Available online at www.sciencedirect.

com

Journal of Power Sources 179 (2008) 793–798

Short communication

Capacity fade analysis of a lithium ion cell

Qi Zhang, Ralph E. White ∗
Department of Chemical Engineering, University of South Carolina, Columbia, SC 29208, USA
Received 3 December 2007; received in revised form 7 January 2008; accepted 8 January 2008
Available online 21 January 2008

Abstract
A physics-based single particle model was used to simulate the life cycling data of a lithium ion cell. The simulation indicates that there are
probably three stages of capacity fade in a lithium ion cell used at low rates. In the first stage, lithium ions are lost to a film formation reaction (e.g.
SEI formation) and, consequently, the cathode becomes less intercalated during cycling. In the second stage, the loss of active cathode material
outpaces the loss of lithium ions and the cathode gradually becomes more intercalated at the end of discharge. The anode is the limiting electrode
in stages one and two and the change in the anode voltage causes the cell to reach end of discharge voltage. In the third stage, the limiting electrode
shifts from the anode to the cathode, and the anode becomes increasingly less discharged at the end of discharge. Thus, more and more “cyclable”
lithium ions are left inside the anode, which causes additional capacity fade.
© 2008 Elsevier B.V. All rights reserved.

Keywords: Life prediction; Limiting electrode

1. Introduction In order to understand how the capacity fade in individual

electrodes would affect the capacity fade at the cell level, we
Secondary lithium ion batteries are being actively researched used a physics-based single particle model [9,10] to analyze the
for the application into aerospace, medical, military and trans- cycling data of a lithium ion cell. In comparison, most empirical
portation applications. Battery life prediction is an important methods such as experimental data extrapolation view the cell
and active research area. A substantial amount of work [1–11] as a whole entity and could not provide us such information.
has gone into understanding the capacity fade and predicting The focus of this work is to show a significant findings based
battery life. For example, Stamps et al. [6] studied the capacity on our data analysis on the cycling data of a lithium ion cell, but
fade of a lithium ion cell over 1600 cycles using a hybrid estima- not to develop a new cell model. Our study reveals the capac-
tion algorithm combining elements of both batch estimation and ity fade of the cell could have three stages under low rate. The
online moving horizon estimation. However, the battery model experimental data show the first two stages, controlled mostly
used in their work was a simplified empirical model which does by loss of active lithium ions and loss of active cathode mate-
not provide insight concerning capacity fade mechanisms. Gang rial, respectively. A third stage with accelerating capacity fade,
et al. [7] used a physics based model to simulate the cycling per- as predicted by our simulation, would happen when the limit-
formance of lithium ion cells. The SEI formation on the anode ing electrode shifts from the anode to the cathode. The detailed
during charge was assumed to account for the capacity fade. discussion on this capacity fade sequence can be found in Sec-
Lee et al. [8] simulated the cycling performance of lithium ion tion 3 of this paper. Although our experimental data have not
cells under low earth orbit conditions using the same model. yet confirmed this three stage capacity fade pattern, Bloom et
Good agreement was achieved between the experimental and al. [4] observed a similar pattern in their cycle life study (Fig. 3
simulated discharge capacities. Other researchers [11] also used in Ref. [4]). However, without further detailed analysis, it is not
equivalent circuit models (ECM) to study the capacity fade of suggested that our findings can be applied directly to their study.
lithium ion cells. To our best knowledge, there has not been any publication
(in terms of simulation) in the literature to show a similar three-
stage capacity fade pattern as ours. We also provide a detailed
∗ Corresponding author. Tel.: +1 803 777 3270; fax: +1 803 777 6769. and plausible explanation on the reasons behind the formation
E-mail addresses: [email protected], [email protected] (R.E. White). of the three-stage pattern. In addition, our findings highlight

0378-7753/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jpowsour.2008.01.028
794 Q. Zhang, R.E. White / Journal of Power Sources 179 (2008) 793–798

