You are on page 1of 14

Journal of The Electrochemical

Society

OPEN ACCESS You may also like


- Least Squares Galvanostatic Intermittent
Galvanostatic Intermittent Titration Technique Titration Technique (LS-GITT) for Accurate
Solid Phase Diffusivity Measurement
Reinvented: Part I. A Critical Review Zheng Shen, Lei Cao, Christopher D.
Rahn et al.

- Porous Electrode Model with Particle


To cite this article: Stephen Dongmin Kang and William C. Chueh 2021 J. Electrochem. Soc. 168 Stress Effects for Li(Ni1/3Co1/3Mn1/3)O2
120504 Electrode
Jing Ying Ko, Maria Varini, Matilda Klett et
al.

- Direct Determination of Diffusion


Coefficients in Commercial Li-Ion Batteries
View the article online for updates and enhancements. Maria Angeles Cabañero, Nicola Boaretto,
Manuel Röder et al.

This content was downloaded from IP address 131.179.156.10 on 10/05/2023 at 08:26


Journal of The Electrochemical Society, 2021 168 120504

Galvanostatic Intermittent Titration Technique Reinvented: Part I.


A Critical Review
Stephen Dongmin Kangz and William C. Chueh
Stanford University, Stanford, California 94305, United States of America

The galvanostatic intermittent titration technique (GITT), introduced in 1977 by Weppner and Huggins, provided a readily
accessible means to measuring the chemical diffusion coefficient of electrochemical electrode materials. The method continues to
be widely used today, but the reported diffusivity values are highly inconsistent, ranging as much as four orders of magnitude for
some Li layered oxide compositions. Even qualitative trends of diffusivity are inconsistent, suggesting significant flaws in the
implementation of the method. Other variants of the GITT method also suffer from similar inconsistency problems. Here we
identify numerous sources of significant error including composition-dependent reaction overpotentials, mathematical flaws in the
relaxation analysis methods, finite-size and non-planar geometry effects, inter-particle inhomogeneity issues, early transient effects,
and surface area uncertainties. We propose a simple relaxation analysis scheme using the time variable t relax + τ − trelax , where
trelax is relaxation time and τ is the galvanostatic pulse duration. We also propose to use dense diffusion-limited samples to isolate
the bulk-diffusion process in the time domain. Chemical diffusivity can be extracted much more reliably with this improved
implementation of the GITT method.
© 2021 The Author(s). Published on behalf of The Electrochemical Society by IOP Publishing Limited. This is an open access
article distributed under the terms of the Creative Commons Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/
by/4.0/), which permits unrestricted reuse of the work in any medium, provided the original work is properly cited. [DOI: 10.1149/
1945-7111/ac3940]

Manuscript submitted September 10, 2021; revised manuscript received November 1, 2021. Published December 3, 2021.

The galvanostatic intermittent titration technique (GITT) is one distance and transition metal repulsion,7–9 but experimental inquiry
of the most widely used methods for measuring chemical diffusivity lags behind in most materials.
in electrochemically active battery materials. Since the first im- Errors in GITT have been attributed to uncertainty in sample
plementation by Weppner and Huggins,1–3 thousands of studies have geometries, Li-composition dependent diffusivity, or non-Fickian diffu-
employed the method, and a majority of the diffusivity values for sion, but these sources only partially account for the literature incon-
battery materials are quoted from GITT measurements. sistencies. For instance, uncertainty in the active material surface
In a GITT experiment, a short pulse of constant current is applied area13,20 should only make the scale of diffusivity uncertain, not the
to electrochemically induce a composition change on the surface of dependency on Li composition (whether an increase in sample area from
an active material in the working electrode. After the pulse, the cell cracking21 compromises the composition-dependency investigation
is rested under an open-circuit condition to approach equilibrium. In could be checked by comparing subsequent cycles). Non-constant
the original recipe,1–3 the cell voltage evolution during a current diffusivity or non-Fickian behavior could require corrections in the
pulse is used to extract a square-root time dependency expected from estimated values, but calculations show that such corrections are minor22
diffusion, and the relaxation period is used to estimate equilibrium relative to the orders-of-magnitude inconsistencies in reported values.
thermodynamic properties. Other variant methods instead extract The inability to explain the inconsistent literature reports suggests that
diffusivity from relaxation curves.4,5 the primary sources of error are yet to be identified.
The sample form in the earliest GITT experiments was dense and Here, we systematically analyze numerous, previously over-
large bulk pieces (e.g. pellets or disks).1,2 Nowadays, multi-particle looked, yet significant error sources. Particularly notable causes
porous electrodes became the standard form for battery devices, and include composition-dependent reaction kinetics, non-diffusion-lim-
the majority of GITT studies apply the technique to porous ited samples, errors in relaxation fitting methods, and misinterpreta-
electrodes. This extension to multi-particle porous systems requires tion of early transient effects. We propose a modified fitting method
a condition that all active particle surfaces remain at similar for the relaxation analysis that uses trelax + τ − trelax as the time
compositions during an electrochemical reaction or relaxation. variable (trelax is relaxation time; τ is the current pulse duration).
This condition is equivalent as requiring each particle reaction to This method mitigates systematic errors present in existing analysis
be limited by solid-diffusion (relative to surface reaction kinetics);6 schemes. We also emphasize the importance of a diffusion-limited
however, this condition is rarely validated or considered. sample. These considerations reveal the ideal setting for evaluating
The most disappointing aspect of the GITT method is that it has chemical diffusivity with the GITT method. An experimental
produced highly inconsistent results in the literature. To illustrate demonstration will follow in Part 2.23
this point, numerous measurement results on the same material
system, LiNi1/3Mn1/3Co1/3O2 (NMC111), are compiled in Fig. 1. Review of the Original Method
Immediately noticeable is the spread of values across four orders of
magnitude. In terms of transport, this uncertainty translates to a Some basic assumptions underlying the original GITT method1–3
difference between 1 and 100 μm-sized particles, an unimaginable become apparent from deriving the time evolution of the working
resolution limit for modern microscopy. Also noticeable is the electrode voltage under a galvanostatic pulse. Suppose that one adds (or
contradicting Li composition dependencies reported in different removes) a neutral species A to (or from) an active material in the
studies. The sky blue markers in Fig. 1 show that opposite Li working electrode through a constant-current electrochemical reaction. If
composition dependencies are reported even in the same study, chemical diffusion occurs through two mobile charge species Az+ and
depending on the current direction. Because of this inconsistency, e−, the time evolution in the chemical potential of the neutral species A
the GITT method as implemented today is not capable of studying at the active material surface is (derived in Appendix A.1.1):
structure-diffusion relationships. Density functional theory investi-
2 −Iv M ∂μA t
gations have predicted a competing interplay between inter-layer μA (t ) − μA (t = 0) = . [1]
π zFS ∂x A D˜

Here, the current pulse starts at time t = 0, I is the Faradaic current


z
E-mail: Stephen.D.Kang@gmail.com defined as positive for oxidation (i.e. negative for species A entering
Journal of The Electrochemical Society, 2021 168 120504

Figure 1. Inconsistent GITT diffusivity values reported in the literature. All data points are from the same material system, LixNi1/3Mn1/3Co1/3O2. In addition to
the orders-of-magnitude inconsistency in the values, the qualitative dependency on Li composition is also discrepant among reports. The sky blue markers
labeled lithiation/delithiation indicate measurements from nominally identical samples but with reversed current directions. Data are from Refs. 10–19.

the material when z > 0), S is the electrochemically active surface including those from charge-transfer or electrical contacts should
area, F is the Faraday constant, xA is the number of species A in the remain minimal relative to the change in the surface chemical
chemical formula, vM is the molar volume (in the solid) of the potential of species A. In the following sections, we highlight several
chemical formula containing species A, and D̃ is the chemical significant error sources in the conventional implementation of the
diffusion coefficient. This solution is for a planar one-dimensional GITT method.
(1D) semi-infinite system, making it valid only for Dt ˜ L2 ≪ 1 in
finite systems with a size of L; the solution is not guaranteed to be a Systematic Errors from Composition-Dependent Overpotentials
good approximation for cylindrical or spherical systems. Equation 1
also requires that all species are added (or removed) at the surface of One essential requirement in using Eq. 2 to extract D̃ is that any
the active material. If the electronic and ionic contacts are not at the change in Rtot (from various sources of overpotentials) during the
same surface, additional considerations could be required depending current pulse is negligible. Whether the changes in overpotentials
on the transference number (see Appendix A.1.1). The active matter or not is determined by the ratio between the change in IRtot
material must also be a single phase; the behavior of a multi-phase and the change in Nernstian voltage corresponding to the surface
material will, in general, not follow24 the simple surface-composi- composition of A:
tion propagation assumed in Eq. 1.
error IδRtot
The voltage evolution of the working electrode (with respect to a = , [ 3]
reference electrode for species A) follows from Eq. 1: signal δV Aeq
eq
2 −Iv M ∂V A t where δ indicates the change during an infinitesimal period of a
V (t ) − V0 = IRtot + . [2] current pulse. The time interval of the current pulse is not relevant
π zFS ∂x A D˜
for assessing this assumption, because both quantities change
Here, V Aeq is the equilibrium cell voltage (from Nernst potential) at a simultaneously during the pulse. By contrast, the magnitude of the
given composition of species A, measured with respect to a current is important as it directly scales with this ratio, but the
reference. V0 is the initial cell voltage before the current pulse, current level is seldom chosen based on this consideration because of
the widespread perception that δRtot is minimal.
ideally equal to V Aeq . Rtot is the resistance of the cell including all
However, Rtot includes components such as charge-transfer
possible sources. Figure 2a shows an example voltage curve for a resistance or electronic resistances, both of which are highly
current pulse applied for a time interval τ. In the original method, a sensitive to the active material composition. To illustrate the
linear fit in V vs t is used to extract D̃ (Fig. 2b). The ∂Veq/∂x factor influence of these factors through an example, Fig. 2a shows a
is estimated by measuring the change in open-circuit voltage after typical GITT curve that one could encounter in an experiment.
full relaxation (the current pulse is usually designed to induce a During a current pulse of τ = 1 in this example, the working
measurable change in overall composition for this estimation). electrode voltage changes with time from two origins: the voltage
Chemical diffusivity D̃ is extracted from the slope of V(t) vs t increases from the change in the surface μA in accordance to Eq. 1 as
under the assumption that Rtot remains constant during the current shown as a dashed line; also, an IR drop component that moderately
pulse. This assumption requires that all changes in the resistances diminishes over time, from a Rtot, here chosen to decrease by 17%
Journal of The Electrochemical Society, 2021 168 120504

