulated
in Figure 70, which correspond to a fully charged
lithium ion cell, wherein the cathode is delithiated
and the anode is lithiated, and a fully discharged
lithium ion cell, wherein the cathode is lithiated and
the anode is delithiated. Apparently, in the former
case, Rct on both the anode and cathode is substan-
tially high, indicating a high polarization for the
lithiation of the anode and delithiation of the cathode,
while the reverse processes for each electrode are
much easier in a fully charged lithium ion cell. Hence,
these authors proposed that the asymmetrical be-
havior of lithium ion cells at low temperature most
likely originated from the concentration-dependence
of Rct instead of the directionality of lithium diffusion
within the bulk graphite anode.512 Since in numerous
reports Rct has been shown to be closely related to
SEI or surface film properties, although its physical
significance remains ambiguous,516 a fact beyond
doubt is that this rate-determining component is
under the influence of electrolyte composition, as
evidenced by Smart et al.’s success.515 On the other
hand, the suggestion by Huang et al. that the
electrolyte and SEI do not control the low-tempera-
ture performance of lithium ion cells seems to be
challenged in these electrode/electrolyte systems.
It must be noted here that the observation made
by Zhang et al. about the dominating magnitude of
Rct could be electrode-specific because Rct is not
always higher than RSEI in all systems and at all
temperatures. At least in the works by Smart et
al.461,462 and Wang et al.,511 Rct in magnitude is
comparable with or even smaller than RSEI. Certainly,
as long as the temperature-dependence and “concen-
tration-dependence” of Rct remain unchanged, the
relative magnitude of Rct versus RSEI should not
reverse the conclusion by Zhang et al. Nevertheless,
in such situations it seems to be necessary to define
a new quantity that would more accurately describe
the overall kinetics of lithium ion intercalation/
deintercalation in the surface film. The interface
exchange current derived from various polarization
techniques that has been used by Smart et al. could
be such a quantity, because both RSEI and Rct are
taken into account.
In most of the situations discussed above, the
graphite anode was investigated as the single and
isolated component, while no or little consideration
was given to the other components where simulta-
neous electrochemical processes occurred. In a full
Figure 68. Nyquist plots of a charged lithium ion cell, a
lithiated graphite/graphite cell, and a delithiated cathode/
cathode symmetrical cell. The inset is an equivalent circuit
used for the interpretation of the impedance spectra.
(Reproduced with permission from ref 512 (Figure 3).
Copyright 2003 Elsevier.)
Figure 69. Comparison for temperature-dependence of the
relative resistances of a charged lithium ion cell, a lithiated
graphite/graphite cell, and a delithiated cathode/cathode
cell. The dashed curves show the “relative resistance” of
the electrolyte, which was taken as the ratio of the
electrolytic conductivity at a specific temperature to the
conductivity at 20 °C: (a) Rb; (b) Rsl; (c) Rct. (Reproduced
with permission from ref 512 (Figure 4). Copyright 2003
Elsevier.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4397

lithium ion cell, which integrates all of those com-
ponents, including electrolyte, separator, cathode,
and anode, the identification of the limiting factor
for low-temperature performance becomes more com-
plicated, and conflicting conclusions have been re-
ported from various authors. Clearly, these conclu-
sions are highly conditional and most likely system-
specific; therefore, caution should always be taken
when interpreting the results of these studies.
Conceptually, besides SEI and charge-transfer
resistance in the interfaces as well as lithium diffu-
sion coefficients in the electrodes, the possible limit-
ing factors that might affect the kinetics of lithium
ion chemistry could also be electrode surface area and
porosity, electrode density and loading, affinity of
binder toward electrolyte, and separator porosity and
lipophilicity. Any material or engineering flaw could
make any of these factors the rate-determining step
of the kinetics; hence, it is not strange that different
limiting factors have been identified for different
lithium ion systems when low-temperature perfor-
mance was investigated.
