Proposal for Oral presentation at the VII WORKSHOP IBERCHIP IWS2001

Title: Authors: Contact: Abstract: The subject of this work is a semiconductor device simulator based on finite differences, to serve as a teaching aid. This software was implemented in Matlab, because of its widespread usage and strong graphic and numerical capabilities. This way the simulator can have an attractive graphic output and a concise, clear source code, encouraging students to experiment, understand and modify the program to suit their research interests. For simplicity and conciseness, analysis was limited to the one-dimensional, steady state case. This allows the simulation of basic devices such as the pndiode, bipolar transistor and two-terminal MOS structure. An overview of basic theory and simulation techniques is presented, and two examples of simulation results are shown, for a pn-diode and a JFET transistor. Simulation of Basic Semiconductor Devices Using Matlab Paulo Tavares, Alberto M. Jorge, Saulo Finco Paulo Tavares (e-mail: ptavares@im.cti.br, phone: +55 19 746-6055)

Simulation of Basic Semiconductor Devices Using Matlab

Paulo Tavares#*, Alberto M. Jorge#, Saulo Finco*
paulo.tavares@ieee.org, martins@fee.unicamp.br, saulo@im.cti.br # Unicamp State University of Campinas, FEEC, DEMIC, Caixa Postal 6101, CEP 13083-970, Campinas SP, Brazil *Centro Tecnolgico para Informtica, IM, LPCI Rodovia D. Pedro I (SP-65), km 143.6, Caixa Postal 6162, CEP 13089-500, CampinasSP, Brazil

Abstract
The subject of this work is a semiconductor device simulator based on finite differences, to serve as a teaching aid. This software was implemented in Matlab, because of its widespread usage and strong graphic and numerical capabilities. This way the simulator can have an attractive graphic output and a concise, clear source code, encouraging students to experiment, understand and modify the program to suit their research interests. For simplicity and conciseness, analysis was limited to the one-dimensional, steady state case. This will allow the simulation of basic devices such as the pndiode, bipolar transistor and two-terminal MOS structure.

as the pn-diode, bipolar transistor and two-terminal MOS structure.

2.

Model for Analysis

In order to perform the analysis, three main equations must be considered: Poisson, continuity and current equations. Their meaning will now be reviewed in more detail.

2.1. Poisson Equation

The Poisson equation relates electric potential with charge density . It can be deduced from one of Maxwells equations, D = , from an equation characterising the electrical behaviour of the material1, D = E , and from the fact that E = . Therefore, Poisson equation is () = / . In semiconductors, electric charge results both from mobile charges (electrons and holes, with concentrations n and p, respectively) and immobile charges (ionised acceptor and donor atoms, with concentrations NA and ND+, respectively). Thus, for semiconductors, 2 = (n p + NA ND+) q / (1)

1.

Introduction

The purpose of this work is the implementation of a semiconductor device simulator based on the finitedifference method, for educational purposes. This kind of simulator obtains the electrical characteristics of a semiconductor device (terminal voltages as a function of current, or vice-versa) from the physical characteristics of the device (dopant concentrations, generation-recombination mechanisms, etc.) The first successful attempts at this technique date back to the sixties [1,2], and thorough surveys can be found on [3,4]. Of course, sophisticated commercial software packages are available. But unfortunately, for pedagogical purposes, these are too expensive and unnecessarily sophisticated. This software was implemented in Matlab. Because of Matlabs widespread usage and strong graphic and numerical capabilities, the simulator can have an attractive graphic output and a concise, clear source code. This will encourage students to experiment, understand and modify the program to suit their research interests. For the sake of simplicity and conciseness, analysis was limited to the one-dimensional, steady state case. This will allow the simulation of basic devices such

2.2. Current Equations

These equations result from a series of hypotheses about the mechanisms for charge transport within a given semiconductor material (see e.g. [5]). A simple form for these equations, known as the drift-diffusion model2, is: Jp = q p p E q Dp p Jn = q n n E + q Dn n where the diffusivity of carriers can be related to mobility according to Einstein equations:
1 2

It assumes that the material is linear and isotropic.