Figure 2. Variants of the GITT method. (a) A typical GITT experiment, where a current pulse is applied for an interval of τ, during which the working electrode
voltage changes due to a change in the surface Nernst voltage (dashed line) in addition to an IR contribution. (b) In the original scheme, the pulse period is
analyzed on a t scale, from which a slope is extracted. Because of contributions from composition-dependent overpotentials as in Eq. 3, the slope can be
significantly different than that induced from only diffusion (dashed line). (c) Literature compilation showing the charge-transfer contribution to Eq. 3 in various
systems (6,12,25). The current for this plot is chosen such that Li fraction of 1 changes over 30 hr, typical for a conventional GITT experiment. The dashed line
marks the level used for the example in (a), (b). (d) Another conventional scheme, analyzing the relaxation with respect to the relaxation time on a t relax scale.
The curve is not linear on this scale, and the slope corresponding to diffusion is only found at trelax → 0. (e) The scheme proposed in the present study, where the
relaxation curve is linear with respect to t relax + τ − trelax . Time and voltage scales are normalized such that τ = 1 and the linear slope from diffusion (as in
(e)) is 2 π when D˜ = 1.

over the course of the current pulse τ. This decrease could represent, ˜ for an increasing overpotential upon reaction,
for example, a change in the charge-transfer resistance of a overestimation in D;
deintercalation reaction. In the standard procedure of fitting the D̃ is underestimated. In Li layered oxides, reaction overpotential
cell voltage with respect to t (Fig. 2b), one would extract a slope tends to decrease upon delithiation (Fig. 2c), implying that the
and convert it to D̃. However, this slope is only half of the diffusion- apparent D̃ will be higher when obtained from charging curves
representing slope (dashed line), resulting in a four-fold overestima- (consistent with literature reports). Autocatalysis effects6 exacerbate
this direction dependency. Therefore, it is unphysical to interpret the
tion in D̃. The reason why such a moderate change in Rtot (only 17%)
direction-dependent GITT measurements as evidence for direction-
resulted in a four-fold error in D̃ is that the overpotential from Rtot is dependent diffusivity.
quite big relative to that induced by diffusion. Thus, a small relative We note that this error from composition-dependent overpoten-
change in overpotential could still contribute significantly to the tials is entirely different than the reaction kinetics contribution to
time-dependency of cell voltage (50% of ΔV Aeq in this example). The GITT discussed previously.33 In that report, the problem describes a
impact of current on the determination of D̃ has also been case when only the difference between the initial and final voltages
demonstrated experimentally.26 from a current pulse interval (ΔV) is utilized for evaluation, rather
Composition-sensitive overpotential is a common feature in than a linear fit using V vs t ; the reaction overpotential was
battery materials. Figure 2c compiles the sensitivity of charge- composition-independent in their evaluation, and it was not sub-
transfer resistance Rct relative to the change in Nernst voltages, from tracted when evaluating the time evolution of voltage. Meanwhile,
studies on different Li intercalation materials. The reported values for potentiostatic techniques, correction methods for Rtot have been
are comparable to the 50% value assumed in the example of Figs. 2a discussed previously;34–36 the primary concern in that correction is
–2b, indicating that the “constant IR-drop assumption” is likely the contribution of series resistances Rtot relative to diffusion-
invalid in many battery systems of interest. It should be noted that induced changes, rather than the change of those series resistances
Rct merely represents one contribution to the ratio in Eq. 3; other (δRtot) during a diffusion measurement.
sources could also add or reduce the effect, and, in either way, they Minimizing the current magnitude is an effective way of reducing
distort the t dependency caused by diffusion. errors from Eq. 3. However, it then becomes difficult to accurately
GITT measurements systematically finding higher diffusivities in estimate the final change in Nernst voltage after full relaxation
one reaction direction than the other has been a puzzling feature (recall the conventional GITT implementation aims to measure both
observed in the literature13,27–32 (see sky blue markers in Fig. 1). the time evolution and change in equilibrium Nernst voltage in the
Composition-dependent overpotentials systematically contribute to same experiment). Increasing the current pulse period τ could
such direction dependency through Eq. 3. As seen from the example compensate the small magnitude, but this approach is not feasible
in Fig. 2a–2b, a decreasing overpotential upon reaction results in an with most particle sizes as we will discuss in Section A.1.3. Overall,
Journal of The Electrochemical Society, 2021 168 120504

analyzing the open-circuit relaxation curve (zero current) is a much the GITT relaxation curve with a more specific time variable
more straightforward solution to this IRtot problem. analogous to t . We solve the diffusion-relaxation problem speci-
fically for the composition profile induced by the current pulse. In
Common Problems in Relaxation Analysis Methods other words, the initial condition for the relaxation problem is
Composition-dependent overpotential problems can be circum- explicitly specified as the terminal condition of the current pulse, as
vented by analyzing the relaxation period, where no net-current is shown in Fig. 4a. The analytical solution of this problem in the
applied. Relaxation analysis has been motivated by this reason, but planar 1D semi-infinite limit42 is:
relaxation fitting has not improved the consistency, as evidenced by eq
2 −Iv M ∂V A 1
the square markers in Fig. 1. An important contributing factor to this V (t relax) − Veq = ( t relax + τ − t relax ). [4]
inconsistency is in the mathematical approach of the relaxation π zFS ∂x A D˜
analysis.
One common approach is to analyze the relaxation with respect Again, trelax = 0 is defined at the start of relaxation (i.e. termination
of the pulse; t − τ = trelax). The solution in Eq. 4 suggests that one
to trelax where, trelax = 0 is defined at the moment of current
can simply find the slope between V and trelax + τ − trelax , as
interruption. However, as shown in Fig. 2d, the relaxation curve is
illustrated in Fig. 2e. Although this solution is analytically
linear with respect to trelax only at trelax → 0. In the method
obtainable,42 this specific time variable has not been exploited in
proposed by Honders et al.4,5 (and also a similar variant proposed analyzing the relaxation curve in GITT experiments.
in Ref. 37), this asymptotic slope is used to extract D;˜ however, this Note that the voltage relaxation is faster for smaller chemical
approach is often impractical because other non-diffusion decay
processes dominate at early relaxation times (we discuss the physical diffusivity values, as seen by the proportionality to 1 D˜ in Eq. 4
origin in the section “Early Transient Effects”). Figures 3a–3b and the slope comparison in Fig. 4b. This inverse dependency might
demonstrate how even a modest transient decay signal superimposed seem counter-intuitive at first glance, but it is because slower
on the diffusion signal can result in significant errors. chemical diffusion induces a steeper composition (or chemical
In general, any relaxation equation not depending on the pulse potential) profile subject to relaxation (Fig. 4a). In other words,
condition is insufficient or incorrect in modeling the relaxation the relaxation behavior reflects the composition profile developed
kinetics: the current pulse determines the initial condition of the during the current pulse. This initial condition-dependency explains
relaxation. For example, approximating the surface composition why relaxation Eqs. independent of the current pulse condition
cannot capture the correct relaxation behavior.
profile as a delta function results in a 1 trelax time dependency
The relaxation solution, Eq. 4, also indicates that the decay “rate”
while being independent of pulse duration τ. These equations fail to
in relaxation is independent of D̃. This independence is illustrated by
capture the correct time dependency expected from a GITT experi-
plotting the relaxation curve on a log-linear scale (Fig 4c), and
ment.
taking the derivative with respect to time (Fig. 4d). It is seen that the
Other model calculations directly fitting the decay of V with
decay rate is not a constant, diverging at trelax → 0, but nevertheless
respect to trelax suffer from the lack of specificity, as we illustrate in
Fig. 3c. Using the hypothetical relaxation curve in Fig. 3a, we fit independent of D̃. Relaxation analysis methods that expect a
with the function that is the exact solution of the diffusion-relaxation relaxation rate proportional to D̃ could thus lead to incorrect
problem and let D̃ and the equilibrium potential Veq vary freely until conclusions.
the best fit is found for the decay curve (turquoise solid line). The advantage of analyzing relaxation with the time variable
Because of the superimposed decay signal, the fit is misled to an trelax + τ − trelax is clearly visible when compared with the
underestimated D̃. One might attempt to “improve” the fit by further conventional analysis (compare Figs. 2d vs 2e). High fitting
varying τ, which results in a nearly perfect fit (dashed line). specificity is demonstrated in Fig. 3d, where the non-diffusion decay
However, this attempt further misleads the fitting, resulting in a signal at shorter times is easily separated from the diffusion signal of
much bigger error. This example illustrates how decay-curve fitting interest at longer time scales. In addition, the determination of the
could be vulnerable to noise in practical implementations, which is a slope does not rely on knowing Veq in this modified analysis method
general challenge in decay analysis. An alternative approach is to fit (Veq is time independent), which is another advantage over conven-
the decay rate (d ln V dt ),38 but this rate is not dependent on D̃ as tional decay fitting methods.
will be discussed in the next section. Although this modified method demonstrates straightforward
Another method is to fit the longer time scale exponential advantages, it still shares a number of other potential error sources
relaxation, a technique originating from Hebb-Wagner chemical prevalent in the GITT implementation. In the following sections, we
polarization/depolarization studies. This approach finds the linear analyze these error sources and discuss how they can be mitigated.
slope between ln (V − Veq ) and trelax, which yields a slope π 2D˜ L2 (L Finite Size and Non-Planar Geometry
is the length between the surface and invariant point in a 1D
system).39,40 This mathematical relation is reliable only when the The GITT method was originally proposed for planar 1D systems
“pulse” τ (chemical polarization time) is long enough such that at conditions where the system is effectively semi-infinite, and this
τD˜ L2 > 0.1; at shorter times, diffusivity can be overestimated by assumption underlies both Eqs. 2, 4. It is nevertheless a widespread
several factors or more, as illustrated in Fig. A·2 and Appendix A.2. practice to apply the method to other geometries, most commonly
However, literature studies11,41 have employed this method using finite-sized spherical particles which are of practical interest for
much shorter pulse scales: reported experimental parameters indicate batteries. Such extension of the method has been justified by arguing
that the pulse duration is short in GITT experiments. This argument
at least an order-of-magnitude overestimation in D̃ from this
is best assessed using a dimensionless time scale:
mathematical error. In general, this chemical depolarization method
requires the diffusion time scale (R2 D˜ ) to be experimentally ˜
Dt
accessible while having a well-defined sample geometry. In addition, tˆ = 2 , [5]
L
an accurate evaluation of Veq must be feasible on that time scale.
where L is the length scale of the system (radius in cylinders or
Simple Method Using trelax + τ − trelax spheres). We then examine how the current pulse length τ used in
We have so far illustrated fundamental limitations in both the literature GITT studies correspond to dimensionless time,
conventional GITT pulse analysis and also various GITT relaxation τˆ = tˆ (t = τ ). The literature compilation, Fig. 5, indicates that the
analysis methods. Both problems can be solved if one can analyze majority of the reports extend the GITT method up to the range of
Journal of The Electrochemical Society, 2021 168 120504