Nomenclature
ce Li+ concentration in the electrolyte (mol cm−3 )
cmax maximum Li+ concentration in the particles
(mol cm−3 )
cs Li+ concentration in the solid particles
(mol cm−3 )
c̄s volume averaged Li+ concentration inside spher-
ical particles (mol cm−3 )
Ds solid phase Li+ diffusion coefficient in the parti-
cles (cm2 s−1 )
Iapp applied current (A)
jint intercalation current density (A cm−2 )
k kinetic rate constant
n number of terms in Eq. (5)
Rs radius of the spherical particles (cm)
U open circuit potential of the electrode (V) Fig. 1. The experimental discharge capacities vs. cycles.
V volume of the electrode (cm3 )
Butler Volmer equation for both electrodes:
x the stoichiometric number at the discharged state 
 
x0 the stoichiometric number at the charged state ∂cs,i 1 ∂ 2 ∂cs,i ∂cs,i 
= 2 Ds,i r , −Ds,i = 0,
∂t r ∂r ∂r ∂r r=0
Greek letters 
ε volume fraction of the active material in the elec- ∂cs,i  jint,i
− Ds,i = (1)
trode ∂r r=Rs,i F
λm positive eigen-value determined from
λm = tan(λm ).
Φ electrode potential (V) jint,i = ki ce0.5 (cmax,i − cs,i |r=Rs,i )0.5 cs,i |r=Rs,i 0.5
    
0.5F −0.5F
Subscripts × exp (Φi −Ui ) −exp (Φi −Ui )
RT RT
i positive or negative electrode (2)
neg negative electrode
pos positive electrode
The intercalation current density jint,i is calculated based on
the equivalent electro-active surface area (Ai ) of the electrode
the importance to consider the interaction between individual “i”:
electrodes and the cell in making the cycle life predictions. We Iapp Iapp
jint,i = = (3)
believe these contributions could provide useful information for Ai 3/Rs,i εi Vi
battery researchers to consider in their own studies.
where Vi is the volume of the electrode and εi is the volume
fraction of the active electrode material. The decrease of the
2. Model volume fraction εi could indicate a loss of active material in the
electrode (i.e., through isolation of active material particles from
The experimental discharge capacities are presented in Fig. 1 the electrode).
for a lithium ion cell (1Ah) with a LiNiCoO2 cathode and a The diffusion equation (Eq. (1)) can be further simplified
carbon anode. The discharge current was 8 mA (<C/100). The from a partial differential equation to a differential and algebraic
discharge data were studied using a so-called single particle equation using the pseudo steady state (PSS) method [12,13].
model [9,10]. The model uses the assumption that the behavior The PSS equations for the diffusion in the electrode “i” are:
of the porous electrodes can be represented by spherical particles
of the active materials in the electrodes. The concentration and ∂c̄s,i 3 jint,i
=− (4)
potential distributions in the solution phase are assumed to be ∂t Rs,i F
negligible, which is a reasonable assumption under low rate situ-
ations. Therefore, the usage of the single particle model is limited jint,i Rs,i 2Rs,i jint,i
(cs,i |r=Rs,i − c̄s,i ) = − +
to low current situations. The detailed description and discus- F 5Ds,i Ds,i F
sion of the single particle model can be found in the literature n 
 1+λ2m
(−1)m sin(λm ) e−λm Ds,i t/Rs,i
2 2
[9,10]. ×
The model equations consist of the solid phase diffusion λm
2
m=1
equation (Fick’s diffusion law) in a spherical particle and the (5)
 Q. Zhang, R.E. White / Journal of Power Sources 179 (2008) 793–798 795

where the number of terms “n” in the summation is typically set Table 1
equal to 5; c̄s,i the volume averaged concentration in the spher- Model parameters used in the study
ical particle, and cs,i |r=Rs,i is the concentration at the surface Parameters Anode Cathode
of the particle. The state of charge (SOC) or the stoichiometric Maximum solid phase concentration 3.056E−2 5.156E−2
number of the electrode is defined as (mol cm−3 ), cmax,i
c̄s,i Solid phase diffusion coefficient 1E−11 2E−10
xi = (6) (cm2 s−1 ), Ds,i
cmax,i Rate constant (cm2.5 mol−0.5 s−1 ), ki 1.07E−4 6.36E−5
Molecular weight (g mol−1 ), MWi 72 97.68
The λm values are determined from the eigen-value equation: Particle size (cm), Rs,i 5E−4 5E−4
Electrode volume (cm3 ), Vi 4.1 3.5
λm = tan(λm ) (7) Initial SOC of the electrode, x0,i a 0.823 0.2925
Volume fraction of active material, εi b 0.3749 0.34
3. Results and discussion Electrolyte concentration (mol cm−3 ), ce 1.0E−3
a Initial SOCs for the anode and the cathode would change during cycling.
3.1. Parameter estimation The value for the first cycle is shown in the table.
b The volume fraction of active material for the cathode decreases with cycling.

