The Drift-Diffusion Transport Model and it’s Numerical Solution

A B. Tech Project Report Submitted

in Partial Fulfillment of the Requirements
of the course

PH 499: Project II

by

Sricharan Veggalam
(120121037)
under the guidance of

Prof. Girish Sampath Setlur and

Prof. Amitabh Chatterjee

to the
DEPARTMENT OF PHYSICS
INDIAN INSTITUTE OF TECHNOLOGY GUWAHATI
GUWAHATI - 781039, ASSAM
 CERTIFICATE

This is to certify that the work contained in this thesis entitled “The Drift-Diffusion
Transport Model and it’s Numerical Solution” is a bonafide work of Sricharan
Veggalam (Roll No. 120121037), carried out in the Department of Physics, Indian
Institute of Technology Guwahati under my supervision and that it has not been submitted
elsewhere for a degree.

Supervisor: Prof. Girish Sampath Setlur and

Prof. Amitabh Chatterjee

Assistant Professor,
April, 2016 Dept. of Electronics and Electrical Engg.
Guwahati. Indian Institute of Technology Guwahati.

i
 DECLARATION

This is to declare that the work contained in this thesis entitled “The
Drift-Diffusion Transport Model and it’s Numerical Solution” submitted to the
Indian Institute of Technology Guwahati is a bonafide work carried out by me under the
supervision of Prof. Girish Sampath Setlur and Prof. Amitabh Chatterjee in
the Department of Physics, Indian Institute of Technology Guwahati.

Sricharan Veggalam

Roll No. 120121037,

April, 2016 Department of Physics,
Guwahati. Indian Institute of Technology Guwahati, Assam.

i
 Introduction

The Drift-Diffusion Transport model consists of five basic equation

which are the poisson equation, continuity equations and current density equations as given
below
q(n − p − c)
∇2 ψ = (1.1)

∂n
∇.J~n − q. = q.Rn ; (1.2)
∂t
∂n
∇.J~p + q. = −q.Rp ; (1.3)
∂t
J~n = q.n.µn .E
~ + q.Dn .∇n (1.4)
J~p = q.p.µp .E
~ − q.Dp .∇p (1.5)
Where: C = ND+ − NA− , n - Concentration of electrons, p - Concentration of holes, R -
net recombination rate of carriers, µp,n - mobility of holes and electrons, Dp,n - diffusion
coefficients.
The current density equations can be substituted in to continuity equations and
we get three coupled equations in ψ, n, p. The physical significance of this mathematical
coupling is the fact that carriers perturb the field which causes their motion. The finite
difference discretization of those three coupled equations using scharfetter-gummel scheme
yields the below equations in 2d
 
h k k h k h q(ni,j − pi,j − Ci,j )
.ψi,j−1 + .ψi−1,j −2. + .ψi,j + .ψi+1,j + .ψi,j+1 −h.k. = 0 (1.6)
k h  h k h k  
ψi,j−1 − ψi,j h ψi−1,j − ψi,j k
ni,j−1 .Dn |i,j−1/2 .B . + ni−1,j .Dn |i−1/2,j .B .
 Vt k   V t h 
ψi,j − ψi,j−1 h ψi,j − ψi−1,j k
−ni,j . Dn |i,j−1/2 .B . + Dn |i−1/2,j .B . +
  Vt  k  Vt  h
ψi,j − ψi+1,j k ψi,j − ψi,j+1 h
−ni,j Dn |i+1/2,j .B . + Dn |i,j+1/2 .B . +
 Vt  h  Vt k
ψi+1,j − ψi,j k ψi,j+1 − ψi,j h
ni+1,j .Dn |i+1/2,j .B . + ni,j+1 .Dn |i,j+1/2 .B .
Vt h Vt k
−Ri,j .h.k = 0 (3.15)