Figure 3. Vulnerability of relaxation analysis methods and the high-specificity of t relax + τ − trelax . (a) A hypothetical relaxation data set (markers)
constructed by superimposing an arbitrary exponential decay signal on top of the true diffusion-relaxation signal induced by a τ = 1 current pulse. Dashed line
shows the true diffusion signal. (b) Fitting to the asymptotic slope at trelax → 0, resulting in an underestimated D̃ . (c) Fitting to the entire decay curve with the
exact solution for diffusion-relaxation. By constraining τ = 1 and fitting D̃ and Veq, an underestimated D̃ is found. By relaxing the constraint of τ, a near-perfect
fit can be found, but the error in D̃ dramatically increases. (d) Analysis using the time variable t relax + τ − trelax , with which the correct diffusion signal can
be easily separated from the superimposed decay signal.

Figure 4. Relaxation reflects the chemical potential profile created by the current pulse. (a) Chemical potential profiles created after applying a galvanostatic
pulse to semi-infinite planar solids with different chemical diffusivities. (b) Relaxation analysis showing slopes inversely proportional to D̃ . (c) log μ vs trelax of
the relaxation curve showing a non-constant decay rate (non-constant slope) with time. (d) The decay rates are independent of D̃ (i.e. cannot be used for
diffusivity extraction). Time and chemical potential scales are normalized such that τ = 1 and the linear slope in (b) is 2 π when D˜ = 1. Length is normalized
such that the integrated area under the curve in (a) is unity. Reference chemical potential is set to 0 at depth → ∞.

τ̂ > 0.25 (note that the as-reported values of D̃ and L are used for the
estimation of τ̂). Even in planar 1D systems, this range is already μ (at surface)
μˆ = . [ 6]
above what has been previously suggested as the “safe limit” for the
semi-infinite assumption.4 In cylindrical and spherical systems, the L ( )(
∂μ
∂x
1 −Iv M
D˜ 0 zFS )
associated errors become even larger, as will be explored here.
To assess finite-size effects in GITT experiments, we introduce a Here D˜ 0 is an arbitrary reference value. Using dimensionless
normalized surface chemical potential (dimensionless): parameters, the semi-infinite planar 1D solution, Eq. 1, reduces to:
Journal of The Electrochemical Society, 2021 168 120504

τ̂ is still preferred in this analysis because of the finite-size effect. At


the end of relaxation, ( tˆrelax + τˆ − tˆrelax ) → 0 in Fig. 6b, Δμ˜
approaches a finite value rather than 0 because the system is finite
and the current pulse induces a net change in the sample composi-
tion. When the pulse is short, as seen in Fig. 6b for τ̂ = 0.02, the
linear region extends to as long as ten times of τ̂. By contrast, with a
longer pulse of τ̂ = 0.2 as in Fig. 6c, the linear region only extends
for a short time span, shorter than τ̂. This short linear region makes
the analysis vulnerable to interference with unwanted decay signals
(such as those illustrated in Fig. 3d).
This analysis reveals the advantage of using a shorter τ̂: either
with shorter current pulses or larger system size L. However, to
separate diffusion processes from other relaxation mechanisms,
increasing L is often the only effective strategy. In the following
sections, we discuss other relaxation mechanisms that interfere with
Figure 5. Current pulse intervals used in literature GITT experiments, the diffusivity measurement.
compared on a dimensionless time scale. Pulse intervals were converted with
Eq. 5 using D̃ and L values as reported in each report. Each bar indicates
maximum and minimum values from a report. The colored region indicates Reaction-Limitation and Inter-Particle Composition Differences
the valid range for a short pulse assumption in planar (1D) and spherical (3D) Applying “single particle” equations, such as Eqs. 1, 2, or 4, to
geometries. The compilation includes reports on various material systems multi-particle electrodes requires that the composition gradients
such as Li(Ni,Mn,Co)O2,13–16,43–45 Li(Ni,Co,Al)O2,44,46 LiCoO2,44,47,48 and
develop inside each particle rather than in between particles. In
others.48–52
other words, the system must be diffusion-limited for the reaction;
diffusion-limitation assimilates the composition of each electroche-
mically active surface and creates intra-particle composition gradi-
2 ⎛ D˜ 0 ⎞ ents as assumed in the GITT diffusion model. However, a recent
Δμˆ1D, ∞ = ⎜ ⎟ tˆ , [7] population dynamics study6 has shown that layered oxide particles
π ⎝ D˜ ⎠ (NMC) with typical sizes and morphology develop composition
gradients between particles (inter-particle), an indication of reaction-
which provides the baseline for comparison between different limited reactions. For GITT studies, this inter-particle inhomo-
systems. Solutions for cylindrical and spherical systems are sum- geneity implies that the time evolution of the working electrode
marized in Appendix A.1.3. voltage represents the surface-exchange kinetics between particles
Figure 6a compares the evolution of surface chemical potential in (mediated through the electrolyte) rather than solid-state chemical
finite-sized planar, cylindrical, and spherical systems. In the cylind- diffusion. In other words, short current pulses in the GITT experi-
rical and spherical cases, the curve is not linear with tˆ . Even for ment do not eliminate these inter-particle processes.
current pulses with dimensionless time scales as small as tˆ = 0.2 In general, if an electrochemical reaction in a sample is not
(which is within the semi-infinite regime for the planar 1D case), the diffusion-limited, it becomes inherently difficult to measure che-
discrepancy in both linearity and slope is apparent, and the mical diffusivity from electrochemical signals. Even by making a
discrepancy continues to grow with time. In other words, to avoid perfectly uniform electrode, one cannot escape this challenge
finite-size effects, the dimensionless time scale must be much shorter because of the stochastic nature triggered by a surface-limited
as the system becomes cylindrical and spherical. At τ̂ = 1, the error process in a multi-particle system.6 This inherent limitation suggests
could be >2 times higher in spherical systems, compared with planar that the GITT method requires much larger particle sizes. Ideally,
1D. As evidenced in Fig. 5, these finite size and dimension effects one can use large and dense bulk samples to ensure that the temporal
contribute to the literature inconsistency. evolution of voltage is determined by chemical diffusion in the bulk
Our modified relaxation scheme mitigates these finite-size and at long time scales.
geometry effects as long as the dimensionless pulse is kept small. As
illustrated in Figs. 6b–6c, the relaxation plot with respect to Early Transient Effects
tˆrelax + τˆ − tˆrelax yields a linear region during the initial relaxa- The transient signals at short time scales can represent physical
tion, which could be used to extract diffusivity. Nevertheless, a short processes different than the diffusion process of interest. Multiple