Ozawa was perhaps the first author who tried to
determine whether the anode or cathode acts as the
kinetic bottleneck in a full lithium ion cell, although
the purpose at the time was not intended for low-
temperature considerations.157 In interpreting the
Nyquist plots obtained in the EIS analysis of the first
generation of Sony 18650 cells, he assigned the larger
semicircle at lower frequency to the anode/electrolyte
interface and the smaller at medium frequency to the
cathode/electrolyte interface on the basis that the
diameter of the former relies on the state-of-charge
of the anode, while the latter becomes larger in
diameter when the surface area of the LiCoO2
cathode decreases. Thus, he concluded that the
charge-transfer process occurring at the anode/
electrolyte interface is the slowest step for the whole
cell chemistry. However, since the EIS study was
conducted on the lithium ion cell without an inde-
pendent reference electrode, the attempt to separate
the contribution from each individual electrode to the
overall cell impedance is deemed unreliable. His way
of assigning the two semicircles to individual elec-
trodes was also questionable, since it is well accepted
that these semicircles correspond to different pro-
cesses rather than different electrode interfaces.134,462
As has been shown in Figure 68, since the time
constants for these two electrochemical components,
RSEI and Rct, are comparable at anode/electrolyte and
cathode/electrolyte interfaces, respectively, the im-
pedance spectra of a full lithium ion could have
similar features in which the higher frequency semi-
circle corresponds to the surface films on both the
anode and the cathode, and the other at lower
frequency corresponds to the charge-transfer pro-
cesses occurring at both the anode and the cathode.512
By incorporating an independent lithium reference
electrode in a commercial lithium ion cell from A&T,
Nagasubramanian managed to separate the contri-
butions from the anode and cathode to cell impedance
and established their individual temperature pro-
files.506 He found that the increase in cell impedance
with decreasing temperature mostly came from the
cathode/electrolyte interface, while the contributions
from the anode/electrolyte interface or the bulk
resistance were negligible. Similar impedance be-
havior was also observed in commercial cells from
Moli and Panasonic.506 Thus, he concluded that the
interfacial resistance at the cathode, which should
include both RSEI and Rct, is mainly responsible for
the poor cell performance at low temperatures. This
conclusion was supported by Chen et al., who inves-
tigated the 18650 lithium ion cell assembled by
Polystor using EIS and identified charge-transfer
resistance at the LiNi0.8Co0.2O2 cathode/electrolyte
interface as the main contributor of overall cell
impedance, although the study was only carried out
at room temperatures.518 An indirect evidence for this
conclusion came from the XPS studies conducted by
Andersson et al. on the cathode and anode surface
from the same lithium ion cell, which showed that
the surface film on the cathode is much thicker
compared with the anode SEI.294
The opposite conclusion was reported by Lin et
al.,165 who used a three-electrode configuration to
study the electrode polarization of the MCMB anode
and LiCoO2 cathode under galvanostatic conditions.
They found that in all cases the polarization at the
MCMB anode surface far outweighs that at the
cathode to such an extent that the potential profile
Figure 70. Asymmetrical behavior of Rct toward lithiation
and delithiation. Comparison between the Nyquist plots
of the anode and cathode symmetrical cells at different
states-of-charge: (a) graphite/graphite; (b) cathode/cathode.
(Reproduced with permission from ref 512 (Figure 5).
Copyright 2003 Elsevier.)
4398 Chemical Reviews, 2004, Vol. 104, No. 10 Xu

of the full lithium ion cell actually mirrors that of
the anode. This dominance of anode polarization
becomes even more severe at low temperatures, and
the logical conclusion should be that the kinetics at
the graphitic anode are the rate-determining step.
In a more recent survey of the commercial lithium
ion cells, Fan believed that the resistance of the
cathode surface layer was the factor limiting the low-
temperature performance of the cell chemistry.507
Unlike the previous researchers who also believed
that the cathode acted as the limiting factor,462,506,512
Fan specifically excluded the role of Rct when he
made the identification, and his major arguments
were based on the summary and interpretation of the
previously published data on lithium diffusion coef-
ficients in both the cathode and anode and the fact
that the surface area of the cathode is normally only
a fraction of the anode. Without data from direct
measurement, the above speculation seems to be
premature; hence, further experimental confirmation
is needed.