1
    
ψi,j−1 − ψi,j h ψi−1,j − ψi,j k
pi,j−1 .Dp |i,j−1/2 .B . + pi−1,j .Dp |i−1/2,j .B .
  Vt k  Vt  h
ψi,j − ψi,j−1 h ψi,j − ψi−1,j k
−pi,j . Dp |i,j−1/2 .B . + Dp |i−1/2,j .B . +
  Vt  k  Vt  h
ψi,j − ψi+1,j k ψi,j − ψi,j+1 h
−pi,j Dp |i+1/2,j .B . + Dp |i,j+1/2 .B . +
 V t  h  V t  k
ψi+1,j − ψi,j k ψi,j+1 − ψi,j h
pi+1,j .Dp |i+1/2,j .B . + pi,j+1 .Dp |i,j+1/2 .B .
Vt h Vt k
−Ri,j .h.k = 0 (3.16)
Where: B(x) is the Bernoulli function defined as
x
B(x) = x
e −1
Here h, k are the dimensions of the mesh in the x-direction and y-direction respec-
tively. The scharfetter-gummel scheme is an optimal scheme for discretization of the basic
semiconductor equations because it assumes that the partial derivative of the potential is
constant between the discrete points(nodes) but for the carrier concentration an equation
was derived which shows that it varies exponentially between nodes. So, this formulation
gives more accurate results than any other scheme for the Drift-Diffusion model. In this
report the all the steps that has to be followed to solve the above equations for a semi-
conductor device in equilibrium and non-equilibrium i.e under applied voltage have been
discussed. Appropriate boundary conditions that are necessary also have been discussed.
At the end, simulation results of pn-junction diode in equilibrium and non-equilibrium
subjected to different boundary conditions are also shown.

2
 Boundary Condition on ψ, n, p
Three boundary one each on ψ, n, p are required to solve the three coupled
equations.The boundary condition on a quantity can be on
• magnitude of the quantity - Dirichlet boundary condition
• derivative of the quantity - Neumann boundary condition
• a combination of the above two - Mixed boundary condition
A semiconductor is generally in contact(connected) with surroundings at some part
of it’s boundary and also some part of the boundary is in non-contact region. These two
boundaries have different boundary conditions which will be discussed below.

2.1 Ideal Boundary Conditions

At the contact the carrier concentrations n, p has to be uniform across the whole
contact and equal to their equilibrium values n = no , p = po . Similarly the potential
ψ at the contacted boundary should be equal to ψo + V where ψo is built-in voltage at
that boundary and V is applied voltage at that boundary. Application of above boundary
conditions imply that the flow lines emanate perpendicular to the contact as shown in fig.1

Fig. 1 Ideal boundary conditions for a device

3
 Since the potential is uniform the electric field at the contact surface can only
be in the direction perpendicular to the surface. Similarly the gradient of n, p should only
be in direction perpendicular to the surface. So the current density which is a combination
of drift and diffusion is also in the direction perpendicular to the surface. All these are
dirichlet boundary conditions.
At the non-contacted surface we have a neumann boundary condition. The gradient
of potential is zero along the direction normal to the surface. This means no electric field
either enters or leave the surface normal to the surface. Also the concentration gradient
of n, p perpendicular to the surface is zero, this evidently means that the carriers cannot
move across the surface which is true since it is a non-contact surface. Equipotential surface
which results due to given ideal boundary conditions is also shown in the example fig1.
Summarizing the ideal boundary conditions we get following
At ideal ohmic contact: n = no p = po ψ = ψo + V
At ideal non-contact: ∇⊥ n = 0 ∇⊥ p = 0 ∇⊥ ψ = 0

2.2 General Boundary Conditions

Ideal boundary conditions can be used for most of the devices particularly which are
of large size(few µm). But in some cases there will be some electric field present across
the device boundary due to the surrounding environment. This results in surface charge
density and some properties of the device like conductance will get modulated.
Consider a semiconductor block as shown below