Figure 6. The influence of finite size and geometry on GITT experiments. (a) Temporal evolution of the normalized chemical potential during a galvanostatic
pulse in finite-size planar (black, labeled 1D), cylindrical (violet, labeled 2D), and spherical (turquoise, labled 3D) systems. The planar semi-infinite case (dashed
line) is shown together as a reference. (b) Modified relaxation analysis after a pulse of τ̂ = 0.02 , showing linear slopes identical to that of the semi-infinite planar
1D case for an extended period of time. (c) Modified relaxation analysis after a pulse of τ̂ = 0.2 , where the linear region is significantly reduced. Especially in the
spherical case, the linear region is limited to tˆrelax < 0.1τˆ .
Journal of The Electrochemical Society, 2021 168 120504

mechanisms contribute to this early time scale, and it is challenging always straightforward to determine the total area, especially in
to single out specific processes in the time domain. Nevertheless, it porous composite electrodes. Because of this uncertainty, some
is important that these early transient effects are separated from the authors decide to quote DS ˜ 2 rather than a specific D̃.20
bulk-controlled diffusion signal dominating the longer time scale Area estimation can differ widely depending on the method.
(given that the sample size is large enough for solid-state diffusion to Common methods are to either use the geometrical information
be the slowest process). It is not uncommon to simply take the initial about the particles or measure adsorption with the Brunauer-
and final voltages during a current pulse (ΔV Δ t ) to analyze GITT Emmett-Teller (BET) method.26 A big ambiguity in the geometrical
measurements, but this approach could inadvertently attribute early determination stems from the fact that neither the primary or
transient effects to the bulk process. Here, we demonstrate the secondary particle size is directly relevant for the active surface
impact of having a thin surface layer with low chemical diffusivity; area estimation. For adsorption methods, the surface area depends on
in general, similar effects could result from many other mechanisms the adsorption species used for the measurement because of the
including capacitive charging. porosity at multiple length scales. The BET adsorption area does
In Fig. 7, we construct a model system passivated with a thin correlate with the electrochemically active area,21 but it is difficult to
surface layer. Chemical diffusivity is assumed to be 10 times smaller determine how the two should be mapped. The ambiguity results in
than the bulk. A current pulse is applied such that τ̂ = 10 for the orders of magnitude difference depending on how one decides the
surface layer and τ̂ = 0.01 for the bulk. The simulation shows two active surface area.26 The area uncertainty has an especially big
distinct linear regions with slopes corresponding to the diffusivity of impact on the magnitude of estimated D̃ values because the area
∂μ
the surface and bulk (note that the factor ∂x of the bulk and surface is factor is squared for determining D̃.
assumed identical here for simplicity.). This distinction is clearly If the primary interest is the diffusivity change with respect to
seen not only during the pulse but also during relaxation. Therefore, parameters tunable within an identical sample, such as Li composi-
one can identify the bulk-dominated time scale by observing this tion or temperature, the uncertainty in the surface area could, in
time dependency. principle, be simply left as uncertainty in the absolute scale.
Similar behavior has been reported in the literature,26 especially Increasing electrolyte penetration over cycles (e.g. micro-cracks21)
at low temperatures; however, it should be noted that the passivation could still cause changes in surface area or the diffusion geometry
layer scenario is only one of the many cases in which this early (meso-scale effect32); however, the extent of this problem can be
transient behavior could be observed. Analogous behavior will be estimated by comparing subsequent cycles. Therefore, uncertainty in
further discussed in Part 2 of this report.23 the surface area does not explain the widely different D̃ vs Li
Overall, this surface layer demonstration points to a general fraction trends displayed in Fig. 1.
challenge for the application of the GITT method in samples tailored Ideally, a non-porous sample with a planar surface would
for device performance rather than diffusion measurements. minimize area estimation issues, as was originally suggested by
Especially in porous electrode systems where multiple length scales Weppner and Huggins.1–3 A precaution to be taken here is to have
and kinetic processes operate simultaneously, isolating the chemical dense samples without electrolyte-penetrating pores.53 The exterior-
diffusion process could be inherently challenging. This is another volume density (also referred to as apparent density) of consolidated
reason why large and non-porous particles or samples are favored for bulk samples are often incorrectly characterized with the
diffusion measurements: the longest time scale could then be more Archimedes method (or, similarly using a pycnometer); but this
safely attributed to a diffusion process. method excludes any open pores from the volume measurement, and
is thus not relevant for the exterior-volume density. For example,
cold-pressed and post-sintered samples often produce nearly 100%
Uncertainty in Surface Area
Archimedes density but only 70%–80% exterior-volume density,5
The total area of the electrochemically active surface needs to be indicating enough porosity to permit liquid penetration into the
known for extracting chemical diffusivity from the voltage evolution consolidated bulk. The exterior-volume density of >90%, measured
curve, as seen in Eqs. 2 and 4 (the S factor). Unfortunately, it is not using the dimensions of the surface boundary of the consolidated

Figure 7. Distinguishing surface and bulk diffusion in GITT experiments. A 1D symmetric bulk system passivated with a thin surface layer is constructed, and
the time evolution of surface chemical potential during a galvanostatic pulse-relaxation experiment is simulated. The surface layer is configured such that the
diffusivity and diffusion lengths are 10 and 1 percent of the bulk values, respectively. The pulse length is defined such that the dimensionless pulse time is 10 and
0.01 for the surface layer and bulk, respectively. From the simulated GITT experiment, two linear regimes corresponding to the surface and bulk are observed in
both the current pulse plot (violet; t scale) and relaxation plot (turquoise; t relax + τ − trelax scale). In an actual experiment measuring the electrode voltage,
the two linear regions may only be distinctive in the relaxation plot because of the addition of time-dependent overpotentials during the current pulse.
Journal of The Electrochemical Society, 2021 168 120504

bulk, is usually required to ensure that the pores are not open and magnitude), such that the change in diffusivity is much less than the
penetrating (which can be confirmed by finding an Archimedes exaggerated example of 50% explored here, the associated uncer-
density similar to the exterior-volume density). tainty in diffusivity should be small relative to other sources of
errors. The maximum change of composition at the surface can be
Contributions from the Counter Electrode inferred from the open-circuit voltage after the current interruption.
GITT experiments are commonly conducted on two-electrode From a D˜ (x) curve obtained from a series of GITT experiments, one
cells employing counter electrodes made of Li metal that also play can retrospectively estimate the maximum error from composition-
the role of a reference electrode. It has been shown from both Li-Li dependent diffusivities.
symmetric cells54 and three-electrode cells26,55 that this practice
adds a non-negligible overpotential to the cell voltage. Since the Parasitic Current
counter electrode overpotential evolves with time, it could distort the We have shown that a small current magnitude (not just pulse
time evolution of the cell voltage during the current pulse and thus length) is preferred in GITT experiments even for the relaxation
the diffusion analysis. By analyzing the relaxation period in large analysis. A big trade-off for using a small current magnitude is the
bulk samples, one can avoid this issue even when using two- increased relative contribution from parasitic currents, which could
electrode cells with a Li-metal counter electrode. With a large be approximated as a baseline current. Since the relaxation slope-to-
diffusion length on the working electrode side, one can analyze data diffusivity conversion requires knowing the magnitude of the
at relaxation time scales much longer than the short time scales Faradaic current (for the reaction of interest), parasitic currents
where the Li metal voltage relaxes due to the electrolyte-Li have a direct contribution to errors. One method to mitigate this error
interface.55 Furthermore, a smaller surface area (relative to porous is to repeat the GITT experiment subsequently with at least three
samples) increases the diffusion-induced voltage evolution from the different current magnitudes, and find the proportionality between
working electrode, overshadowing any contributions from the the relaxation slopes (i.e. slopes from V vs trelax + τ − trelax in
counter electrode-electrolyte interface. each pulse-relaxation experiment) and the applied currents con-
trolled by the source (Icrtl). That is:
Composition-Dependent Diffusivity
eq
Departure from linearity in the GITT method is frequently dV 2 v M ∂V A 1
= (−Ictrl + Ipar ).
attributed to composition-dependent diffusivity. To investigate the d ( t relax +τ − t relax ) π zFS ∂x A D˜
extent of this error, we simulate GITT experiments where the solid [8]
experiences an increase or decrease in diffusivity by a maximum of
50% (the surface experiences the maximum change) relative to that By finding this scaling behavior between relaxation slopes and
of the initial composition. Using two study cases of increasing and applied current, one can separate contributions from baseline
decreasing diffusivity, as described in Fig. 8a, the change in surface parasitic currents.
Li chemical potential is calculated for each case during a current
pulse (Fig. 8b) and the following relaxation (Fig. 8c). ∂V eq
Evaluation of the Factor
We find that errors from composition-dependent diffusivity can ∂x

be sufficiently minimized by using the modified relaxation analysis ∂V eq 1


GITT measures the combined factor ∂x D˜
, requiring separate
and by minimizing the surface composition change with a short ∂V eq
current pulse. The pulse analysis from the composition-dependent evaluation of the factor ∂x to extract D̃. This factor is related to the
diffusivity simulation (Fig. 8b) shows that the curve is no longer so-called thermodynamic factor Γ (or enhancement factor) by
linear with respect to t . By contrast, in the modified relaxation ∂VAeq
analysis (Fig. 8c), the linearity is (approximately) maintained. The ΓA =
cA v M zF
RT (− ), accounting for the fact that the actual driving
∂xA
slope of the linear region (1.4 and 0.6 for the D̃-increasing and force for diffusion is a chemical potential gradient rather than
-decreasing case, respectively) corresponds to a diffusivity value ∂V eq
concentration gradient. With an experimental measurement of ∂x ,
bound by the initial and final values of the diffusivity at the surface,
weighted closer toward the diffusivity value at the end of the current the evaluation of D̃ becomes independent from whether the mobile
pulse. As long as the current pulse is kept small (both in time and species is dilute or not. In the original GITT method, this factor is

