Complete electrical and optical simulation of electronic paper

Tom Bert
a,
*
, Herbert De Smet
a
, Filip Beunis
b
, Kristiaan Neyts
b
a
TFCG Microsystems-Elis, Ghent University and IMEC vzw, Technologiepark (building 914), 9052 Zwijnaarde, Belgium
b
Elis-Liquid Crystals and Photonics Group, St Pietersnieuwstraat 41, 9000 Ghent, Belgium
Received 20 April 2005; received in revised form 12 October 2005; accepted 13 October 2005
Available online 8 November 2005
Abstract
A complete model for the internal particle distribution of electrophoretic image displays (EPIDs) is presented. From this model, simulation
formulas for electrical and optical properties are calculated. These formulas incorporate both time dependence and voltage dependence. They
make it possible to optimize display properties since the model is built up with physical parameters. Results are shown that prove that this model is
indeed capable of complete optical and electrical transient simulation.
The importance of centrifugation during display manufacturing is proven as a way to improve the switching speed of the display.
q 2005 Elsevier B.V. All rights reserved.
Keywords: Centrifugation; Electrophoresis; Diffusion
1. Introduction
Electrophoretic image displays (EPIDs) are a thin, exible
reective display type based on the movement of colored
pigments in a colored solvent. Its properties make it the most
likely candidate in the search for a display that combines the
benets of printed paper and an image display [1]. Just like
printed paper it is lightweight, exible and easy to read in all
lighting conditions. Just as an image display its content can be
altered fast and an innite number of times.
In an EPID colored pigments move in a contrasting liquid
through the laws of electrophoresis in an externally applied
eld. When the observer side of the pixel is coated with
pigment, the pixel has the pigment color. When the voltage
switches sign, the pigments move to the opposite side and the
pixel has the liquid color. A matrix of these pixels combined
makes a two color reective display.
2. Problem denition
Electronic paper has a lot of clear advantages. It is bistable:
once attached to an electrode the pigments remain at that
position even when no holding voltage is applied. This is a
great improvement in the eld of power consumption: constant
refreshing of the image is no longer needed.
It is also very thin and lightweight: the liquid is placed
between two ITO-covered sheets, only 100 mm apart. These
sheets can be made of plastic, so the whole display can be
exible. Since, its reective properties are based on the same
technology as paper printing, reection of the surrounding light
on colored pigments, it has a very good reectivity. Even in
bright outdoor environments.
There are however also some major drawbacks:
There is no threshold in the switching properties of the
pigments. The pigments respond to every applied voltage.
Therefore normal passive matrix addressing is not possible.
The optical switching speed is very low. Even with applied
voltages above 10 V, a pixel needs more then 500 ms to
fully reach its contrasting color. This impedes the use of
EPIDs at video rate.
The system inside an electronic paper pixel has a very high
electrochemical complexity. A full model of the internal
particle distribution has not yet been developed. Neither the
optical or electrical response properties are fully under-
stood. Optimization of the switching speed is a pure
chemical and physical process: a good understanding of the
internal physics is necessary to do this.
In this article a solution for the last two problems will be
given while we elaborate on the development of a complete
simulation tool for EPIDs. A method to increase the internal
Displays 27 (2006) 5055
www.elsevier.com/locate/displa
0141-9382/$ - see front matter q 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.displa.2005.10.001
* Corresponding author. Tel.: C32 92645368; fax: C32 92645374.
E-mail address: [email protected] (T. Bert).
eld will be given, this will increase the pigment velocity and
hence the optical response.
3. Display model
Two kinds of charged particles are present inside the pixel
of an EPID [2]
Charged pigments: neutral pigments covered with surfac-
tant ions. The pigments are macromolecules with a radius of
approximately 0.5 mm. After addition of surfactant, a shell
of surfactant ions attaches itself to the pigment. It hereby
acquires a charge of about 3000e (eZ1.60EK19C) which
makes it possible to move the pigment in an electrical eld.
This also avoids occulation of the pigment particles.
Charged inverse micelles: these are formed through
occulation of surfactant molecules. They form ions with
a charge of 1e and a radius that is much smaller than that of
the pigments.
The charged pigments and inverse micelles have opposite
charge: on application of an external eld they move in
opposite directions.
As a rst approximation, we assume that the charged
micelles and pigments follow a normal (Gaussian) distribution
inside the pixel. This is in very good accordance with reality [5]
and greatly improves the ease of computation. A 1D simulation
will be done: particles only move along the x-axis. Since, the
pixel dimension parallel to the eld is much smaller than
perpendicular to the eld, it is safe to make this 1D assumption.
By adjusting the standard deviation it is possible to model a
uniform (large standard deviation) or packed (small standard
deviation) distribution [3,4].
To model diffusion a time-dependent standard deviation is
dened: the charge distribution spreads out with a diffusion
velocity v
diffusion
. The standard deviation s(t) then equals
(v
diffusion
t).
To model drift due to the external eld a time dependent
average value for the normal distribution is dened by means
of a drift velocity v
drift
. The average value then equals (v
drift
t).
A delay time in the particle movement has to be used. This
was proposed by Hopper and Novotny [3,4]. This is used to
model the eld screening due to Poissons law: not all charges
move at the same maximum velocity from the start, this can be
modeled with a normally distributed delay time
Rt
d
Z
1