Fig. 2 semiconductor block

4
 In general at any boundary the condition is on the current density, for example at a
boundary x = 0 in fig2 the condition is as follows
Jn⊥ (0) = qDn ∇⊥ n|0 + qn(0)µn E⊥ (0) (2.1)
This is because we can talk in terms of how much current is leaving the boundary
or entering the boundary perpendicular to the boundary. At contacts we have non-zero
current passing through and at non-contacts there is zero current. Using the conditions
on current density we can derive the conditions on n, p, ψ. The equation for non-contact
boundary at x = 0 in fig2 is
Jn (0) + Jp (0) = 0 (2.2)
It is important to note that the total current should be zero as there can be some
instances where individual currents of electrons and holes is non-zero but they add up to
zero. Below some cases are discussed on how to get the conditions on n, p, ψ.
Case I: When the surface recombination velocity ’s’ is specified at the boundary
x = 0 in fig2
This specification sets Jn⊥ (0) = qs(n(0) − n0 ) (2.3)
i.e. qDn ∇⊥ n|0 + qn(0)µn E⊥ (0) = qs(n(0) − n0 ) (2.4)
Above equation can be manipulated to express either n(0)(preferred at a contact) or
∇⊥ n|0 (preferred at a non-contact) as
(i) n(0) = (sn0 + Dn ∇⊥ n|0 )(s − µn E⊥ (0))−1 - equation preferred at a contact (2.5)
At ideal contact, s → ∞ ⇒ n(0 = n0 )
(ii) ∇⊥ n|0 = s(n(0)−n0 )Dn−1 −n(0)Vt E⊥ (0) - equation preferred at a non-contact (2.6)
At ideal non-contact, s → 0, E⊥ (0) = 0 ⇒ ∇⊥ n|0 = 0
Case II: When Gs (surface generation rate) is specified at the boundary at x = 0
in fig2
This specification sets Jn⊥ (0) = −qGs (2.7)
i.e qDn ∇⊥ n|0 + qn(0)µn E⊥ (0) = −qGs (2.8)
Again above equation can be manipulated to express either n(0) or ∇⊥ n|0 as

5
 n(0) = −(Gs + Dn ∇⊥ n|0 )(µn E⊥ (0))−1 - equation preferred at a contact (2.9)
∇⊥ n|0 = −(Gs Dn−1 + n(0)Vt E⊥ (0)) - equation preferred at a non-contact (2.10)
Conditions on Potential
At a contact with applied voltage ’V ’ at the boundary x = 0 in fig2
ψ = ψ0 + V − rc J⊥ ((0) where rc → contact resistance (2.11)
At non-contact at boundary x = 0 in fig2:
Some times we can have electric field outside and inside the device due to the practical
setup around the device. So the below condition has to be taken care
s E⊥ (0+ ) = ρs (0) + a E⊥ (0− ) (2.12)
ρs → surface charge density, a → permittivity outside the semiconductor device.
E⊥ = −∇⊥ ψ relation can be used in above equation to get the condition on ψ.
At ideal non-contact, we assume a = 0, ρs = 0 ⇒ E⊥ (0+ ) = 0. But this assumption
might not be correct as there can be field passing across the boundary.
E⊥ (0+ ) is needed to determine n(0) or ∇⊥ n|0 , if Gs 6= 0 or 0 < s < ∞ and
E⊥ (0+ ) 6= 0. This can be done using above equation.

6
 Algorithm to solve the
discretized equations
The carrier concentrations in terms of Quasi-Fermi levels and conduction and
valence band edges(Nc and Nv ) including Fermi function and Boltzmann approximation is
given by
 
Ef n − Ec
n = Nc . exp (3.1)
 k.T 
Ev − Ef p
p = Nv . exp (3.2)
k.T
The band edges can be split in to essentially three parts:
Ec = Eco − δEc − q.ψ (3.3)
Ev = Evo + δEv − q.ψ (3.4)
Eco , Evo denote the band edges for pure material, i.e. an intrinsic semiconductor.
δEc and δEv describe shifts of the band edges caused by a non-uniform composition of the
semiconductor under consideration such as dopants. These quantities have to be assumed
to be functions of position, where as Eco and Evo are usually not position dependent. ψ
represents the electrostatic potential as already used in the previous sections. The Fermi
energies Ef n , Ef p are usually separated into two parts:
Ef n = Ei − q.φn (3.5)
Ef p = Ei − q.φp (3.6)
where: φn , φp are the quasi-Fermi potentials of electrons and holes. They describe
deviations of the corresponding distribution function from the equilibrium state.
Substituting the equations (3.3), (3.4), (3.5), (3.6) in (3.1) and (3.2) we get:
   