Figure 8. The effect of composition-dependent diffusivity on GITT experiments. (a) Two cases are simulated: a diffusivity increasing (turquoise) and decreasing
(violet) upon increase in chemical potential. The open markers indicate the diffusivities of the surface composition at the end of the current pulse, which change
by 50% during the pulse. (b) Surface chemical potential change during a current pulse period of τ = 0.1. The dashed line is a reference case for a constant D˜ = 1.
Comparison to the reference curve shows that in both cases the curve is not linear. (c) The modified relaxation plot, where the relaxation curve is approximately
linear in the semi-infinite regime. The diffusivity measured from the linear region is a weighted average between the bounds of the initial and final properties of
the surface. For example, the 1.4 value measured for the increasing D̃ case (turquoise) is between 1 (initial) and 1.5 (final). Scales are normalized such that time
coincides with dimensionless time tˆ , and dμLi d t = 2 π corresponds to D˜ = 1.
Journal of The Electrochemical Society, 2021 168 120504

obtained from the same pulse experiment which is used to also diffusion can also produce distributed impedance components that
∂V eq 1 resemble a diffusion-like impedance feature. For example, the early
obtain D˜
; the change in the open-circuit voltage is measured
∂x transient signal in Fig. 7 translates to a Warburg-like feature in the
after the pulse and long relaxation, and the change in voltage is frequency domain at a higher frequency than bulk diffusion. A
divided by the number of charges passed during the current pulse. distribution of double-layer capacitors or a distribution of reaction
Although this approach may seem attractive because both factors are constants could also create Warburg-like features. One solution to
extracted from a single experiment, significant errors are produced avoid misinterpretation of an apparent Warburg feature is to increase
from this practice. the length scale of the sample and ensure that the diffusion process
The underlying reason is the conflicting sample requirements and of interest corresponds to the lowest frequency among all possible
∂V eq 1 processes. In this extremely low frequency limit, however, it usually
pulse conditions needed for the measurement of ∂x D˜
(from the
becomes easier to measure in the time-domain.
∂V eq
time-dependent voltage) and (from the voltage after full
∂x
Consistency between multiple methods is often considered a
equilibration). For the correct time-dependency extraction, a short validation of a parameter extraction. Contrary to the common belief,
pulse and a large sample dimension are required. By contrast, to such consistency does not necessarily validate the interpretation of a
measure changes in the equilibrium Nernst voltage, long relaxation measurement. For example, assume that a distribution of reaction
times and small sample dimensions are required. Note from the constants produces a Warburg-like feature in the impedance
example in Fig. 6b, where the equilibrium Nernst voltage is still not spectrum. Suppose this feature is then mistaken for a diffusion
reached after relaxing for 500 times longer than the pulse duration. feature, and one extracts an apparent D̃ using Eq. 9. One might then
From the pulse magnitude perspective, the current must not be too try the relaxation analysis from GITT and also extract D̃ using Eq. 8.
big to avoid introducing unwanted contributions to the GITT As long as the time scale (i.e. frequency regime) agrees between the
measurement. By contrast, to produce a measurable change in the two methods, one will find consistency although the actual deter-
Nernst voltage, the pulse must be large enough, especially because mining kinetic process is not diffusion. Since the two methods are
the pulse duration must be kept short. Above all, to ensure diffusion- mathematically equivalent, it does not add validation to the model; it
limitation, large samples or particle sizes are imperative, but large only validates the raw measurement.
samples make it difficult to measure the fully equilibrated voltage
within a practical time frame. As a result, it is challenging to reliably The Recommended GITT Implementation
measure both factors from the same pulse-relaxation experiment.
∂V eq Summarizing the considerations so far, here we list the recom-
A simple solution is to measure the ∂x factor from a separate mended procedure for measuring chemical diffusivity using the
experiment. To increase the measurement reliability of the Nernst GITT method. Part II23 will demonstrate the experimental applica-
voltage vs composition curve (from which the enhancement factor is tion of these procedures to NMC111.
derived), one can utilize small-particle samples and longer pulses to
ensure sufficient capacity per pulse while still being able to reach • Sample For the diffusion measurement, the diffusion length of
equilibrium voltages. In fact, several authors use the GITT method the sample must be large enough to ensure diffusion-limitation.
as a means to only estimate the Nernst voltage but not chemical Large bulk samples are required, and they should be dense enough to
diffusivity.56–59 This approach yields enhancement factors more prevent electrolyte penetration. The electrochemical cell should be
consistently. designed such that the electrolyte or counter electrode has negligible
effects on the voltage relaxation corresponding to diffusion.
Comparison to Other Electroanalytical Methods
• Material composition To homogenize the composition prior
to a pulse-relaxation measurement, annealing processes could be
Other electroanalytical methods commonly used to extract required to avoid long or unpractical relaxation times. Electrolyte
chemical diffusion coefficients include the potentiostatic intermittent removal could be necessary for annealing. The composition must
titration technique (PITT),3,34,60 electrochemical impedance spectro- correspond to a single-phase region in the phase diagram.
scopy (EIS),61 or cyclic voltammetry (CV).62,63 In principle, if the • Pulse design The minimum current pulse magnitude should be
diffusion component can be isolated from other kinetic or transport at least a few times larger than the parasitic current baseline. It
processes in the system, all methods could be equivalent.4,64 should also induce a change in the working electrode voltage large
However, this separation of processes is not always possible, as enough for reliable measurement. The pulse interval should be kept
we have reviewed from the conventional GITT method. Unlike our as short as possible while avoiding non-diffusion transient effects at
modified GITT method where analyzing the relaxation period the shorter time scale.
allowed separation of the current-related overpotentials, other • Slope extraction For each current pulse, the relaxation slope
time-domain methods like PITT or CV cannot use this particular should be obtained from a voltage vs trelax + τ − trelax plot. The
trick to avoid misinterpretation35 because they do not have zero- slope corresponding to the bulk process (not surface) should be
current relaxation periods. identified. At least three consecutive pulse-relaxation experiments
Frequency-domain methods like EIS, on the other hand, could with different magnitudes are recommended (pulses do not have to
potentially provide a better opportunity to separate different kinetic be the same current direction). Then, from a slope vs current
processes. Assuming that a diffusion-related component can be magnitude plot, one can exclude parasitic current contributions using
reliably isolated from an impedance spectrum, an EIS measurement Eq. 8. One can test the consistency between both current directions
is mathematically equivalent to a time-domain measurement. The to ensure that autocatalytic processes are not contributing.
• ∂V∂x factor measurement A separate GITT experiment should
eq
Warburg impedance from diffusion ZDiffusion in the semi-infinite
regime scales inversely with the square-root of angular frequency be conducted using small-size particles. Typical porous electrode
ω:61 composites are more eqfavorable for this measurement compared with
∂V
bulk samples. The ∂x factor is obtained by taking a derivative in the
eq
d∣ Z Diffusion ∣ v ∂V A 1 Nernst voltage vs composition curve.
= M . [ 9]
d (ω )− 1 2 zFS ∂x A D˜ • Iterative check with measured values After extracting D̃
from the measurement, one should calculate the dimensionless pulse
This scaling factor is identical to that of Eq. 8 except for a numerical time using Eq. 5. For planar 1D, τ̂ < 0.25 is recommended. For
factor. Nevertheless, the remaining challenge is to ensure that a cylindrical and spherical systems, a shorter τ̂ (one order of
Warburg-looking feature is from the bulk diffusion process of magnitude smaller) is required. One should also ensure that the
interest. In electrochemical systems, mechanisms other than bulk electrochemical reaction is diffusion-limited.
Journal of The Electrochemical Society, 2021 168 120504

Acknowledgments
This work was supported by the Toyota Research Institute ⎛ ∂μ ⎞−1
JA = −D˜ ⎜ A ⎟ ∇μA . [A·9]
through the Accelerated Materials Design and Discovery program. ⎝ ∂cA ⎠
The authors would like to thank: Devi Ganapathi and Xiao Cui for
contributing to improving the clarity of the manuscript; Daniel Note that this expression resembles Fick’s first law if we convert μ
Cogswell for suggesting useful references. to c (JA = − D˜ ∇cA).
The boundary condition of a GITT experiment is a constant flux
Appendix of the neutral species at the surface, which we can now apply using
A.1.1. Derivation of chemical potential evolution under gal- Eq. A·9: JA(x = 0; t) = JS. Here, x = 0 is the surface with material
vanostatic conditions.—Here, we derive equations 1 starting from extending semi-infinitely in x > 0, and the sign of J is thus positive
general driving force and flux Eqs. for each charged species, rather for influx into the material (see Appendix A.1.3 for finite-size
than only considering the diffusion of neutral species. This general effects). The diffusion problem Eq. A·3 with this flux boundary
approach makes it easier to see the assumptions underlying the condition is now a standard form (see 42). The solution for the
simplified description of just the neutral species. temporal evolution is:
Suppose that stoichiometry change in the reacting material occurs
μA (x; t ) − μA (x; t = 0)
through the diffusion of an ionic species Az+ and electronic species
e−. The neutral species responsible for stoichiometry changes in the ⎡ 2 ∂μ t ⎛ x2 ⎞ x ⎛ x ⎞⎤
solid is A, with a chemical potential μA = μ˜ Az + + zμ˜ e−. The flux of =JS ⎢ A
exp ⎜− ⎟ − erfc ⎜ ⎟ ⎥. [A·10]
⎢⎣ π ∂cA D˜ ⎝ 4Dt
˜ ⎠ D˜ ˜ ⎠ ⎥⎦
⎝ 2 Dt
the neutral species A, denoted JA, requires the flux of the two
charged species, driven by their respective driving forces:
Since the measured cell voltage represents the surface property, we
J Az + = −L ii ∇μ˜ Az + = −L ii (∇μA − z ∇μ˜ e− ) [A·1] let x = 0, and introduce I as the Faradaic current (positive for
oxidation current, following the international convention for electro-
−I S
Je− = −L ee ∇μ˜ e− , [A·2] chemistry) to substitute JS = zF . Also, dcA may be replaced with
dxA/vM, where vM is the molar volume of species A in the solid and
where, Lʼs are coefficients proportional to the conductivity of each xA is the number of species A in the chemical formula corresponding
species (ionic conductivity σi = z2F2Lii; electronic conductivity to each mole unit. With these substitutions, Eq. 1 is obtained.
σe = F2Lee). Cross coefficients of L are assumed to be negligible. Next, we consider the case where non-zero net current flows
From the continuity equation and preservation of local charge through the active material because of a configuration like Fig. A·1b.
density, the master equation for diffusion can be obtained (derived in Since the electrical and electrolyte contacts are on the opposite sides
Appendix A.1.2; note that a zero net current condition is not required of the active material, net current must flow through the material for
here): an electrochemical reaction to happen. If we let i denote the net
∂μA current in the solid,
= D˜ ∇2 μA , [A·3]
∂t i
= zJ Az + − Je−. [A·11]
F
where D̃ is
By substituting the fluxes with Eqs. 1–A, ∇μ˜ e− can be expressed in
L ee L ii ∂μA ⎛ 1 ⎞ σ σ ∂μA terms of μA:
D˜ = = ⎜ 2 2⎟ e i , [A·4]
2
L ee + z L ii ∂cA ⎝ z F ⎠ σe + σi ∂cA
ti iF
∇μ˜ e− = ∇μ + , [A·12]
and cA is the concentration of neutral species A. Note that the Fick’s z A σ
second law is a special case of this general equation Eq. A·3.
To solve the diffusion equation, the fluxes should also be where we introduced total conductivity σ = F2(Lee + z2Lii). Then,
expressed in terms of μA. If we constrain our interest to the case from ∇μA = ∇μ˜ Az + + z ∇μ˜ e−, we can also express ∇μ˜ Az + in terms of
of zero net current inside the material for having ionic and electronic μA:
contacts on the same surface (Fig. A·1a), then zJ Az + − Je− = 0 (this
assumption will be revisited below). Combining this condition with iFz
∇μ˜ Az + = te ∇μA − . [A·13]
Eqs. 1–A, σ