Three model parameters were found to change with cell The value is shown for the first cycle.
capacity fade when fitting the discharge data. They are the initial
SOCs for the cathode and the anode at the beginning of charge
(x0,pos and x0,neg ), and the volume fraction of the cathode (εpos ).
The volume fraction of the anode does not change much with
cycling. Fig. 2 compares the experimental and simulated dis-
charge curves at selected cycles and the model parameters are
listed in Table 1.
Fig. 3 shows the change of the three model parameters with
cycling. The anode and the cathode become consistently less
charged with cycling (x0,pos increases and x0,neg decreases). It
is more severe for the anode than for the cathode. The volume
fraction of the cathode εpos decreases, indicating that the cathode
loses active material with cycling.
Fig. 4 shows the predicted values of SOCs for the anode
(xneg ) and the cathode (xpos ) at the end of discharge (EOD) for
each cycle. A two-stage pattern is clearly seen in Fig. 4 for the
electrode SOCs. The cathode becomes less intercalated at EOD
in the first stage, then more intercalated in the second stage.
Fig. 3. The change of model parameters with cycles (obtained from fitting the
Although the anode SOC changes in a similar pattern, it must be discharge data for each cycle). Both the anode and the cathode become less
charged, as x0,neg decreases and x0,pos increases. The volume fraction of the
cathode εpos decreases, indicating that the cathode loses some active material
with cycling.

Fig. 2. Comparison of the experimental and simulated discharge curves at Fig. 4. The simulated SOCs for the anode and the cathode at the end of discharge
selected cycles. The symbols are experiment data and the lines are the simulation (EOD) predicted by the model using parameters shown in Fig. 3. A two-stage
predictions. The discharge current was 1 mA. behavior is clearly seen from the figure for the electrode SOCs.
796 Q. Zhang, R.E. White / Journal of Power Sources 179 (2008) 793–798

Table 2
The extrapolation expressions for the parameters shown in Fig. 6
Parameters Expression a b c

x0,pos axb +c 3.656E−5 1.221 0.2928

x0,neg axb + c −4.262E−3 0.7507 0.8267
εpos axb + c −3.468E−4 1.053 0.3406

Fig. 5. The change of stoichiometric windows (or SOCs) for the electrodes with
cycling. The cross marker ( ) indicates the movement for x0,pos and x0,neg and
the circle marker ( ) indicates the movements for xpos and xneg (SOC at the
EOD). Note the cathode OCP is plotted against 1 − x in Li1−x NiCoO2 . Both
electrodes become less charged with cycling. A two-stage pattern is observed
for the electrode SOC at the EOD (refer to Fig. 4 for xpos and xneg ).

noted that the anode SOC at EOD changes on a much smaller Fig. 7. The discharge capacities predicted by the model compared to the one
scale, between 0.042 and 0.045, as compared to the cathode predicted by the empirical expression. A stage of accelerated capacity fade is
predicted by the physics-based model.
SOC at EOD. A detailed discussion on the two-stage pattern is
presented below. Movements of the stoichiometric windows for
the electrodes are summarized in Fig. 5. The predicted discharge capacity as a function of cycle num-
ber is shown in Fig. 7 using the extrapolated model parameters.
A stage of accelerating capacity fade is predicted after about
3.2. Model prediction 390 cycles. Also shown in Fig. 7 is a prediction of the discharge
capacities obtained by using an empirical equation (see Table 3).
The model parameters are extrapolated in Fig. 6 (see Table 2). It should be mentioned that researchers [4] have observed capac-
The assumption is that the capacity fade mechanisms remain ity fade patterns shaped in a similar means to that shown in
unchanged during the period of our prediction. Fig. 7.
Fig. 8 shows the simulated electrode SOCs at the EOD using
the extrapolated model parameters. Please note that the scale for
xneg in Fig. 8 is different from the one used in Fig. 4. A third
stage is observed for the change of xpos and xneg with cycling.
The SOC of the cathode at the EOD becomes mostly fixed at
0.98 in this stage, indicating that the cathode is close to being
fully intercalated. The SOC of the anode, however, increases
dramatically (less discharged) at the EOD, as compared to the
negligible movement in the first two stages. Fig. 9 shows the
movement of the SOCs for the electrodes in all the three stages.
By comparing Figs. 6 and 7, it is found that the appear-
ance of the three stages for the electrode SOCs at the EOD

