Proceedings of the 13th WSEAS International Conference on COMPUTERS

Numerical Modeling of Neural Switch Programming Field Using Finite

Element Analysis
MAHMOUD Z. ISKANDARANI
Faculty of Science and Information Technology
Al-Zaytoonah Private University of Jordan
P.O.BOX 911597, Post Code: 11191
JORDAN
m_iskandarani@yahoo.com

Abstract: - The design and numerical modeling using Finite Element Analysis (FEA) of electric field strength
in programmable neural switch is carried out. The obtained model provided good approximation to the derived
complex analytical solution. Effect of electrode separation and field spread in both x and y directions are
studied and explained. Boundary effect on field strength representation is discussed and numerically reduced
through increasing the number of nodes for each element in the finite grid. Edge effect on field strength is also
eliminated using semi-infinite coplanar electrode approximation.

Key-Words: - Neural, Numerical Modeling, Finite Element, Synapse, Memory, Information Processing

1 Introduction excellent chemical and thermal stability and has

The architecture of the brain can be viewed with
greater or less granularity. Although understanding
of lower-level processing is relevant to higher
levels, it is possible to begin understanding the
higher levels without a complete picture of the lower
ones. At the level of single neurons, the several ion
species present are assumed to obey a set of partial
differential equations. Such equations can be
numerically modeled using Finite Element Analysis,
which proved to be a good approximation to the
complex analytical solution and a fast tool for
validation of the analytical model and future
variations [1, 2, 3, 4].
In this paper, the programming field of a polymer
ability to switch between two states, namely
conducting (oxidized) and insulating (reduced),
similar to the real neuron.
Combining these polymers produced an ionic-
electronic interacting mixed conductor. The
resulting device was tested and proved to be able to
store the injected charge, affecting its conduction
and switches states due to device resistance change.
Such a change would be stable occupying a
specified level until another charge is injected or
removed [10, 11, 12, 13].
Ag Gate

based neural switch device is numerically modeled

using various types of elements. Effect of electrode
separation and distance from the origin are analyzed. PEO10LiCF3SO3

Mathematical representation of FEA is illustrated Au Source

PPY Film
Au Drain

and discussed. SiO2 Substrate

Fig.1: Neural Switch Structure

2 The Modeled Neural Switch
Figure 1 illustrates the designed and modeled neural
switch, which is made as realistic in construction as
3 Numerical Modeling
The used numerical model approximates an
possible using polyethylene oxide - lithium
analytical solution of the neural switch described in
trifluoromethane sulphonate (PEO10LiCF3SO3),
equation 1 using semi-infinite co-planar electrode
which is known as material for high energy, long
approximation with SCHWARZ-CHRISTOFFEL
life, flexible, and re-chargeable just like a neural
transformation as illustrated in figure 2.
switch and has good mechanical and stable
A check is carried out that covers model continuity,
electrochemical properties [5, 6, 7, 8, 9]. The
distortion index, element connectivity, and
material is used together with a highly conducting
orientation, in conjunction with the properties of the
polymer, called polypyrrole (ppy) that possess

ISSN: 1790-5109 227 ISBN: 978-960-474-099-4

 Proceedings of the 13th WSEAS International Conference on COMPUTERS

polymeric material used and analytical solution β,x

boundary conditions to ensure that the model is { } = [B] {β e} (5)
physically and numerically correct [18, 19, 20].
β, y

Where [B] is given by:

dw d ⎛ Vo -1 ⎛ z ⎞ ⎞
|E|= E = = ⎜⎜ cosh ⎜ ⎟ ⎟⎟
dz dz ⎝ π ⎝ d ⎠⎠ ⎡ N1 , x, N 2 , x N3 , x N 4 , x ⎤
⎢ ,y ,y ,y ,y⎥ (5.1)
⎣ N1 N 2 N3 N 4 ⎦
v
= (5.1) is the displacement matrix where, x and y are
π (z2 - d 2 )1 / 2 given by:

4 4
x = ∑ N i X i , y = ∑ N i Yi
Z-Plane Y (Im)
(5.2)
i =1 i =1

I K N L
X (Re)
Now, the element characteristic matrix [K], with
2d material property q and element thickness t is given
M
J by:
-Π

Fig. 2: Geometrical Construction. [K] = ∫ ∫ [B ]T q[B] tdxdy

(6)
3.1 Four-Node Plane Isoparametric Element 1 1
= ∫ ∫ [B ]T q[B] tJdudv
Let the field quantity be β = β(x,y) = β(u,v), where u -1 -1
and v are dimensionless cartesian co-ordinates used
to generate elements in the (x,y) plane. If an Where; J is the Jacobian of the transformation from u-
element has side lengths of 2a and 2b then u and v v plane to x-y plane.
are given by:
In equation (6) second form, [B] is a function of u and
x y v and β is expressed in terms of u and v so equation
u= ,v= (2) (5) is re-written as:
a b
β,u
For a four-node element, β is interpolated from nodal { } = [D N ] {β e} (7)
values {βe} = [β1,β2,β3,β4]T, β,v

4 Where,
β = [N] β e or β = ∑ Ni β i (3)
i =1
⎡ N1 u, N 2 , u N3 , u N4 , u ⎤
[D N ] = ⎢ ⎥ (7.1)
Where Ni is the interpolation or shape function given ⎣ N1 , v N 2 , v N3 , v N 4 , v⎦
by:
From equation (3):
1 1 (4.1)
N1 = (1 - u) (1 - v), N2 = (1 + u) (1 - v)
4 4 1 1
1 1 (4.2) N1 , u = - (1 - v), N1 , v = - (1 - u) (7.2)
N3 = (1 + u) (1 + v), N 4 = (1 - u) (1 + v) 4 4
4 4
And so on.
Where u = + 1, v = + 1
Now, the relationship between (β,u), (β,v) and (β,x),
(β,y) is given by:
For an isoparametric element:

ISSN: 1790-5109 228 ISBN: 978-960-474-099-4

 Proceedings of the 13th WSEAS International Conference on COMPUTERS

β,x β,u β,u 3.2 Eight-node plane isoparametric element

{ } = [J ]-1 { } = [Γ ] { } (8)
β, y β,v β,v By placing an extra node on each side of the four-
node quadrilateral element described previously a
From equations (5), (7) and (8) we obtain: quadratic quadrilateral element is produced. As
with the four-node bilinear element, the sides of the
[B] [ β e] = [Γ] [D N ] [ β e] (9) quadratic element are at u = + 1 and v = + 1. Axes u
and v may be curved in the quadratic element like
From (9): the ones used in the analysis.
The complete set of shape functions is as shown in
[B] = [Γ] [D N ] (9.1) table 1:

Where; 1 1
N1 = (1 - u) (1 - v) - ( N8 + N5)
4 2
Γ11 = u, x, Γ12 = v, x, Γ21 = u, y, Γ 22 = v, y (9.2)
1 1
So, the required JACOBIAN matrix [J] is given by: N2 = (1 + u) (1 - v) - ( N5 + N6)
4 2
⎡ 4 4
⎤ 1 1
⎢ ∑ Ni , ux i ∑ N , uy ⎥ N3 = (1 + u) (1 + v) - ( N6 + N7)
⎡ x, u y, u ⎤ ⎢ i =1 i =1
i i
⎥
(10) 4 2
[J] = ⎢ ⎥ =
⎣ x, v y, v ⎦ ⎢ ⎥
4 4