2p
p
s
delay
e
K1=2 t
d
KT
delay

2
=s
delay

2

(1)
R(t
d
) is the number of particles with a certain delay time t
d
,
T
delay
is the average delay time, s
delay
the standard deviation.
3.1. Particle distribution
The distribution of the charged particles inside the pixel is
modeled as follows:
nx; t Z
N

2p
p
v
diffusion
t
e
K1=2xKv
drift
t
2
=v
diffusion
t
2
(2)
N is the total amount of particles that is present. Micelles and
pigments have the same distribution, however with different
parameters: they are different in number, have different
properties and move in the opposite direction.
The mean deviation of a group of particles with mobility m
spreading out through diffusion (starting from a delta Dirac
distribution) varies as [5].
s Z

2Dt
p
(3)
where the relationship between diffusion coefcient D and
mobility m is given by
D Z
kT
q
m (4)
This means that the diffusion velocity is given by
v
diffusion
t Z
ds
dt
Z
D

2Dt
p (5)
When starting from a distribution with a distribution larger than
0 (no Dirac delta), the following formulas have to be used
st Z

2Dt Ct
0

(6)
and
v
diffusion
t Z
ds
dt
Z
D

2Dt Ct
0

p Z
D

2D t C
s
2
0
2D

(7)
where s
0
is the starting standard deviation (at tZ0).
The relationship between s
0
and V
1
(the voltage applied in
the previous half period) can be approached through
Coulombs law:
E Z
Q
4p3R
2
0s
0
zR Z

Q
p

4p3
p
1

E
p (8)
The charges are pushed apart through Coulomb forces and
pushed together by V
1
. The standard deviation of the
distribution gives an idea of how far apart the charges are.
Therefore:
s
0
z

Q
p

4p3
p

cst
p

V
1
p (9)
where cst is an unknown constant which has meter (m) as a
unit.
Using Poissons law it is possible to obtain how this varying
charge distribution interferes with the internal eld.
dE
dx
Z
r
3
0
dE
dt
dt
dx
z
NQ
3
1
s
2

0
dE
dt

2Dt
p
D

Z
NQ
3
1
2Dt
0
dE
dt
Z
NQ
3
1
2

2D
p
1
t
3=2
(10)
Here, the following approximation was used:
rZ
dQ
dVolume
z
NQ
Dx
2
z
NQ
s
2
:
T. Bert et al. / Displays 27 (2006) 5055 51
This gives the following time dependence of the internal
eld screening due to diffusion:
DEt z
NQ
3
1