J Az + = −L ii te ∇μA [A·5] By substituting Eqs. 12–A into Eqs. 1–A, we can obtain the net
current case of Eqs. 5–A:
L ee ti
Je− = − ti ∇μA , [A·6] J Az + = −L ii te ∇μA + i [A·14]
z zF
where we have defined the ionic and electronic transference
L ee t
numbers: Je− = − ti ∇μA − e i. [A·15]
z F
z 2L ii σi
ti = = [A·7] Note that the Eqs. converge to Eqs. 5–A when i → 0. From the
L ee + z 2L ii σe + σi additional current term, it is seen that the applied current may not
necessarily correspond to the boundary flux of neutral species A; the
te =
L ee
=
σe
. [A·8] constant flux condition may not hold either. If the charge transport is
2
L ee + z L ii σe + σi ion-limiting (i.e. Lee ? z2Lii and ti → 0), then the ionic current from
the electrolyte terminates at the active material-electrolyte interface
Since JA = J Az + under a zero net current condition, we can rewrite to combine with the electronic current. The Faradaic current at this
Eq. A·6 as: interface is equal to the applied current, also equal to JA at the
Journal of The Electrochemical Society, 2021 168 120504

Figure A·1. Configurations for a diffusion measurement in an electrochemical cell. (a) Net current is zero inside the active material if the electrical and ionic
contacts are made at the same interface. (b) If the electrical and ionic contacts are made at opposite sides of the active material, the compositional evolution and
its correspondence to cell voltage depend on the transference numbers of the mixed conductor.

electrolyte side of the active material. In this ion-limiting case, the A.1.3. Dependence on finite size and geometry.—To straight-
diffusion problem remains identical to that of the zero net current forwardly compare systems with different geometries, we use a
condition (the configuration of Fig. A·1a). In the opposite case of (dimensionless) normalized surface chemical potential μ̂ and time tˆ ,
electron-limited transport inside the active material (ti → 0), ionic as defined in Eqs. 5–6. The semi-infinite planar 1D solution, Eq. 1,
current from the electrolyte transmits through the active material reduces to Eq. 7 with these dimensionless parameters. In a planar 1D
toward the electrical contact interface. Not only does this ion case with finite length L, the solution for surface chemical potential
conduction of the active material move the Faradaic reaction to is modified to:
the electrical contact side of the material, but it also changes the
interface that represents the cell voltage. Nevertheless, the applied
D˜ 0 ⎡⎢ exp (−n2π 2tˆ) ⎤⎥

current still corresponds to JA despite being at the other side of the 1
material, and the diffusion problem can be solved similarly. In a Δμˆ1D = tˆ + −2∑ . [A·21]
D˜ ⎢⎣ 3 n2π 2 ⎥⎦
mixed case (ti ≈ te), however, the flux is no longer confined to one n=1
interface, making the diffusion problem more complex. In this mixed
For a cylinder (infinite length but finite radius L), the corre-
case, the configuration must be modified to that of Fig. A·1a to apply
sponding solution is:
the diffusion analysis.

A.1.2. Derivation of the chemical diffusion equation.—The D˜ 0 ⎡⎢ exp (−αn2 tˆ) ⎤



1 ⎥,
chemical diffusion process described in Eq. A·3 is based on Δμˆ 2D = 2tˆ + −2∑ [A·22]
continuity of the mobile species and preservation of local charge D ⎢⎣
˜ 4 n=1
αn2 ⎥⎦
density. If both negative and positive species change locally by equal
amounts, the local charge density remains unchanged: where αn (n = 1, 2, ⋯ ) are the positive roots of J1(αn) = 0, and J1 is
a Bessel function.
z ∇J Az + = ∇Je−. [A·16] For a sphere:
From Eqs. 1–A, this condition is equivalent to:
D˜ 0 ⎡⎢ exp (−βn2 tˆ) ⎤

1 ⎥,
1 Δμˆ 3D = 3tˆ + −2∑ [A·23]
L ii ∇ μA = (zL ii + L ee)∇2 μ˜ e− .
2
[A·17] D˜ ⎢⎣ 5 βn2 ⎥⎦
z n=1

The continuity of the charged species requires: where βn (n = 1, 2, ⋯ ) are the positive roots of tan βn = βn .
The solution for relaxation in the corresponding systems can also
∂ c Az + be expressed with similar series solutions. Relaxation in a finite
= −∇J Az +. [A·18]
∂t planar 1D system following a galvanostatic pulse of duration τ̂ is:
The left hand side can be expressed in terms of μA:
D˜ 0 ⎡ ⎧ exp (−n2π 2 [tˆrelax + τˆ])
∂ c Az + ∂c ∂c ∂μ Δμˆ1D = ⎢tˆrelax − 2 ∑∞ ⎨
= A = A A, [A·19] D˜ ⎣ n = 1
⎩ n2π 2
∂t ∂t ∂μA ∂t
exp (−n2π 2tˆrelax ) ⎫ ⎤
while the right hand side can also be expressed in terms of μA using − ⎬ ⎥. [A·24]
n2π 2 ⎭⎦
Eq. A·1 and A·17:
1 L Similarly, for cylinder and spherical,
−∇J Az + = − ∇Je− = ee ∇2 μ˜ e−
z z
D˜ 0 ⎡⎢ ˆ ⎧ exp (−α 2 [tˆrelax + τˆ])
2t relax − 2 ∑n = 1 ⎨
L ee L ii ∞ n
= ∇2 μA . [A·20] Δμˆ 2D =
L ee + z 2L ii D ⎢⎣
˜ ⎩ αn2

exp (−αn2 tˆrelax ) ⎫ ⎤


Combining Eqs. 19–A, the governing equation for neutral species A − ⎬ ⎥, [A·25]
is obtained as Eq. A·3 with Eq. A·4 for D̃. αn2 ⎭ ⎥⎦
Journal of The Electrochemical Society, 2021 168 120504

D c ∂μ
D˜ = i i A . [A·28]
RT ∂cA
∂μA
We emphasize that, even in this ion-limited case, the factor ∂cA
is
z+
still that of the neutral species A rather than that of ions A (e.g.
∂μ Az +
∂ci
): a commonly made error in the literature.

A.2. Time Requirements in the Hebb-Wagner Depolarization


Analysis.—It is clearly recognized in the literature that the analysis
using ln (V − Veq ) vs trelax requires that the preceding chemical
polarization time is long enough such that τˆ = τD˜ L2 is large.
However, the specific criterion or the impact of not fulfilling this
condition is not widely recognized. Observation of linearity in
ln (V − Veq ) vs trelax is sometimes incorrectly considered as valida-
tion. Here, we briefly illustrate the pitfall of this perception with an
example shown in Fig. A·2. In the regime where larger τD˜ L2 is
accessible, it is seen that Veq is obtainable from a relaxation period
similar to the pulse duration τ, leading to a reasonable estimation of
D̃ (Fig. A·2a). When τD˜ L2 is small, either due to slow diffusion or a
large sample, the measured V is still significantly different than Veq,
even after 10 times the pulse period (Fig. A·2b). This inaccuracy in
Veq leads to an overestimation in D̃ by several factors (or, even
orders of magnitude in general), despite the apparent linearity in a
Figure A·2. Impact of polarization time on the depolarization analysis. ln (V − Veq ) vs trelax plot. If one can afford to have the sample relax
(a) Relaxation curve in a finite-size planar 1D system after a galvanostatic for a much longer time until the true Veq is reached, in principle, the
pulse of τ̂ = 0.5. After relaxing for a period equal to the pulse duration, the
cell potential reverts reasonably close to the equilibrium value. The
correct slope for diffusion can be recovered; however, such analysis
logarithm voltage plot using the voltage at trelax = 0.5 yields a linear region would require the accurate determination of extremely small values
with a slope similar to that obtained by using the exact equilibrium value. for ln (V − Veq ) (e.g., -9 or smaller) at relaxation times more than 100
(b) Relaxation curve after a galvanostatic pulse of τ̂ = 0.005. Even after times longer than τ. Therefore, this analysis scheme using Veq is
relaxing for 10 times the pulse duration, the voltage is discrepant from the impractical, if not impossible, for small τD˜ L2 values.
true equilibrium value by many factors, although the absolute difference One quick way to recognize the reliability problem from an
between the two is small (and possibly non-distinguishable in experiments).
By using the measured voltage at the end of the relaxation curve does yield a insufficient range of τD˜ L2 values is to find the y-intercept from the
linear region (turquoise solid curve, bottom panel of (b)), but the slope is linear region in ln (V − Veq ) vs trelax. If τD˜ L2 is sufficiently large, the
much larger than the expected slope π2. Even by hypothetically assuming intercept should be close to -1 (or smaller in absolute value). Notice
that one can resolve the true equilibrium value Veq, the slope (black dashed in the example of Fig. A·2b, the intercept is ≈ − 3.7.
line, bottom panel of (b)) still overestimates D̃ by a factor of 2. Time scales
are normalized such that tD˜ L2 = 1. ΔV is normalized such that a pulse of ORCID
τ̂ = 1 induces a change in Veq by 1.
Stephen Dongmin Kang https://orcid.org/0000-0002-7491-7933
William C. Chueh https://orcid.org/0000-0002-7066-3470
and
References
⎡ ⎧ exp (−β 2 [tˆrelax + τˆ])