⎢∑ Ni , Vx i ∑ Ni , Vyi ⎥
⎣ i =1 i =1 ⎦ 1 1
N4 = (1 - u) (1 + v) - ( N7 + N8)
4 2
⎡ x1 y1 ⎤ 1
⎢ ⎥ 2
N5 = (1 - u ) (1 - v)
⎢ x 2 y2 ⎥ 2
= [D N] (10.1)
⎢ x 3 y3 ⎥
⎢ ⎥
⎢⎣ x 4 y4 ⎥⎦ 1
N6 = (1 + u) (1 - v2)
2
Where; DN is given by: 1 2
N7 = (1 - u ) (1 + v)
2
1 ⎡- (1 - v) (1 - v) (1 + v) - (1 + v)⎤ (10.2)
[DN] =
4 ⎢⎣- (1 - u) - (1 + u) (1 + u) (1 - u)⎥⎦
1
N8 = (1 - u) (1 - v2)
From equations (9.2) and (10), Γ can be written as: 2
Table 1: Eight Nodes Shape Function
1 ⎡J 22 - J12 ⎤
[Γ] = [J ]-1 = (11)
∝ ⎢⎣ - J 21 J11 ⎥⎦
To determine the characteristic matrix, the same
procedures are followed as before. It is important to
realize that isoparametric elements are geometrically
Where; ∝ = det [J] = (J11 J22 - J21 J12) isotropic, so, the numerical values of coefficients in
[K] do not depend on the order of numbering of
The JACOBIAN is a function of u and v and it is nodes. However, the order must be maintained and
regarded as a scale factor that transforms the area must run counter clockwise if J is not to become
dxdy in the first form of (6) from dudv in the second negative over part or the entire element.
form of (6).
Now all the necessary expressions are developed
and by substituting them into equation (6) the
4 Analysis, Discussion and Conclusion
characteristic matrix of a four-node quadrilateral
element is evaluated using GAUSS When programmed, the bias applied between the
QUADRATURE formula. gate and the shorted source/drain terminal will set
up a field that will oxidize or reduce the polymer,
and hence when the programming voltage is

ISSN: 1790-5109 229 ISBN: 978-960-474-099-4

 Proceedings of the 13th WSEAS International Conference on COMPUTERS

removed and a constant current is applied between Numerically Simulated Electric Field Strength (En) for
the source and drain, the measured conductance will d=7.5 um
be related to the previously applied bias and will be 0.6
fixed until another latching potential is introduced

Electric Field Strength

0.5
[14, 15, 16]. Such actions are related to the 0.4

(mV/um)
En(Y=2.5)
established electric field within the switching device 0.3 En(Y=4.5)

where a slow and gradual (power) decrease in the 0.2

En(Y=6.5)
En(Y=8.5)
programming field value is realized (figure 3) as we 0.1 En(Y=10.5)

travel along the x-axis. This is consistent with all 0

En(Y=12.5)
En(Y=14.5)
field theories and with the geometrical co-planar 0 10 20
electrode construction shown in figure 2. Also, a Travelled Distance in um ( X-axis)
peak point occurred at x=5µm, which corresponds to
the internal edge of the electrode structure as it Fig. 4: Field Characteristics for Neural Switch
equals half the separation between the electrodes (2d
= 10µm) according to figure 2. Such a peak point is Numerically Simulated Electric Field Strength
valid as the field strength at the inner edge is at its (En) for X=0 um

peak, while the one at the far edge is at its lowest. 0.7

Electric Field Strength

This is also consistent with the semi-infinite co- 0.6
planar electrode assumption. A decrease in the field 0.5

(mV/um)
0.4
value as we travel up the y-axis (figure 4) is evident.
d=5 um
0.3 d=7.5um
This is also consistent with field theory and the 0.2
increment of distance away from the field lines 0.1
linking the device structure. 0
0 5 10 15
Further analysis lead to the following conclusions Travelled Distance along Y-axis (um)
regarding the modeled field strength:
(1) As the separation between the electrodes Fig. 5: Effect of Electrode Separation on the
increases, the field value decreases as shown in Programming Field of Neural Switch
figure 5. Numerically Simulated Electric Field Strength (En)
(2) The decrement in the field value as a for X=3 um
function of distance along the y-axis is gradual and 0.7
has a slow, power like characteristics, where no
Electric Field Strength