Table 3
Parameter values for the empirical expression used in Fig. 7
Expression AxB + Cx + D
A −0.00971
B 0.5007
C −8.03 × 10−3
Fig. 6. The extrapolation of the model parameters shown in Fig. 3. The expres- D 0.9946
sions used for the extrapolation are listed in Table 2.
 Q. Zhang, R.E. White / Journal of Power Sources 179 (2008) 793–798 797

film on the anode. The major capacity fade mechanisms in the

cell simulated here are expected to be the loss of lithium ions to
a film formation reaction and the loss of active material in the
cathode.
The loss of lithium ions to a film formation reaction controls
the capacity fade in the first stage, because the film formation
reaction rate is relatively high on a fresh cell. It is estimated that
during the first charge/discharge cycle, about 18% of the charge
capacity is irreversibly lost to the film formation [14]. In our case,
the cathode is charged from Li(1.0) NiCoO2 to Li(0.3) NiCoO2
(x0,pos ≈ 0.3 in Fig. 3) on the first charge. Then the cathode is only
intercalated back to Li(0.89) NiCoO2 at the end of first discharge
(xpos ≈ 0.89 in Fig. 4). The loss ratio is (1–0.89)/(1–0.3) = 0.16.
That is, 16% of the charge capacity (1–0.3) is irreversibly lost
to the film formation for the first cycle, which is in good agree-
Fig. 8. The simulated SOCs for the anode and the cathode at the EOD using ment with the 18% estimation in Ref. [14]. After the first cycle,
extrapolated model parameters shown in Fig. 6. The symbols represent the SOC the side reaction rate gradually decays, because the SEI film on
data plotted in Fig. 4. Please note the scale for xneg is different from the one in the anode partially impedes the further occurrence of the side
Fig. 4. A third stage shown for xpos and xneg is predicted by the model.
reaction. This is supported by the observation that the capacity
fade rate decreases initially, and the decrease of x0,neg slows with
corresponds to the different capacity fade stages shown in our cycling.
discharge capacity predictions. Next, we will analyze the capac- In the first stage of capacity fade, the anode is the limiting
ity fade of the cell based on the information obtained using the electrode. Here the limiting electrode refers to the one which is
physics-based model. fully discharged at the EOD and causes the cell to reach end of
discharge voltage. The circle marker ( ) in Fig. 9 shows clearly
3.3. Capacity fade analysis that the anode is fully discharged at EOD. The cathode, on the
other hand, is not fully intercalated, because some lithium ions
Our study shows that there could be multiple stages in the have been irreversibly consumed by the film formation reaction.
capacity fade of the cell. In this section, the scenarios behind As the loss of lithium ions becomes larger, the cathode is con-
those stages for cell capacity fade are presented. tinuously less intercalated at the EOD, leading to a decreasing
The parameter estimation study shows that the volume frac- xpos as shown in Fig. 8.
tion of the cathode decreases with cycling (εpos in Fig. 3), In the second stage, the controlling capacity fade mecha-
indicating that the cathode loses some of the active material nism shifts from the loss of lithium ions to the loss of active
in the electrode. In addition, it is well known that there are cathode material. As mentioned previously, the film formation
side reactions happening on the anode during charge process, reaction rate decays in the first stage after the SEI film forms
which would consume the cyclable lithium ions to form SEI on the anode, leading to a slower loss of lithium ions. The loss
of active cathode material then gradually outpaces the loss of
lithium ions, which causes the cathode to become more inter-
calated at the EOD (xpos increases in Fig. 8). Fig. 8 shows this
transition point as a minimum in the xpos versus cycles, where
the loss of cathode material starts to outpace the loss of lithium
ions. However, the anode is still the limiting electrode in the sec-
ond stage. Fig. 8 shows that the anode is mostly discharged to
the SOC below 0.05 in the second stage, which can be regarded
as fully discharged at the EOD. Although the SOC of the cath-
ode (xpos ) at the EOD increases with the cycling, it is not the
cathode that causes the cell to reach the EODV in the second
stage.
In the third stage, the capacity fade mechanisms remain the
same as in previous stages. However, the limiting electrode has
been shifted from the anode to the cathode. Fig. 8 shows that
the cathode is nearly fully discharged (to Li1.0 NiCoO2 ) at the
beginning of the third stage. The anode becomes increasingly
less discharged at the EOD. More and more “cyclable” lithium
ions are left inside the anode and cannot be intercalated back
Fig. 9. The simulated SOCs for the electrodes at the EOD using extrapolated into the cathode. The holding capacity of the cathode decreases
model parameters. further as the cathode loses more active material. In return, more
798 Q. Zhang, R.E. White / Journal of Power Sources 179 (2008) 793–798