Δμˆ 3D = 0 ⎢3tˆrelax − 2 ∑n = 1 ⎨

∞ n
1. W. Weppner and R. A. Huggins, “Determination of the Kinetic Parameters of
D˜ ⎢⎣ ⎩

βn2 Mixed-Conducting Electrodes and Application to the System Li3Sb.”
J. Electrochem. Soc., 124, 1569 (1977).
exp (−βn2 tˆrelax ) ⎫ ⎤ 2. W. Weppner and R. A. Huggins, “Electrochemical investigation of the chemical
⎬ ⎥,

− [A·26] diffusion, partial ionic conductivities, and other kinetic parameters in Li3Sb and
βn2 ⎭ ⎥⎦

Li3Bi.” J. Solid State Chem., 22, 297 (1977).
3. W. Weppner and R. A. Huggins, “Electrochemical Methods for Determining
Kinetic Properties of Solids.” Annual Review of Materials Science, 8, 269 (1978).
respectively. 4. A. Honders and G. H. Broers, “Bounded diffusion in solid solution electrode
powder compacts. Part I. The interfacial impedance of a solid solution electrode
(Mx SSE) in contact with a M+-ion conducting electrolyte.” J. Solid State Ionics,
A.1.4. Chemical diffusivity vs self-diffusivity.—The relation 15, 173 (1985).
between chemical diffusivity D̃ and self-diffusivities of the involved 5. A. Honders, J. M. der Kinderen, A. H. van Heeren, J. H. W. de Wit, and G.
charged species kʼs, Dk, can be obtained by using the Nernst-Einstein H. Broers, “Bounded diffusion in solid solution electrode powder compacts. Part II.
The simultaneous measurement of the chemical diffusion coefficient and the
relation: thermodynamic factor in LixTiS2 and LixCoO2.” J. Solid State Ionics, 15, 265
(1985).
Dk ck σ
= 2k 2, [A·27] 6. J. Park, H. Zhao, S. D. Kang, K. Lim, C.-C. Chen, Y.-S. Yu, R. D. Braatz, D.
RT nk F A. Shapiro, J. Hong, M. F. Toney, M. Z. Bazant, and W. C. Chueh, “Fictitious
phase separation in Li layered oxides driven by electro-autocatalysis.” Nat. Mater.,
20, 991 (2021).
where nk is the charge valency of species k. For the D̃ of neutral 7. A. Van der Ven, “Lithium Diffusion in Layered LixCoO2.” Electrochem. Solid-
species A = Az+ + e−, one can substitute Eq. A·27 for the two State Lett., 3, 301 (1999).
8. A. Van der Ven, G. Ceder, M. Asta, and P. D. Tepesch, “First-principles theory of
species Az+ and e− into Eq. A·4. If σe ? σi (ion-limited), then D̃ of ionic diffusion with nondilute carriers.” Physical Review B, 64, 184307 (2001).
neutral species A can be expressed with the self-diffusivity of only 9. K. Kang and G. Ceder, “Factors that affect Li mobility in layered lithium transition
the ions Az+: metal oxides.” Physical Review B, 74, 094105 (2006).
Journal of The Electrochemical Society, 2021 168 120504