0.6

peak point is present as the plot considers the effect 0.5

(mV/um)

0.4
of travel along the y-axis as shown in figure 6. 0.3
d=5 um
d=7.5um
(3) Increasing the nodes per modelling element 0.2

in the finite mesh resulted in more uniform field 0.1

0
strength and better representation of the derived 0 5 10 15

analytical solution, as field approximation and Travelled Distance along Y-axis (um)

numerical simulation is affected by the total number

Fig. 6: Effect of Electrode Separation on the
of nodes used in constructing the grid needed for
Programming Field of Neural Switch
analysis. Figure 7 illustrates the reduction of error as
a function of used nodes.
Numerical Simulation
Numerically Simulated Electric Field Strength (En) for
d=5um 0.5

0.7 0.4
Average Error
Electric Field Strength

0.6
En(Y=2.5) 0.3
0.5 En(Y=4.5)
(mV/um)

0.4 En(Y=6.5) 0.2

En(Y=8.5)
0.3 0.1
En(Y=10.5)
0.2 En(Y=12.5)
0.1 En(Y=14.5) 0
0 20 40 60 80 100 120
0
0 5 10 15 20 Percentage nodes used in the simulation
Travelled Distance in um (X-axis)
Fig. 7: Effect of Number of Nodes on modeling
Fig. 3: Field Characteristics for Neural Switch Accuracy