2D
p
1
t
1=2
(11)
And when starting from a non-zero distribution:
DEt z
NQ
3
1

2D
p
1
t C
Q

cst
8p3DV
1

1=2
(12)
It is important to notice that the value of this screening eld can
never exceed the value of the externally applied eld. At its
maximum strength the Poisson eld complete blocks the
external eld, resulting into a zero internal eld.
This formula shows how the redistribution of charges leads
to eld screening. Using this formula easy to measure
properties can be calculated: average velocity, delay time and
diffusion velocity
3.1.1. Average velocity
For the average particle velocity we can write (using Eq.
(12)):
v ZmE ZmE
external
KDEt
Zm E
extern
K
NQ

2D
p
3
1

t C
Q

cst
8p3DV
1

: (13)
The external electrical eld is only dependent on the applied
voltage V
2
.
Using the equations derived above and a rst order
approximation we nd that
v
average
Z
m
d
V
2
K
mN
3

4p3Q
cst

V
1

(14)
where d is the thickness of the pixel.
The average particle velocity:
Is both dependent on V
2
and V
1
: it becomes bigger with
increasing V
2
and smaller as a bigger V
1
is applied. V
2
increases the drift velocity, V
1
increases the eld screening.
Becomes lower when more charged particles are present
(bigger N): this is because the eld screening becomes
stronger when more charged particles are present inside the
pixel. The average velocity also drops as the charge of the
particles (Q) increases. This is for the same reasons.
3.1.2. Diffusion velocity
From Eq. (6):
st Z

2D t C
s
2
0
2D

2D t C
Q

cst
8Dp3V
1

D
2

8Dp3V
1
Q

cst

t (15)
Through a rst order approach (last step) we nd
v
diffusion
Z

V
1

4D
2
p3
Q

cst

(16)
Hence, the diffusion velocity:
Is only dependent on V
1
and increases with V
1
. As a bigger
V
1
is applied, the charges are packed more closely together
and they diffuse at a higher velocity.
Increases with increasing mobility: when charges are more
mobile they can get to the uniform distribution faster.
3.1.3. Delay time
The delay time is the time difference between a particle that
moves at its maximum velocity in the externally applied eld
and one that moves in the internal eld that has a lower value
due to eld screening [6].
Dt Z
Ds
v
slow
K
Ds
v
fast
ZDs
1
v
fast
KDv
K
1
v
fast

zDs
1
v
fast
KDv

(17)
Using Eq. (13) we nd n
fast
ZmE
external
Zm(V
2
/d).
From Eq. (12) for tZ0 (since the delay time occurs then) we
nd:
Dv ZmDEt zm
NQ
3
1

2D
p
1
0 C
Q

cst
8p3DV
1

1=2
ZmN

4pQ
3

cst

V
1

This gives for the total delay time, from Eq. (17):
Dt Z
Ds
m
1
V
2
d
KN

4pQ
3

cst

V
1
p
(18)
By using a rst order approximation this becomes:
Dt z
Ds
m
d
2
N

4pQ
3

cst

V
1
p
V
2
2
(19)
Ds is a non-physical length that is used to match a difference in
velocity to a difference in time. It cannot be measured nor has it
a value in the pixel. Its value will have to be matched to t
measurements when optimizing the simulation.
The average delay time
Increases with V
1
: as charges are packed more together it
takes longer for them to loosen. For V
1
Z0 micelles are
T. Bert et al. / Displays 27 (2006) 5055 52
uniformly distributed at the beginning, there is no eld
screening, all charges immediately move at the drift
velocity, there is no delay time.
It decreases with V
2
: as a bigger drift force is applied,
charges loosen more quickly. When an innitely big drift
force is applied, there is no delay time since, the external
eld then overtakes all internal elds.
When calculated, the same voltage dependence for the
standard deviation of the delay time as for the average delay
time is found.
3.2. Optimization and summary
In all the previous formulas more physical properties of the
particles can be incorporated by using the following
relationship between mobility (m), viscosity (h), charge (Q)
and particle size (R):
m Z
Q
6phR
(20)
Together with Eq. (4), this gives for the formulas calculated
above:
Average velocity (from Eq. (14))
v
average
Z
Q
6phR
1
d
V
2
K
N
3