“cyclable” lithium ions are kept inside the anode at the EOD, Acknowledgement
causing accelerating capacity fade.
We are grateful for the financial support of the project by the
4. Conclusions National Reconnaissance Office (NRO) under contract # NRO-
000-03-C-0122.
A physics-based single particle model is used to study the
cycling data of a lithium ion cell. Parameter estimation technique References
is applied to extract model parameters from the experiment dis-
charge data. The study on the experiment data shows that there [1] R.B. Wright, C.G. Motloch, J.R. Belt, J.P. Christophersen, C.D. Ho, R.A.
are two stages in the cell capacity fade. Model simulation fur- Richardson, I. Bloom, S.A. Jones, V.S. Battaglia, G.L. Henriksen, T. Unkel-
haeuser, D. Ingersoll, H.L. Case, S.A. Rogers, R.A. Sutula, J. Power
ther indicates that there is a third stage of capacity fade yet to Sources 110 (2002) 445.
come where the limiting electrode shifts from the anode to the [2] P. Nelson, I. Bloom, K. Amine, G. Henriksen, J. Power Sources 110 (2002)
cathode. 437.
The loss of lithium ions to a film formation reaction (SEI [3] I. Bloom, S.A. Jones, V.S. Battaglia, G.L. Henriksen, J.P. Christophersen,
formation) and the loss of active cathode material are the major R.B. Wright, C.D. Ho, J.R. Belt, C.G. Motloch, J. Power Sources 124
(2003) 538.
capacity fade mechanisms. In the first stage, the loss of lithium [4] I. Bloom, B.G. Potter, C.S. Johnson, K.L. Gering, J.P. Christophersen, J.
ions to the film formation reaction is the controlling mechanism. Power Sources 155 (2006) 415.
However, the side reaction rate decays after the SEI film forms [5] Q. Zhang, R.E. White, J. Power Sources 173 (2007) 990.
on the anode. The cathode becomes less intercalated at the EOD [6] A.T. Stamps, C.E. Holland, R.E. White, E.P. Gatzke, J. Power Sources 150
as cyclable lithium ions are irreversibly consumed in the film (2005) 229.
[7] N. Gang, R.E. White, B.N. Popov, Electrochim. Acta 51 (2006) 2012.
formation reaction. The anode is the limiting electrode in the [8] J.W. Lee, Y.K. Anguchamy, B.N. Popov, J. Power Sources 162 (2006)
first stage. In the second stage, the loss of active cathode material 1395.
outpaces the loss of lithium ions because of the decay of the film [9] G. Ning, B.N. Popov, J. Electrochem. Soc. 151 (2004) A1584.
formation reaction rate. The cathode becomes more intercalated [10] S. Santhanagopalan, Q. Guo, P. Ramadass, R.E. White, J. Power Sources
at the EOD. The anode is still the limiting electrode. In the third 156 (2006) 620.
[11] B.Y. Liaw, R.G. Jungst, G. Nagasubramanian, H.L. Case, D.H. Doughty,
stage, the limiting electrode shifts from the anode to the cathode. J. Power Sources 140 (2005) 157.
Because the anode could not be fully discharged, “cyclable” [12] Q. Zhang, R.E. White, J. Power Sources 165 (2007) 880.
lithium ions are left inside the anode, causing the accelerating [13] S. Liu, Solid State Ionics 177 (2006) 53.
capacity fade. [14] V. Srinivasan, J. Newman, J. Electrochem. Soc. 151 (2004) A1530.