ISSN: 1790-5109 230 ISBN: 978-960-474-099-4

 Proceedings of the 13th WSEAS International Conference on COMPUTERS
References:
[1] Misha I Rabinovich, R D Pinto, Henry D I
Abarbanel, Evren Tumer, Gregg Stiesberg, R
Huerta and Allen I Selverston,” Recovery of
hidden information through synaptic dynamics”,
Network: Comput. Neural Syst. 13 (2002) 487–
501 PII: S0954-898X (02)53151-1
[2] Alexandre Pouget and Peter
Latham,”Digitized Neural Networks: long-term
stability from forgetful Neurons”, nature
neuroscience, volume 5 no 8, august 2002.
[3] Christoph von der Malsburg, “Neural Network
Self-organization II: Perception in Self-
organizing Nervous Networks”, WS 2002/2003,
22nd October 2002
[4] Y. Karklin, M. S. Lewicki, "A hierarchical
Bayesian model for learning non-linear statistical
regularities in non-stationary natural signals",
Neural Computation, 17 (2): 397-423, 2005.
[5] O. Du¨ rr and W. Dieterich, “Coupled ion and
network dynamics in polymer electrolytes:
Monte Carlo study of a lattice model”, Journal
Of Chemical Physics Volume 121, Number 24
22, December 2004.
[6] Anne-Vale´rie G. Ruzette,a Philip P. Soo,
Donald R. Sadoway, and Anne M. Mayesz, “
Melt-Formable Block Copolymer Electrolytes
for Lithium Rechargeable Batteries”, Journal of
The Electrochemical Society, 148 ~6! A537-
A543 -2001.
[7] S. C. Mui,a, P. E. Trapa,a, B. Huang,a P. P.
Soo,a M. I. Lozow,a T. C. Wang,b
R. E. Cohen,b A. N. Mansour,c, S. Mukerjee,d, A.
M. Mayes,a, and D. R. Sadowaya,z,”Block
Copolymer-Templated Nanocomposite
Electrodes for Rechargeable Lithium Batteries”,
Journal of The Electrochemical Society, 149 -12!
A1610-A1615 -2002.
[8] Yosslen Aray, Manuel Marquez, Jesus
Rodrı´guez, David Vega, Yamil Simo´n-Manso,
Santiago Coll, Carlos Gonzalez, and David A.
Weitz, “ Electrostatics for Exploring the Nature
of the Hydrogen Bonding in Polyethylene Oxide
Hydration”, J. Phys. Chem. B, 108, 2418-2424,
2004
[9] M A.V. Shevade, M.A. Ryan, M.L. Homer,
A.M. Manfreda, H. Zhou, K.S, ” Manatt olecular
modeling of polymer composite–analyte
interactions in electronic nose sensors”, Sensors
and Actuators B 93 (2003) 84–91
[10] R. Segev, Y. Shapira, M. Benvenisto, and E.
Ben-Jacob, “Observations and modeling of
synchronized bursting in two dimensional neural
networks,” Phys. Rev. E, vol. 64, 011920, pp 1-
9, 2001.
ISSN: 1790-5109
[11] Buitenweg, J.R., Rutten, W.L., Marani, E.
“Modeled channel distributions explain
extracellular recordings from culture neurons
sealed to microelectrodes,” IEEE Trans Biomed
Eng. 49, 1580-90, 2002.
[12] M. A. Corner, J. van Pelt, P. S. Wolters, R.
E. Baker, and R. Nuytinck, “Physiological
effects of sustained blockade of excitatory
synaptic transmission on spontaneously active
developing neuronal networks - an inquiry into
the reciprocal linkage between intrinsic
biorhythms and neuroplasticity in early ontogeny
,” Neurosci. Biobehav. Rev., vol. 26, pp. 127-
185, 2002.
[13] J. van Pelt, P. S. Wolters, W. L. C. Rutten,
M. A. Corner, P. van Hulten, and G. J. A.
Ramakers, “Spatio-temporal firing in growing
networks cultured on multi-electrode arrays”.
Proc. World Congr. Neuroinformatics, Rattay F
(Ed) Argesim Report no. 20, Argesim/Asim
Vienna pp. 462-467, 2001.
[14] Q Bai, K. D Wise, “Single-unit neural
recording with active microelectrode arrays”,
IEEE Trans. Biomed. Eng., vol. 48, Aug. 2001.
[15] D. Meeker, S. Cao, J.W. Burdick, and R. A.
Andersen, “Rapid plasticity in the parietal reach
region with a brain–computer interface,” Soc.
Neuroscience Abstracts, no. 357.7, 2002.
[16] Mohammad Mojarradi, David Binkley,
Benjamin Blalock, Richard Andersen, Norbert
Ulshoefer, Travis Johnson, Member, and Linda
Del Castillo, “A Miniaturized Neuroprosthesis
Suitable for Implantation Into the Brain”,
Transactions On Neural Systems And
Rehabilitation Engineering, Vol. 11, No. 1,
March 2003.
[17] Kaihsu Tai, Stephen D. Bond, y Hugh R.
MacMillan, y Nathan Andrew Baker,z Michael
Jay Holst, and J. Andrew McCammon, " Finite
Element Simulations of Acetylcholine Diffusion
in Neuromuscular Junctions", Biophysical
Journal Volume 84 April 2003 2234–2241
[18] Volker Scheuss and Erwin Neher,
"Estimating Synaptic Parameters from Mean,
Variance, and Covariance in Trains of Synaptic
Responses", Biophysical Journal Volume 81
October 2001 1970–1989.
[19] Kevin M. Franks, Thomas M. Bartol, Jr.,
and Terrence J. Sejnowski, "A Monte Carlo
Model Reveals Independent Signaling at Central
Glutamatergic Synapses", Biophysical Journal
Volume 83 November 2002 2333–2348
[20] Serdar Kuyucak, Matthew Hoyles, and
Shin-Ho Chung, "Analytical Solutions of
231 ISBN: 978-960-474-099-4
 Proceedings of the 13th WSEAS International Conference on COMPUTERS

Poisson’s Equation for Realistic Geometrical

Shapes of Membrane Ion Channels", Biophysical
Journal Volume 74 January 1998 22–36

ISSN: 1790-5109 232 ISBN: 978-960-474-099-4