It assumes, namely, small deviations from equilibrium (low fields), unless and D are made field-dependent.

Dp = p q , Dn = n q . Thus, applying E = , kT Jp = q p ( p q p) , Jn = q n ( n +
kT n) q

kT

kT

(2.a) (2.b)

2.3. Continuity Equations

The equations of continuity, one for each type of carrier, positive and negative, state a simple fact: for each infinitesimal volume element, the decrease with time of the amount of carriers within the volume is due to (i) the local difference between the flux of carriers leaving the volume and those entering it, and (ii) the difference between recombined and generated carriers. The rate of decrease with time is c / t dV , where c is a concentration (either n or p). To the first contribution corresponds div f dV (f is the flux of one of the carriers), and to second corresponds (R G) dV. Then, t dV = div f dV + (R G) dV or div f = G R t dV. (3) Equation (3) will be rewritten in terms of density of current J. For holes, Jp = qf , and (3) must be multiplied by q, whereas for electrons Jn = qf , which means multiplying (3) by q. Therefore div Jp = q (G R) q t ,
n t p c c

p n = Nc Nv eEG /kT. The product pn in equilibrium is commonly written ni2. Besides, in equilibrium, in spite of the random thermal motion of carriers, Jp 0, Jn 0. Equilibrium can be disturbed by applying an electric field, by injecting (n p > ni2) carriers or by extracting them (n p < ni2).

2.6. One-dimensional Steady State Model

In steady state all derivatives with respect to time are null. And because its a 1-D model, differential operators reduce to total derivatives in x:
Poisson equation: Continuity equations: Current equations:

''= (n p +NB) (6)

Jp' = q U (7.a) Jn' = q U(7.b) Jp= q p (p' p' VT) (8.a) Jn= q n (n' + n' VT) (8.b)
kT

where NB = NA ND+, U = G R and VT = q .

3.

Numerical solution

3.1. Choice of variables

The procedure consists of solving equations (6), (7) and (8) for variables (, n, p). Often, during calculations these variables are replaced by variables (, n, p), where p = ni e (p )/VT, (9.a) n = ni e ( n)/VT. (9.b) Thus,

(4.a) (4.b)

div Jn = q (G R) + q

2.4. Recombination Model

The simplified Shockley-Read-Hall recombination model [6] will be assumed, i.e. ni2 n p GR= (5) (2 ni + n + p) For a recombination model to be qualitatively correct, the following conditions must be met: injection (ni2 > n p) G R > 0 extraction (ni2 < n p) G R < 0 These ensure that the product np tends to return to its equilibrium value.

2.5. p-n Junction in Equilibrium

A p-n junction is said to be in equilibrium when temperature throughout the material is uniform and there are no external disturbances (such as light or a bias voltage) In this case, the values of n e p vary in such a manner that their product is independent of ND and NA, varying only with temperature, according to (see e.g. [7]):

p = + VT log( p/ni ), (10.a) n = VT log( n/ni ). (10.b) Variables n and p are called quasi-Fermi potentials. It is implicit in the equations above that the relation between potential energy and occupancy of states in the conductance band and in the valence bands (from which n and p result) is determined by Boltzmann statistics (for more details see e.g.[7]), but in the context of this problem, they can be seen simply as a variable substitution. This substitution is useful because n and p vary exponentially with . Because n and p are obtained from the logarithm of n and p, problems due to overflow or underflow become less likely.