10. S.-L. Wu, W. Zhang, X. Song, A. K. Shukla, G. Liu, V. Battaglia, and determine the diffusion time constant and external resistance from potential step
V. Srinivasan, “High Rate Capability of Li(Ni1/3Mn1/3Co1/3)O2 Electrode for Li- (PITT) experiments.” Journal of Electroanalytical Chemistry, 572, 299 (2004).
Ion Batteries.” J. Electrochem. Soc., 159, A438 (2012). 37. Y.-C. Chien, H. Liu, A. S. Menon, W. R. Brant, D. Brandell, and M. J. Lacey, “A
11. R. Amin and Y.-M. Chiang, “Characterization of Electronic and Ionic Transport in fast alternative to the galvanostatic intermittent titration technique.” Energy (2021).
Li1-xNi0.33Mn0.33Co0.33O2(NMC333) and Li1-xNi0.50Mn0.20Co0.30O2(NMC523) as a 38. J. S. Horner, G. Whang, D. S. Ashby, I. V. Kolesnichenko, T. N. Lambert, B.
Function of Li Content..” J. Electrochem. Soc., 163, A1512 (2016). S. Dunn, A. A. Talin, and S. A. Roberts, “Electrochemical Modeling of GITT
12. P.-C. Tsai, B. Wen, M. Wolfman, M.-J. Choe, M. S. Pan, L. Su, K. Thornton, Measurements for Improved Solid-State Diffusion Coefficient Evaluation.” ACS
J. Cabana, and Y.-M. Chiang, “Single-particle measurements of electrochemical Appl. Energy Mater., 4, 11460 (2021).
kinetics in NMC and NCA cathodes for Li-ion batteries.” Energy & Environmental 39. J. Maier, Physical Chemistry of Ionic Materials: Ions and Electrons in Solids
Science, 11, 860 (2018). (Wiley, New York, NY) (2004).
13. Z. Li, C. Ban, N. A. Chernova, Z. Wu, S. Upreti, A. Dillon, and M. S. Whittingham, 40. I. Yokota, “On the Theory of Mixed Conduction with Special Reference to Conduction
“Towards understanding the rate capability of layered transition metal oxides in Silver Sulfide Group Semiconductors.” J. Phys. Soc. Jpn., 16, 2213 (1961).
LiNiyMnyCo1-2yO2.” Journal of Power Sources, 268, 106 (2014). 41. R. Amin, D. B. Ravnsbæk, and Y.-M. Chiang, “Characterization of Electronic and
14. S. Cui, Y. Wei, T. Liu, W. Deng, Z. Hu, Y. Su, H. Li, M. Li, H. Guo, Y. Duan, Ionic Transport in Li1-xNi0.8Co0.15Al0.05O2 (NCA).” J. Electrochem. Soc., 162,
W. Wang, M. Rao, J. Zheng, X. Wang, and F. Pan, “Optimized Temperature Effect A1163 (2015).
of Li-Ion Diffusion with Layer Distance in Li(NixMnyCoz)O2 Cathode Materials for 42. H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids (Oxford University
High Performance Li-Ion Battery.” Adv. Energy Mater., 6, 1501309 (2015). Press, Oxford) 2nd ed. (1959).
15. D. Westhoff, T. Danner, S. Hein, R. Scurtu, L. Kremer, A. Hoffmann, A. Hilger, 43. G. Assat, D. Foix, C. Delacourt, A. Iadecola, R. Dedryvère, and J.-M. Tarascon,
I. Manke, M. Wohlfahrt-Mehrens, A. Latz, and V. Schmidt, “Analysis of “Fundamental interplay between anionic/cationic redox governing the kinetics and
microstructural effects in multi-layer lithium-ion battery cathodes.” Mater. thermodynamics of lithium-rich cathodes.” Nat. Commun., 8, 2219 (2017).
Charact., 151, 166 (2019). 44. M. A. Cabañero, N. Boaretto, M. Röder, J. Müller, J. Kallo, and A. Latz, “Direct
16. L. Peng, Y. Zhu, U. Khakoo, D. Chen, and G. Yu, “Self-assembled Determination of Diffusion Coefficients in Commercial Li-Ion Batteries.”
LiNi1/3Co1/3Mn1/3O2 nanosheet cathodes with tunable rate capability.” Nano J. Electrochem. Soc., 165, A847 (2018).
Energy, 17, 36 (2015). 45. A. Yaqub, Y.-J. Lee, M. J. Hwang, S. A. Pervez, U. Farooq, J.-H. Choi, D. Kim, H.-
17. H.-J. Noh, S. Youn, C. S. Yoon, and Y.-K. Sun, “Comparison of the structural and Y. Choi, S.-B. Cho, and C.-H. Doh, “Low temperature performance of graphite and
electrochemical properties of layered Li[NixCoyMnz]O2 (x = 1/3, 0.5, 0.6, 0.7, 0.8 LiNi0.6Co0.2Mn0.2O2 electrodes in Li-ion batteries.” J. Mater. Sci., 49, 7707 (2014).
and 0.85) cathode material for lithium-ion batteries.” Journal of Power Sources, 46. D. W. Dees, S. Kawauchi, D. P. Abraham, and J. Prakash, “Analysis of the
233, 121 (2013). Galvanostatic Intermittent Titration Technique (GITT) as applied to a lithium-ion
18. W. Zheng, M. Shui, J. Shu, S. Gao, D. Xu, L. Chen, L. Feng, and Y. Ren, “GITT porous electrode.” Journal of Power Sources, 189, 263 (2009).
studies on oxide cathode LiNi1/3Co1/3Mn1/3O2 synthesized by citric acid assisted 47. J. Cho, J.-G. Lee, B. Kim, T.-G. Kim, J. Kim, and B. Park, “Control of AlPO4-
high-energy ball milling.” Bulletin of Materials Science, 36, 495 (2013). nanoparticle coating on LiCoO2 by using water or ethanol.” Electrochimica Acta,
19. A. V. Ivanishchev, I. A. Bobrikov, I. A. Ivanishcheva, and O. Y. Ivanshina, “Study of 50, 4182 (2005).
structural and electrochemical characteristics of LiNi0.33Mn0.33Co0.33O2 electrode at 48. Y.-M. Choi, S.-I. Pyun, J.-S. Bae, and S.-I. Moon, “Effects of lithium content on the
lithium content variation.” Journal of Electroanalytical Chemistry, 821, 140 (2018). electrochemical lithium intercalation reaction into LiNiO2 and LiCoO2 electrodes.”
20. K. Märker, P. J. Reeves, C. Xu, K. J. Griffith, and C. P. Grey, “Evolution of Journal of Power Sources, 56, 25 (1995).
Structure and Lithium Dynamics in LiNi0.8Mn0.1Co0.1O2 (NMC811) Cathodes 49. Y. You, H.-R. Yao, S. Xin, Y.-X. Yin, T.-T. Zuo, C.-P. Yang, Y.-G. Guo, Y. Cui,
during Electrochemical Cycling.” Chemistry of Materials, 31, 2545 (2019). L.-J. Wan, and J. B. Goodenough, “Subzero-Temperature Cathode for a Sodium-Ion
21. S. Oswald, D. Pritzl, M. Wetjen, and H. A. Gasteiger, “Novel Method for Battery.” Adv. Mater., 28, 7243 (2016).
Monitoring the Electrochemical Capacitance by In Situ Impedance Spectroscopy 50. C. M. Subramaniyam, Z. Tai, N. Mahmood, D. Zhang, H. K. Liu, J.
as Indicator for Particle Cracking of Nickel-Rich NCMs: Part I. Theory and B. Goodenough, and S. X. Dou, “Unlocking the potential of amorphous red
Validation.” J. Electrochem. Soc., 167, 100511 (2020). phosphorus films as a long-term stable negative electrode for lithium batteries.”
22. B. Han, A. V. der Ven, D. Morgan, and G. Ceder, “Electrochemical modeling of Journal of Materials Chemistry A, 5, 1925 (2017).
intercalation processes with phase field models.” Electrochimica Acta, 49, 4691 51. R. Qing, L. Liu, C. Bohling, and W. Sigmund, “Conductivity dependence of lithium
(2004). diffusivity and electrochemical performance for electrospun TiO2 fibers.” Journal
23. S. D. Kang, J. J. Kuo, N. Kapate, J. Hong, J. Park, and W. C. Chueh, “Galvanostatic of Power Sources, 274, 667 (2015).
Intermittent Titration Technique Reinvented: Part II. Experiments.” J. Electrochem. 52. C. Bommier, D. Leonard, Z. Jian, W. F. Stickle, P. A. Greaney, and X. Ji, “New
Soc., 168, 120503 (2021). Paradigms on the Nature of Solid Electrolyte Interphase Formation and Capacity
24. M. Z. Bazant, “Thermodynamic stability of driven open systems and control of Fading of Hard Carbon Anodes in Na-Ion Batteries.” Advanced Materials
phase separation by electroautocatalysis.” Faraday Discussions, 199, 423 (2017). Interfaces, 3, 1600449 (2016).
25. M. D. Levi, G. Salitra, B. Markovsky, H. Teller, D. Aurbach, U. Heider, and 53. E. Talaie, P. Bonnick, X. Sun, Q. Pang, X. Liang, and L. F. Nazar, “Methods and
L. Heider, “Solid-State Electrochemical Kinetics of Li-Ion Intercalation into Protocols for Electrochemical Energy Storage Materials.” Chemistry of Materials,
Li1-xCoO2: Simultaneous Application of Electroanalytical Techniques SSCV, 29, 90 (2016).
PITT, and EIS.” J. Electrochem. Soc., 146, 1279 (1999). 54. K.-H. Chen, K. N. Wood, E. Kazyak, W. S. LePage, A. L. Davis, A. J. Sanchez, and
26. A. Nickol, T. Schied, C. Heubner, M. Schneider, A. Michaelis, M. Bobeth, and N. P. Dasgupta, “Dead lithium: mass transport effects on voltage, capacity, and
G. Cuniberti, “GITT Analysis of Lithium Insertion Cathodes for Determining the failure of lithium metal anodes.” Journal of Materials Chemistry A, 5, 11671
Lithium Diffusion Coefficient at Low Temperature: Challenges and Pitfalls.” (2017).
J. Electrochem. Soc., 167, 090546 (2020). 55. E. Trevisanello, R. Ruess, G. Conforto, F. H. Richter, and J. Janek, “Polycrystalline
27. N. Ding, X. Wang, Y. Hou, S. Wang, X. Li, D. W. H. Fam, Y. Zong, and Z. Liu, and Single Crystalline NCM Cathode MaterialsQuantifying Particle Cracking,
“Rational design of a high-energy LiNi0.8Co0.15Al0.05O2 cathode for Li-ion Active Surface Area, and Lithium Diffusion.” Adv. Energy Mater., 11, 2003400
batteries.” Solid State Ionics, 323, 72 (2018). (2021).
28. C. Hong, Q. Leng, J. Zhu, S. Zheng, H. He, Y. Li, R. Liu, J. Wan, and Y. Yang, 56. S. Saha, G. Assat, M. T. Sougrati, D. Foix, H. Li, J. Vergnet, S. Turi, Y. Ha, W. Yang,
“Revealing the correlation between structural evolution and Li+ diffusion kinetics J. Cabana, G. Rousse, A. M. Abakumov, and J.-M. Tarascon, “Exploring the
of nickel-rich cathode materials in Li-ion batteries.” Journal of Materials Chemistry bottlenecks of anionic redox in Li-rich layered sulfides.” Nat. Energy, 4, 977 (2019).
A, 8, 8540 (2020). 57. W. Yin, A. Grimaud, G. Rousse, A. M. Abakumov, A. Senyshyn, L. Zhang,
29. S. Wi, J. Park, S. Lee, J. Kim, B. Gil, A. J. Yun, Y.-E. Sung, B. Park, and C. Kim, S. Trabesinger, A. Iadecola, D. Foix, D. Giaume, and J.-M. Tarascon, “Structural
“Insights on the delithiation/ lithiation reactions of LiMn0.8Fe0.2PO4 mesocrystals in evolution at the oxidative and reductive limits in the first electrochemical cycle of
Li+ batteries by in situ techniques.” Nano Energy, 39, 371 (2017). Li1.2Ni0.13Mn0.54Co0.13O2.” Nat. Commun., 11, 1252 (2020).
30. S. Tang, Z. Wang, H. Guo, J. Wang, X. Li, and G. Yan, “Systematic parameter 58. Q. Pang, A. Shyamsunder, B. Narayanan, C. Y. Kwok, L. A. Curtiss, and L.
acquisition method for electrochemical model of 4.35 V LiCoO2 batteries.” Solid F. Nazar, “Tuning the electrolyte network structure to invoke quasi-solid state
State Ionics, 343, 115083 (2019). sulfur conversion and suppress lithium dendrite formation in Li-S batteries.” Nat.
31. H. Zhou, F. Xin, B. Pei, and M. S. Whittingham, “What Limits the Capacity of Energy, 3, 783 (2018).
Layered Oxide Cathodes in Lithium Batteries?” ACS Energy Lett., 4, 1902 (2019). 59. D. Kim, J.-M. Lim, Y.-G. Lim, J.-S. Yu, M.-S. Park, M. Cho, and K. Cho, “Design
32. M. Chouchane, E. N. Primo, and A. A. Franco, “Mesoscale Effects in the Extraction of Nickel-rich Layered Oxides Using d Electronic Donor for Redox Reactions.”
of the Solid-State Lithium Diffusion Coefficient Values of Battery Active Chemistry of Materials, 27, 6450 (2015).
Materials: Physical Insights from 3D Modeling.” The Journal of Physical 60. C. J. Wen, B. A. Boukamp, R. A. Huggins, and W. Weppner, “Thermodynamic and
Chemistry Letters, 11, 2775 (2020). Mass Transport Properties of LiAl.” J. Electrochem. Soc., 126, 2258 (1979).
33. E. Deiss, “Spurious chemical diffusion coefficients of Li+ in electrode materials 61. C. Ho, I. D. Raistrick, and R. A. Huggins, “Application of A-C Techniques to the
evaluated with GITT.” Electrochimica Acta, 50, 2927 (2005). Study of Lithium Diffusion in Tungsten Trioxide Thin Films.” J. Electrochem. Soc.,
34. C. Montella, “Discussion of the potential step method for the determination of the 127, 343 (1980).
diffusion coefficients of guest species in host materials Part I. Influence of charge 62. K. Aoki, K. Tokuda, and H. Matsuda, “Theory of linear sweep voltammetry with
transfer kinetics and ohmic potentialdrop.” Journal of Electroanalytical Chemistry, finite diffusion space.” Journal of Electroanalytical Chemistry and Interfacial
518, 61 (2002). Electrochemistry, 146, 417 (1983).
35. C. Montella, “Apparent diffusion coefficient of intercalated species measured with 63. A. J. Bard and L. R. Faulkner, Electrochemical Methods : Fundamentals and
PITT.” Electrochimica Acta, 51, 3102 (2006). Applications (Wiley, New York, NY) (2001).
36. M. A. Vorotyntsev, M. D. Levi, and D. Aurbach, “Spatially limited diffusion 64. M. D. Levi and D. Aurbach, Potentiostatic and Galvanostatic Intermittent Titration
coupled with ohmic potential drop and/or slowinterfacial exchange: a newmethod to Techniques (Wiley , New York, NY) 2nd ed., 913 (2012).

You might also like