4p3Q
cst

V
1

(21)
Diffusion velocity (from Eq. (16))
v
diffusion
Z

V
1

4p3
Q

cst

kT
6phR
(22)
Average delay time and standard deviation of the delay time
(from Eq. (19)):
Dt Z
6phR
Q
Dsd
2
N

4pQ
3

cte

V
1
p
V
2
2
(23)
Both the screening eld and delay time increase with the
amount of charge that is present (dened by N and Q). It is
therefore important to remove all excess charge that is present
inside the pixel. This can be done by carefully centrifuging the
liquid. The charged pigments are much heavier than the
micelles. They can be separated from the rest of the liquid after
centrifugation. When the supernatant is replaced with fresh,
uncharged solvent, the minimum amount of charge is present in
the liquid: only charged pigments. Then the internal eld
approaches the external eld in the best way and the optical
response is at its fastest. Hence, centrifugation is a simple way
to improve the display response speed. In a previous article we
already proved that centrifugation can also be used to improve
the image quality by reducing pattern formation in the display
liquid [9].
3.3. Current simulation
The distribution of the inverse micelles inside the pixel and
how this distribution varies in time and with the applied
voltages V
1
and V
2
was calculated above. Since, the movement
of charged micelles forms the major contribution to the current,
the voltage and time dependence of the current can be
calculated from the charge movement.
When incorporating a delay time, the formula for micelle
distribution becomes:
nx; t; V
1
; V
2
ZRt
d
; V
1
; V
2
n
micelles
x; tKt
d
; V
1
; V
2
(24)
Due to the forces of drift and diffusion the micelles acquire a
certain time and position dependent velocity. This is given by:
vx; t; V
1
; V
2
Z
dx
dt
Z
dnx; t; V
1
; V
2
=dt
dnx; t; V
1
; V
2
=dx
(25)
The current is given by the total amount of charged particles
(b) (a)
0
0.2
0.4
0.6
0.8
1
0.5 1 1.5 2
Fig. 1. Simulation (Fig. 1(a), left) and measurement (Fig. 1(b), right) of the optical response on application of a voltage step of 20 V.
T. Bert et al. / Displays 27 (2006) 5055 53
that is present inside the pixel, weighted with their velocity:
Jt; V
1
; V
2
Z

d
0
vx; t; V
1
; V
2
nx; t; V
1
; V
2
dx (26)
With this formula the current response in time on application of
a voltage V
2
after a voltage V
1
has been applied can be
calculated.
3.4. Optical simulation
Using the knowledge of the pigment distribution in time and
how it is voltage dependent, a formula to calculate the optical
response in time can be developed. Again, the pigments are
assumed to be normally distributed (as in Eq. (2))
The intensity of the reected light on a particle varies
exponentially with the distance to the observer. When the
pigment distribution and a delay time are incorporated, this
yields for the reected light intensity, hence for the optical
response:
It; V
1
; V
2
ZI
0
s

t
0
Rt
d

d
0
nx; tKt
d
; V
1
; V
2
e
KaadKx
dx

N
d
nx; tKt
d
; V
1
; V
2
dx

dt
d
(27)
I
0
is the input intensity coming from the light source, a is
the extinction coefcient of the medium and a is a
geometrical factor that varies with the incidence angle of
the light and dielectric properties of the medium.
5e08
1e07
1.5e07
2e07
(a) (b)
0.5 1 1.5 2 2.5 3
0.0E+00
2.0E07
4.0E07
1 0 2 3 4
t (s)
C
u
r
r
e
n
t

(
A
)
1V
2V
4V
Fig. 2. Simulation (Fig. 2(a), left) and measurement (Fig. 2(b), right) of the transient current response on application of a voltage block wave with varying amplitude.
5e08
1e07
1.5e07
2e07
2.5e07
(a) (b)
2 4 6 8 10
3.0E08
1.8E07
3.3E07
4.8E07
0 1 2 3 4 5 6 7 8 9 10
t (s)
C
u
r
r
e
n
t