3.2. Normalisation
Another way to limit the range of numeric values is to perform a normalisation. There are several

possibilities for this. Is this case we will be using the following: Potentials are divided by VT : , * = / V T p* = p*/ VT n* = n*/ VT Concentrations are divided by N, where N = ni . Lengths are normalised according to x* = x /L where L, the intrinsic Debye length3, is given by

By normalising, NB q *'' VT/L2 = (e* en* ep* e *+ n ) ni i Taking (11) into account results in NB *'' = e* en* ep* e* + n (14) i Because the equation is non-linear in *, it will still be necessary to linearise it and solve it iteratively afterwards. The procedure consists of replacing * by *new = * + * in equation (13). Then, *'' + *'' = e* + * en* ep* e* * + NB ni Now a linear equation can be obtained, by considering e* 1 + *, and so NB *''+*''= e*en*(1+*) ep*e*(1 *) + n i . By grouping the terms in *,

VT (11) q ni Extra care must be used when normalising derivatives... d 1 *' = dx* { VT [x(x*)] }.
L= By applying the chain rule, one obtains

*' = *' =
and therefore

1 d d (x) dx* VT dx L ' VT L VT L VT L2 VT

x(x*)

*'' = *'' = *'' =

d {'[x(x*)]} dx* d d '(x) dx* x(x*) dx

''(x)

3.3. Domain discretisation

The finite-differences method will be used, where differential operators will be approximated in the following manner (h is the grid interval): F(x + h) F(x) F'(x) h or F'(x) and F''(x)
F(x + h) 2F(x) + F(x h) . h2 F(x) F(x h) h

*''(e*n*+ ep**) *= NB e*n* ep**+ *''. ni When discretising the equation, the remaining terms that multiply by * appear, because, according to (12), *i1 2*i +*i+1 *'' h2 and *i1 2*i +*i+1 h2 It is essential to apply the boundary conditions properly. These conditions use potential, although they could also be expressed as electric field values. Note that boundary conditions are imposed to initial values *left and *rigth, but because the system is solved in terms of *, boundary conditions in such systems must be *left = *right = 0. *''

, (12)

Grid interval h must be made small enough to resolve charge variations in space. Therefore, it should be significantly smaller than the Debye length.

3.5. Solving the continuity equations

To begin with, variable substitution of (, n, p) by (, n, p) is performed on the current equations, which will then be substituted in the continuity equations. Let us first consider holes. From equation (9.a), p' ' p' = p , VT which results in Jp = q p [p ' p (p' ')] = q p p p' = q p ni e(p)/VT p'

3.4. Solving the Poisson equation

To begin with, variables (, n, p) will be substituted by (, n, p). Applying (9) in (6) results in

'' = [ni e( n)/VT ni e(p )/VT + NB]

or

NB q '' = (e/VT en/VT ep/VT e/VT+ n ) ni . i

3

Corresponds to the distance over which a small unbalanced charge decays (ref. [8], page 639).

= q p ni e/VT ep/VT p' = q p ni e VT (p'/VT) e (15) = q VT p ni e/VT (ep/VT)' The procedure for electrons is similar, using (9.b),
/VT

4.
p/VT

Simulation Results

A silicon p-n junction at 300K will be considered. Doping levels and the resulting V-I curve are shown below.
10 9 8 7 x 10
17

' n' , n' = n VT and considering

Jn = q n [ n ' + n (' n')] = q n n n' = q n ni e
(n)/VT
concentration [m^-3]

Impurity ions

Ionized acceptors Ionized donors

6 5 4 3 2 1 0 0

n'

= q n ni e/VTen/VTn' = q n ni e/VT VT (n'/VT) en/VT (16) = q VT n ni e/VT (en/VT)' In order to simplify notation in subsequent manipulations, terms with no dependence on p or n will be isolated and an additional variable substitution will be performed, p = ep/VT, n = en/VT, (17) from which the current equations can be rewriteen as Jp = ap(x) p' , (18.a) Jn = an(x) n' , (18.b) where ap(x) = q VT p ni e/VT, (19.a) an(x) = q VT n ni e/VT. (19.b) To substitute the current equations, as rewritten in (18), into the continuity equations (7), it is convenient to notice that the latter present a similar basic form for both electrons and holes: J' = q U [a(x) ']' = q U When discretising this equation, the derivative of will use differences between points in the interleaving mesh, for greater accuracy (refer to [9,5] for details). That is equivalent to consider Ji+ Ji =qU h where, in turn, i+1 i , h i i1 . Ji = ai h If the values of current density are not to be stored, the discretisation can be written as Ji+ = ai+
ai+ i+1h i ai i h i1