Ei − Eco + δEco q.(ψ − φn )
n = Nc . exp . exp (3.7)
 kT   kT 
Evo + δEvo − Ei q.(φp − ψ)
p = Nv . exp . exp (3.8)
kT kT

7
3.1 Device in Equilibrium Conditions

For a semiconductor, in the absence of dopants(intrinsic) and external forces Poisson’s

equation (1.1) reduces to the trivial form:
p(ψ = φp = 0) − n(ψ = φn = 0) = 0 (3.9)
Using the expressions just derived for the carrier concentrations we obtain:
   
Ei − Ec Ev − Ei
Nc . exp = Nv . exp (3.10)
kT kT
For many applications it is convenient to define a so-called intrinsic concentration
ni as the geometric average of the carrier concentrations in a semiconductor in equilibrium
√
ni = n.p (3.11)
√
 
Eg
Where: ni = Nc .Nv . exp − (3.12)
2kT
The carrier concentrations can now be written as below:
 
q.(ψ − φn )
n = ni . exp (3.13)
 kT 
q.(φp − ψ)
p = ni . exp (3.14)
kT
Substituting above equations in Poisson’s equation we get:
       
 2 q.ψ −q.φn −q.ψ q.φp
.∇ ψ − ni .(exp . exp − exp . exp )+C = 0 (3.15)
q kT kT kT kT
In the simulation code we use the following values in the kth iteration ! for above equn
k+1 k k+1 k
q.φp
     
 2 k+1 q.ψ −q.φn −q.ψ
.∇ ψ −ni .(exp . exp −exp . exp )+C = 0 (3.16)
q kT kT kT kT
Again we substitute the values of the kth iteration using the equations (3.13) and
(3.14) then we get the following equation:
 2 k+1 ψ k+1 − ψ k ψ k − ψ k+1
∇ ψ −(nk .(exp( )−1)−pk .(exp( )−1))−(nk −pk −C) = 0 (3.17)
q Vt Vt
The exponential function in the above equation can be further approximated to get
the below equation which is called linearized Poisson’s equation:
 2 k+1 nk + pk k+1
.∇ ψ −( .(ψ − ψ k )) − (nk − pk − C) = 0 (3.18)
q Vt
For a semiconductor device in equilibrium the Fermi level of both the electrons and
holes is equal and uniform throughout the device. So, we can choose the Fermi level to be
reference level i.e. Ef n = Ef p = 0. Then we get the equation for the carrier concentration

8
as:
 
q.(ψ)
n = ni . exp (3.19)
 kT 
q.(−ψ)
p = ni . exp (3.20)
kT
which are in terms of the potential. So, substituting these values in the equation
(3.18) and solving that equation iteratively subjected to appropriate boundary conditions
on potential and carrier concentration we get the required ψ, n, p all over the device in
equilibrium.

3.2 Device in Non-Equilibrium Conditions

In equilibrium conditions the values of Jn and Jp is zero across the whole device, this
is the reason we don’t need continuity equations for equilibrium. But in non-equilibrium
the currents densities are non-zero. All the parameters like the field dependent mobilities,
recombination/generation rate has to be taken care.
Gummel’s method gives a nice procedure to convert coupled non-linear equations into
decoupled linear equations. First, we solve the non-linear Poisson’s equation(3.18). The
potential obtained from this solution is substituted into the continuity equations, Which
are now linear, and are solved directly to conclude the iteration step. The result is then
substituted back into Poisson’s equation and the process is repeated until the convergence is
reached. To check for convergence, one can calculate the residuals obtained by positioning
all the terms to the left-hand side of the equations and substituting the variables with
iteration values. For the exact solution, the residuals shoud be zero. Convergence is
assumed when the residuals are smaller than a set tolerance. The rate of convergence of
gummel method is faster when there is little coupling between different equations. The
computational cost of one Gummel iteration is one matrix solution for each carrier type
plus one iterative solution for the linearized Poisson’s equation. Note that in conditions
of equilibrium, only the solution of Poisson’s equation is necessary, since the equilibrium