With the aim of gaining insight into the kinetics
of lithium intercalation/deintercalation of both an-
odes and cathodes in novel low-temperature electro-
lytes, Smart et al. carried out the Tafel polarization
experiments on MCMB and LiNiCoO2 electrodes as
a function of temperature in a three-electrode cell,
and the results are summarized in Figure 71. In the
temperature range from ambient to -40 °C, the
limiting current densities observed on the anode
remained higher than those observed on the cathode.
In other words, under the condition of galvanostatic
cycling at low temperatures, the cathode would be
preferentially polarized and very likely serves as the
bottleneck for the kinetics of lithium ion chemistry.
As pointed out earlier, these polarization current
densities reflect the resistances of both the surface
film and charge-transfer on anodes and cathodes,
respectively. Interestingly, the temperature-depend-
ences established for the limiting current densities
on anodes and cathodes showed that, as temperature
decreases, the gap between the polarizations at
anode/electrolyte and cathode/electrolyte interfaces
rapidly closes (Figure 71), predicting a switch of rate-
determining step in the temperature ranges below
-40 °C if other factors such as the precipitation of
bulk electrolyte components do not intervene.
The EIS studies using symmetrical cells by Zhang
et al. presented a third answer to the question about
whether the cathode or the anode is the limiting
factor.512 They showed that the controlling factor for
each electrode, Rct, is comparable in magnitude at
room temperature for a graphite as well as a LiNiO2
electrode (Figure 68); and with decreasing tempera-
tures down to -20 °C, these charge-transfer quanti-
ties also decline following a matching profile (Figure
69c). Therefore, merely on the basis of the impedance
measurement, it would be difficult to tell which
interface is rate-determining. Thus, the authors
concluded that charge-transfer processes in both the
anode and the cathode limit the capacity utilization
at subambient temperatures.
8.4.2. High-Temperature Performance
Compared with the efforts spent on the low-
temperature performance, less attention has been
paid to the applications of lithium ion technology at
elevated temperatures, with perhaps storage stability
as the only exception. Cycling tests at temperatures
above 50 °C have been rarely reported in the litera-
ture, most likely owing to the chemical instability of
LiPF6 in the organic solvents at elevated temperature
and the difficulty of replacing it with new lithium
salts.
Using the thermally stable salt LiBOB, Xu et al.
showed that a full lithium ion cell can operate at
temperatures up to 70 °C with limited capacity
fading, while LiPF6-based cells suffer obvious per-
manent capacity loss, as shown in Figure 72.155,492
The authors believed that the chemical stability of
BOB anion and the absence of reactive decomposition
products such as HF and PF5 confer upon the
electrolyte stable performance at elevated tempera-
tures. Liu et al. also reported the improved cycling
performance of lithium ion cells based on the spinel
cathode and LiBOB-containing electrolyte at elevated
temperatures.490 Since Mn2+-dissolution caused by
the HF from LiPF6 has resulted in a severe capacity
fading problem for spinel-based cathode materials,
especially at elevated temperatures, the application
of this promising cathode material has been pre-
vented; therefore, the thermal and chemical stability
of it in the presence of LiBOB might have special
significance.
Another salt that is less sensitive to moisture than
LiPF6, LiBF4, was also tested as an electrolyte solute
intended for high-temperature applications. Zhang
et al. reported that electrolytes based on this salt
could allow the lithium ion cells to cycle at temper-
atures up to 70 °C.132 Irreversible reactions occurred
at temperatures above 80 °C, and the cells lost
capacity rapidly, which was accompanied by the rise
of cell impedance simultaneously.
As the structurally modified version of LiPF6,
LiFAP has also been reported to be less chemically
sensitive due to the partial replacement of fluorine
with the more stable perfluorinated alkyls.496,497
Aurbach and co-workers investigated the stability of
Figure 71. Tafel polarization measurement at different
temperatures performed on MCMB and LiNiCoO2 elec-
trodes in 1.0 M LiPF6/EC/DEC/DMC/EMC (1:1:1:3). (Re-
produced with permission from ref 515 (Figure 8). Copy-
right 2003 Elsevier.)