(
A
)
0,5V from -10V
2V from -10V
10V from -10V
Fig. 3. Simulation (Fig. 3(a), left) and measurement (Fig. 3(b), right) of the transient current dependence on V
2
, the applied voltage, with constant V
1
, the previously
applied voltage.
T. Bert et al. / Displays 27 (2006) 5055 54
4. Computer simulation
We have incorporated our calculations in a simulation tool.
Some results are given below.
Fig. 1 shows the agreement between the model and
measurements in optical response. The simulation curve is a
very good match to the measurement. Delay time, response
speed and shape are simulated correctly. The dependence of the
optical response on the physical parameters (charge, viscosity,
.) can be visualized through temperature dependent measure-
ments. When we compare the variation of the optical response
when temperature varies to the theoretical prediction of the
optical response calculating the temperature dependence of the
parameters, the physical correctness of the model is conrmed.
This temperature dependent modeling will be described in a
later article.
Figs. 24 show the agreement between the model and
measurements in electrical response. The dependence on V
1
and
V
2
of the height, shape and position of the current peak is
simulated correctly. The dependence on the amplitude of a block
wave with varying amplitude is correctly simulated by the
model. Again can the model correctly simulate the temperature
dependence (hence parameter dependence) of the electrical
transient responses. This conrms the correctness of the
electrical model and will be elaborated on in a later article.
5. Conclusion
We managed to model the internal physics of electronic
paper. A complete study and explanation for both electrical
and optical properties is given. The formulas contain
physical, measurable parameters.
The importance of the presence of the charged inverse
micelles is shown: removal of these particles by careful
centrifugation is very important in order to maximize the
internal eld strength and hence the pigment velocity.
The pigments are much heavier than the micelles: through
centrifugation it is possible to separate these heavy particles
from the rest, hereby creating a liquid without contami-
nating ionic particles. Another benet of careful removal of
the contaminating charged particles is the lowering of the
electrical current ow through the pixel. This will improve
the power consumption.
Using this model it will be possible to calculate in advance
the use of other display materials (liquid viscosity, pigment
radius, .) and process parameters to optimize the display
production.
Acknowledgements
This research was nanced by a Phd grant of the Institute for
the Promotion of Innovation through Science and Technology in
Flanders (IWT-Vlaanderen), and partially funded by the Federal
Ofce for Scientic, Technical and Cultural Affairs (OSTC,
Belgium) through the IUAP (phase V) Contract No. P5/11.
References
[1] B. Comiskey, J.D. Albert, H. Yoshizawa, J. Jacobson, An electrophoretic
ink for all-printed reective electronic displays, Letter to Nature, 394, 1998
p. 253.
[2] Bert T., Beunis F., De Smet H., Neyts K., Transient Current Properties in
Electronic Paper, SID-ME 2004 10th Anniversary Meeting, Technical
Presentations, 2004.
[3] M.A. Hopper, V. Novotny, An electrophoretic display, its properties,
model, and addressing, IEEE Transactions on Electron Devices 26 (1979)
1148.
[4] V. Novotny, M.A. Hopper, Optical and electrical characterization of
electrophoretic displays, Journal of The Electrochemical Society 126
(1979) 2211.
[5] C. Colpaert, Geleidingsprocessen in nematische vloeibare kristallen, PhD
Thesis, Ghent University, 1997.
[6] T. Bert, H. De Smet, Pattern formation in electronic paper, Proceedings of
the Society for Information Display 2005 International Symposium, 36,
2005, p. 765.
2e08
4e08
6e08
8e08
1e07
(a) (b)
1.2e07
1.4e07
3.0E08
9.0E08
1.5E07
2.1E07
0 1 2 3 4 5 6 7 8 9 10
t(s)
C
u
r
r
e
n
t

(
A
)
2V from -2V
2V from -5V
2V from -8V
2 4 6 8 10
Fig. 4. Simulation (Fig. 4(a), left) and measurement (Fig. 4(b), right) of the transient current dependence on V
1
, the previously applied voltage, with constant V
2
, the
applied voltage.
T. Bert et al. / Displays 27 (2006) 5055 55