9
Fermi level is constant and coincides with both quasi-Fermi levels.

Fig. 3 Gummel’s iteration scheme

3.3 Mesh Size and Scaling

The mesh size must be smaller than the Debye length since one has to resolve the
charge variations in space. A simple example is the carrier redistribution at an interface
between two regions with different doping levels. Carriers diffuse into the lower doped
region creating an excess carrier distribution, which, at equilibrium, decays in space down
to the bulk concentration with approximately exponential behavior. The spatial decay

10
constant is the Debye length
r
kB T
LD = , N → doping density (3.21)
q2N
For example in GaAs and Si at room temperature, the Debye length is approximately
400 Åwhen N = 1016 cm−3 and decreases to about only 50 Åwhen N = 1018 cm−3 .
Scaling the variables is very important, it is almost inevitable for every simulator.
Scaling means every variable is divided by a appropriate quantity of same units and the
discretized equations also changes slightly. This is done to make the algorithms more effi-
cient and to avoid numerical overflow and underflow. It is advisable to input the data in
MKS or practical units (the use of centimeters is, for instance, very common in semicon-
ductor practice, instead of meters) and then provide a conversion block before and after
the computation blocks to normalize and denormalize the variables. The most common
scaling factors for the normalization of semiconductor equations are listed in fig4.

Fig. 4 Scaling Factors

11
 Simulation Results
A code in Matlab has been written to simulate the basic device pn-junction diode in
equilibrium and non-equilibrium conditions.

1. Equilibrium Results at Na=1016 and Nd = 1017

2. Non-equilibrium Results

Electron and Hole Densities at Applied Bias = 0.625V

18
Electron & Hole Densities vs Position - at Applied Bias(0.625V)
10

16
10
X: 3.365
Electron & Hole Densities [1/cm3]

14 Y: 2.602e+013
10
X: 3.126
Y: 2.237e+014
12
10

10
10

8
10

6
10 X: 0.1936
Y: 2.25e+004 n X: 6.475
4
10 p Y: 2250

2
10
0 1 2 3 4 5 6 7
x [um]
I-V Characteristics - Total Current vs Applied Voltage

I vs V Plot X: 0.6226
4.5 Y: 4.177

3.5
Total Current Density [Amp/cm2]

2.5

1.5

0.5

-0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
VA [V]

Total Charge Density at Applied Bias = 0.625V

-3
x 10 Total Charge Density vs Position - at Applied Bias(0.625V)
12

10

8
Total Charge Density [C/cm3]

4 X: 3.354
Y: 0.004766

-2

-4 X: 3.322
Y: -0.003202

-6
0 1 2 3 4 5 6 7
x [um]
Electric Field at Applied Bias = 0.625V

4
x 10 Field Profile vs Position - at Applied Bias(0.625V)
0.5

0
X: 3.054 X: 3.432
-0.5 Y: -51.58 Y: -18.2

-1
Electric Field [V/cm]

-1.5

-2 X: 3.322
Y: -1.719e+004

-2.5

-3

-3.5

-4

-4.5
0 1 2 3 4 5 6 7
x [um]
 Conclusion
From this project I understood how to design a simulator for a semiconductor
device using the Drift-Diffusion Model and to apply appropriate boundary conditions to
simulate it’s physical behavior. The Drift-Diffusion Model is widely used model which can
be used to simulate many devices like BJT and MOSFET by including appropriate mobility
and recombination/generation models. But modern devices need even more fundamental
equations like Boltzmann Transport Equation to simulate correctly. The results obtained
from the code were also verified to be correct.

12
References
[1] Analysis and Simulation of Semiconductor Devices by Siegf reid Selberherr

[2] Computational Electronics by Dragica V asileska

[3] Fundamentals of carrier transport by M ark S. Lundstrom

13