Electrolytes for Lithium-Based Rechargeable Batteries Chemical Reviews, 2004, Vol. 104, No. 10 4399

interest in nonflammability of electrolytes.524-532 The
overwhelming choice of these compounds has cer-
tainly originated from the in-depth knowledge about
the combustion of organic materials accumulated in
the polymer industry, where halogenated and
organophosphorus compounds have been identified
as the most effective flame retardants.521,522,533 Tra-
ditionally, two major models have been proposed to
explain the flame retardation achieved by these
compounds: (1) char-formation, which builds up a
thermal barrier between the condensed and gaseous
phases, and (2) radical-scavenging, by which the
chain reaction is inhibited in the gaseous phase due
to the radical traps formed by the decomposition
products of these additives.533
Wang et al. seemed to favor the second mechanism
when the effectiveness of organophosphate in flame
retardation was discussed in their work.524 Using
trimethyl phosphate (TMP, R1-3 ) CH3 in Table 14)
as the flame retardant additive/cosolvent, they con-
firmed that when TMP content was higher than a
certain threshold value, which depends on the flam-
mability of the baseline solvents, the electrolyte could
be rendered nonflammable. On the basis of the
previous findings that the radical species containing
phosphate have been detected in the MS, they
proposed that such radicals act like a trap to scav-
enge the main active agent for flame propagation, H¥
radicals. To estimate the minimum amount (Nlimit)
of TMP needed in any binary electrolyte composition
to achieve nonflammability, they even derived an
empirical equation:
The unitless quantity in eq 14, CPTH/CHTP, is the so-
called “nonflammability index” defined by the authors
using the atom content of H or P in the two electro-
lyte components and their boiling points, respectively.
Qualitatively, this equation is of general significance
in that the effectiveness of a certain flame retardant
is proportional to the percentage of P in its molecule
and inversely to its bp, while, for the baseline
components, their flammability is proportional to
their bp and inversely to the H content in their
molecule. The lower flammability of the electrolyte
formulated with TMP was also confirmed to yield
higher thermal stability by calorimetry tests, in
which the thermal reaction between LiPF6/EC/DEC
and the LiCoO2 cathode was apparently suppressed
due to the presence of 20% TMP.
Unfortunately, TMP was found to be cathodically
unstable on a graphitic anode surface, where, in a
manner very similar to PC, it cointercalated into the
graphene structure at 1.20 V and then decomposed
to exfoliate the latter, although its anodic stability
did not seem to be a problem. For this reason, TMP
has to be used in amounts less than 10% with EC
and other carbonates in high concentration in order
to achieve decent performance in lithium ion cells.
However, capacity fading caused by the increase of
cell impedance cast doubt on the application of this
flame retardant in a lithium ion cell.524 To avoid the
poor cathodic stability of TMP on graphitic anodes,
the possibility of using it with other amorphous
carbon electrodes was also explored by the authors.525
The above flame retardants, HMPN and TMP,
along with another commercially available alkyl
phosphate, triethyl phosphate (TEP), were system-
atically characterized by Xu et al.526 To quantify the
flammability of the electrolytes so that the effective-
ness of these flame retardants could be compared on
a more reliable basis, these authors modified a
standard test UL 94 HB, intended for solid polymer
samples, and measured the self-extinguishing time
(SET) instead of the universally used flame propaga-
tion rate.520a Compared with the UL 94 HB, this new
quantity is more appropriate for the evaluation of the
electrolytes of low flammability, since the electrolytes
that are determined to be “retarded” or “nonflam-
mable” by this method all showed zero flame propa-
Table 14. Flame-Retarding Additives or Solvents
log Nlimit ) 2.6 - 9.3[CPTHCHTP] (14)
4402 Chemical Reviews, 2004, Vol. 104, No. 10 Xu