3 Position [m]

5 x 10

6
-5

14

x 10

-18

p-n junction

12

10

Current [A]

-2 -0.4

-0.3

-0.2

-0.1

0 Voltage [V]

0.1

0.2

0.3

0.4

But even this basic 1-D simulation can be used to obtain insight into the workings of more complex device such as a JFET. By saving all the results obtained during a a voltage sweep, a "1-D" simulation is obtained. Plotting n(x,V) shows the enlargement of the depletion zone under reverse bias:
Electron concentration n(x,V)

x 10 2.5 2 1.5 1 0.5 0 0.4

18

Concentration [m^-3]

0.2 0 -0.2 2 -0.4 0 Position [m] 4

x 10

-5

h or ai+ i+1 (ai+ + ai) i + ai i1 = q U h2 (20)

= qU

Voltage [V]

Now, we can consider that the previous simulation was actually the simulation of half of a symmetric JFET transistor having such a long channel that 2-D effects can be neglected in most of the channel region.

The figure below renders pinch-off quite visible:

x 10
-4

x 10

-4

Electron concentration [m^-3]

Electron concentration [m^-3]

1 1 0.8 0.8

position x [m]

Position [m]

0.6

0.6

0.4 0.4 0.2 0.2 0 0 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 Voltage [V] -0.3 -0.2 -0.1 0 -4.5 -4 -3.5 -3 -2.5 -2 position l [m] -1.5 -1 -0.5 x 10
-7

Note that there is a non-linear relation between voltage drop and distance along the channel, because dVDS = I dR, where 1 dR = dl W q n xmaxn(x)dx xmin If we call f(V) = q n xmaxn(x)dx , then xmin I W dVDS = dl l(V) = Vf(V')dV' W f(V) I 0 This function l(V) is displayed below:
0 -0.5 -1 -1.5 -2 x 10
-7

Note that the values of l would seem to indicate that the initial hypothesis of a long channel is inadequate, but that is not quite true, because the value is relative to W = 1m, I = 1A.

References
[ ] P. Antognetti, G. Massobrio (ed.s), Semiconductor Device Modelling with SPICE, 1988, McGraw-Hill, NewYork [1] H. K. Gummel, A Self-Consistent Iterative Scheme for One-Dimensional Steady State Transistor Calculations, IEEE Transactions on Electron Devices, Oct. 1964, pp. 455-465 [2] G. D. Hachtel, R. C. Joy, J. W. Cooley, A New Efficient One-Dimensional Analysis Program for Junction Device Modelling, Proceedings of the IEEE, Jan. 1972, pp. 86-98 [3] S. Selberherr, Analysis and Simulation of Semiconductor Devices, 1984, Springer Verlag, Wien [4] C. M. Snowden, Introduction to Semiconductor Device Modelling,1986, World Scientific Publishing, Singapore [5] U. Ravaioli, Review of Conventional Semiconductor Models, 1994, online at: http://wwwncce.ceg.uiuc.edu /tutorials/bte_dd/html /bte_dd.html (visited on May 2000) [6] A. M. Jorge, Fsica e Modelos de Componentes Bipolares, 1999, Editora da Unicamp, Campinas Brazil. [7] A. S. Grove, Physics and Technology of Semiconductor Devices, 1967, John Wiley & Sons, New York [8] S. M. Sze, Physics of Semiconductor Devices, 2nd ed., 1981, John Wiley & Sons, New York [9] U. Ravaioli, Introduction to the Numerical Solution of Partial Differential Equations, 1994, online at: http://wwwncce.ceg.uiuc.edu /tutorials/pde/html/pde.html (visited on May 20)

l(V)

l [m]

-2.5 -3 -3.5 -4 -4.5 -5 -1

-0.8

-0.6

-0.4 V [V]

-0.2

0.2

The spatial distribution of n can now be correctly displayed: