Sie sind auf Seite 1von 8

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 39, NO.

2, FEBRUARY 1992

33 1

A New Recombination Model for Device Simulation


Including Tunneling
G. A. M. Hurkx, D. B. M. Klaassen, and M. P. G. Knuvers

Abstract-A new recombination model for device simulation


is presented. This model includes both trap-assisted tunneling
(under forward and reverse bias) and band-to-band tunneling
(Zener tunneling). The model is formulated in terms of analytical functions of local variables which makes it easy to implement in a numerical device simulator. The trap-assisted tunneling effect is described by an expression that for weak electric
fields reduces to the conventional Shockley-Read-Hall (SRH)
expression for recombination via traps. Compared to the conventional SRH expression, the proposed model has one extra
physical parameter, vis. the effective mass m*.For m* = 0.25m0
the model correctly describes the experimental observations associated with tunneling, including the distinctly different temperature behavior of trap-assisted tunneling and band-to-band
tunneling. The band-to-band tunneling contribution is found to
be important at room temperature for electric fields larger than
7 X lo5V/cm. It is shown that for dopant concentrations above
5 X 10
or, equivalently, for breakdown voltages below
approximately 5 V, the reverse characteristics are dominated
by band-to-band tunneling.

1. INTRODUCTION
ECENT developments in both bipolar and MOS technologies, such as lateral downscaling, shallow-junction formation, and the use of self-alignment techniques,
have led to an increase in electric field strength around
p-n junctions in these devices. In bipolar transistors it is
particularly the emitter-base junction at the emitter periphery where the maximum electric field can reach values
as high as lo6 V/cm, while in MOS transistors such
strong fields can occur at the drain. In addition, the high
intrinsic-base dopant concentration possible in Si/SiGe/Si
heterojunction bipolar transistors also gives rise to strong
electric fields at the intrinsic emitter-base junction.
It is a well-known fact that in a strong electric field,
tunneling of electrons through the bandgap can significantly contribute to carrier transport in a p-n junction [ 11,
[2]. Both transitions directly from band to band (Zener
tunneling) and transitions via traps (trap-assisted tunneling) can be important. Tunneling not only adversely affects the leakage currents (e.g., the so-called Zener
breakdown) but it can also lead to an anomalously high

10-@
~

NA=~XIO@

\,

N~=10@

Manuscript received February 28, 1991. Part of this work was funded
by ESPRIT Project 2016. The review of this paper was arranged by Associate Editor D. D. Tang.
The authors are with Philips Research Laboratories, 5600 JA Eindhoven,
The Netherlands.
IEEE Log Number 9104679.

nonideal current under forward bias (forward-biased tunneling) [2]-[4]. The latter is shown in Fig. 1 , where the
current at 0.3 V forward bias and at room temperature is
plotted versus the zero-bias depletion layer width for literature data and for our own measurements [3], [5]-[7].
Details of this figure are given in Section IV. Other characteristic features of this high nonideal forward current
are a reduced temperature dependence and a high nonideality factor [3], [4].
For CAD purposes it is of crucial importance that these
effects are properly taken into account in a numerical device simulator. Since these effects can basically be considered as the generation or recombination of electronhole pairs, they must be incorporated into the recombination term in the electron and hole continuity equations.
Existing models for trap-assisted tunneling [2], [3] give a
semi-empirical relation between the current density and a
certain exponential function of the applied voltage. These
models, however, suffer from the following drawbacks:
Since these models describe tunneling by means of a
current density, they are only suitable for post-processing
calculations and cannot be incorporated into the continuity equations.

0018-9383/92$03.00 O 1992 IEEE

332

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 39, NO. 2, FEBRUARY 1992

They only describe the voltage dependence of the


current density, while its magnitude must be obtained
from experiments.
The predicted temperature dependence is too weak
[41.
In this paper we present a recombination model which
takes into account band-to-band tunneling in reverse-bias
and trap-assisted tunneling in both forward and reverse
bias. In situations of a weak electric field (i.e., lowly
doped junctions) the model reduces to the conventional
Shockley-Read-Hall expression for recombination via
traps. In [4], [SI we have established the basic physics
behind the model. In this work we concentrate on the formulation of the model for device simulation purposes and
on the comparison of simulation results with experiments.
In our model the total net recombination rate is given
by
where Rtrapis the contribution of transitions via traps (including the conventional SRH recombination mechanism)
and Rbb,is the band-to-band tunneling contribution. In the
following sections we discuss these two terms in detail,
while in Section IV a comparison is made between simulation results and experiments.
11. MODELING
TRAP-ASSISTED
TUNNELING
The net recombination rate via traps is determined by
the density of carriers captured per unit of time and the
probability per unit of time of emitting a free carrier from
a trap. To obtain an expression for Rtrapwe start with the
following general phenomenological expression for the net
recombination rate resulting from a dynamic balance between the net rate of captured electrons and that of holes.
This expression reads (see, e.g., [9])

where NT is the trap density, while n, and p t are the densities of electrons and holes which have the capture rates
c, and cp, respectively. The quantities e,l and ep are the
respective probabilities per unit of time for the emission
of an electron or a hole. Both the density of captured carriers and the emission probability per unit of time are increased by tunneling.
In a weak electric field the carrier densities at a certain
location in a depletion layer are given by the conventional
density of free carriers in the conduction and valence
bands. However, in a strong electric field the density of
carriers at a certain location within the depletion layer increased due to the finite probability of carriers tunneling
into the gap. For instance an electron at location x l in Fig.
2(a) has a certain probability of tunneling to a trap at x,
where it has a chance of being captured. In highly doped
junctions, which have a narrow depletion layer, the tunneling distances are relatively short, and, hence, this tunneling effect becomes important. In order to obtain an

neutral n

neutral p

(b)
Fig. 2. Energy-band diagram of a depletion layer around a forward-biased
junction (a) and around a trap in a reverse-biased (b) junction. In (a) tunneling of an electron from x, to a trap at location x is indicated. In (b)
tunneling-enhanced emission of an electron from a trap is indicated. The
solid line in (b) denotes the potential well of the trap without Coulomb
interaction and the dashed line with Coulomb interaction.

expression for the tunneling current, in [2], [3] only the


probability of tunneling directly through the depletion
layer, i.e., the transition from x1 = 0 to x = W , is considered. For this reason the temperature dependence of
these models is too weak [4]. In [SI we have derived an
expression for the carrier density in a depletion layer, including the tunneling contribution, from the solution of
the effective-mass Schrodinger equation for a linear potential. For electrons this expression reads

(3)

where Ai is the Airy function. In y = (2qFm*h-2)3, F


is the average electric field and m* is the effective mass.
The first term on the right-hand side of the above expression is the conventional density of electrons in the conduction band, while the second term is the tunneling contribution [SI. The physical meaning of Ai2[y(x xl)]/Ai2 (0) is the probability that an electron at x1 will
tunnel to a trap at x. The integration in (3) is performed
over all locations x1 from which electrons can tunnel to
location x. The value of 6 depends on the relative position
of the trap level and the conduction-band minimum at the
neutral n side. In Fig. 2(a) the trap level at location x is
below the conduction-band minimum and, hence, 6 = 0.
When the trap level is above the conduction-band minimum at the n side, 6 > 0 and tunneling to x can occur
only over a part of the depletion layer. This will be discussed in more detail below. For holes, a similar expression can be derived.

HURKX er al.: A NEW MODEL FOR DEVICE SIMULATION INCLUDING TUNNELING

The emission of electrons and holes from a trap is enhanced by the phonon-assisted tunneling effect (see Fig.
2(b)) [lo], [4]. Instead of thermal emission over the entire
trap depth E, - ET, which is the only escape mechanism
possible in the absence of a field, carriers can also be
emitted by thermal excitation over only a part of the trap
depth (transition P + P in Fig. 2(b)), followed by tunneling through the remaining potential barrier (transition
P -+ P ). Following the approach of Vincent et al. [lo],
the expression for the enhancement of the emission probability is given by an integral over the trap depth of the
product of a Boltzmann factor, which gives the excitation
probability of a carrier at the trap level to an excited level
E, and the tunneling probability at that energy level from
the trap to the band. For electrons the emission probability reads
Ai2(2m*y-2h-2E)

Ai (0)

dE1

(4)
where e,,Ois the emission probability in the absence of an
electric field. Again, the value of A E,, depends on the relative position of the trap level and the conduction-band
minimum at the neutral n side. For the situation sketched
in Fig. 2(b), tunneling at all levels between ET and E, is
possible, so AE, = E, - ET.
In order to make (3) and (4) suitable for implementation
into a numerical device simulator, we must express these
tunneling effects in terms of analytical functions which
depend on local variables only. For a linear potential it
can be shown that both the carrier concentration and the
emission probability are enhanced by the same factor, i.e.
en - ? =en0

r,,+ 1

-rp+i

ep -- p t =
epo

(54
(5b)

where we have introduced the field-effect functions rnand


rp.Following the same derivation used to obtain the conventional SRH expression from ( 2 ) [9], we arrive at

where E = ET - Ei , i.e., the difference between the trap


level and the intrinsic level. The quantities r, and rp are
the recombination lifetimes of electrons and holes, respectively, while nie is the intrinsic carrier concentration.
For weak electric fields, rn,p
<< 1 and (6) reduces to the
conventional SRH recombination formula.
Using the asymptotic behavior of the Airy function
Ai(y)
exp (-(2/3)y3I2), the expressions for r, and
r, can be written as

333

where

The quantity F is the local electric field. Analytical approximations for the integral in (7) are given in the Appendix.
Because the conduction-band minimum E&) and the
valence-band maximum E&) are a function of the position x in the depletion layer, the absolute value of the trap
level ET(x) is also position-dependent. This implies that
also the integration intervals A E,(x) and A Ep(x) are position-dependent. For the determination of these integration intervals we must distinguish between two situations:
For the situation of a trap at location x in Fig. 2(a), which
is important in forward-biased junctions, tunneling can
occur only at an energy level between the local conduction-band minimum E,(x) and the conduction-band minimum at the neutral n side E,,, because below E,, there are
no states available from (and into) which an electron can
tunnel. In the case where the trap level Ej-(x) lies above
E,, (most important in reverse bias, see Fig. 2(b)) the integration interval is the whole trap depth, i.e. , A E,(x) =
E,(x) - ET(x). For holes, a similar criterion holds. The
expression for the integration intervals can be written as

AEn(x) =
=

E&)

Ern,

ET@) 5 Ern

ET@),

ET(x) > Ern

(9a)

and

AEp(x) = Evp
=

> Evp

Ev(x17

ET(x)

ET(x) 5 Evp.

(gb)

For device simulation the quantities E&) , E&), and


ET(x) can easily be determined from the electrostatic potential (i.e., the isrinsic Fermi level), the local value of
the bandgap and E , which is the relative position of the
trap level with respect to the intrinsic Fermi level. At high
dopant concentrations the Fermi level in a neutral region
nearly coincides with the corresponding band edge. For
this reason and because under low and medium fonvard-

bias conditions, where tunneling is important, the quasiFermi levels are approximately constant in the depletion
region, E,, and Eup can be replaced by -q+,(x) and
- qq$(x), respectively. The quantities +,(x) and +p(x) are
the local quasi-Fermi levels of electrons and holes. In reverse bias these levels are not constant but their relative
position with respect to the trap level is such that they do
provide the correct criterion for the integration intervals.
Using the analytical approximations for the integral in
(7), together with (8) and (9), (6) is readily suitable for
incorporation into a numerical device simulator. How-

334

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 39, NO. 2. FEBRUARY 1992

ever, considering the validity of (6), together with (7),


two questions may arise.
The first one concerns the Poole-Frenkel effect, which
is the mechanism whereby in the case of Coulomb interaction between the free carrier and the trap, the effective
trap depth is lowered (see dashed line in Fig. 2(b)). This
is not taken into account in the above expressions. At a
strong electric field this effect is much weaker than the
tunneling effect, but at a weak field it can greatly enhance
the emission probability [lo]. This Poole-Frenkel effect
occurs in processes where the trap is neutral when it is
occupied by a carrier (either a hole or an electron) and
charged when the carrier is being emitted. However, since
a trap is either donor- or acceptor-like only one of the two
field-enhancement factors in (6) increases due to the
Poole-Frenkel effect [9]. The maximum influence of the
Poole-Frenkel effect on Rtrapoccurs when one of the two
field-enhancement factors in (6) becomes so large that Rtrap
is determined solely by the slowest process, i.e., the other
term in the denominator. That term only experiences the
tunneling effect. This implies that the overall influence of
the Poole-Frenkel effect on the net recombination rate is
fairly limited (maximum a factor of 2 for midgap states
and equal lifetimes for electrons and holes).
The second question which may arise concerns the validity of using a linear potential to calculate the tunneling
probability. This can be investigated by replacing the tunneling probability evaluated for a constant field Ai2 [y(x
- xl)]/ A i 2 (0) (or its asymptotic approximation in (7)) by
the WKB expression [ 111

Tt

exp /-2

i:,

IK(X')I

dx'j

(10)

which is valid for an arbitrary potential. In (lo), I K ( x ' ) (


is the absolute value of the wave vector of the carrier in
the gap, which is determined by the actual potential distribution between xI,and x . Fig. 3 shows the numerically
calculated value of rn for a trap in the middle of a forward-biased, linearly graded junction versus the depletion
layer width (i.e., the doping gradient) for three cases:
1) the tunneling probability evaluated for a constant
field is used and for F the local electric field at the
trap is taken (dotted line);
2) the tunneling probability evaluated for a constant
field is used and for F the average electric field in
the depletion layer is taken (dashed line);
3) the tunneling probability as given by (10) is used
(solid line).
From Fig. 3 we observe that the choice of the local electric field gives results which agree better with the WKB
calculations than the results obtained with the average
field. At this point is should be noted that both (7) and
(lo), as well as the expression of Vincent et al. [lo] accounting for the Poole-Frenkel effect, are obtained in a
one-dimensional (1D) approach. A three-dimensional
(3D) numerical treatment of these problems show that a

1014
1012

i
0
c

10'0

*
3
c

108

c
c

106
1 o4
102
0

200

400

600

depletion width (A)

800

1000

Fig. 3 . The field-effect function I', in the case of AE,, = 0.4 eV versus
the depletion-layer width of a forward-biased, linearly graded junction for
two temperatures. The solid lines are obtained by using (10) for the tunneling probability, while the other lines are obtained by using the tunneling
probability for a constant electric field (dashed lines: average field; dotted
lines: local field).

small variation of the effective mass in the 1D expressions


can account for the 3D effects. To account for the abovementioned effects, the value of the effective mass to be
used in (8) is obtained from a comparison of simulations
with experiments. Using the local value of the electric
field in (8), the experimentally obtained value of m* is
0.25mo (see Section IV and Fig. 7), which is quite a plausible one.
111. MODELING
BAND-TO-BAND
TUNNELING
For the band-to-band tunneling contribution Rbbrwe
base ourselves on the theoretical work of Keldysh and
Kane [ 121-[ 141. Since silicon is an indirect semiconductor whose direct bandgap is much larger than its indirect
gap, indirect transitions including electron-phonon interaction are predominant. Keldysh calculated the transition
rate on the basis of a solution of the time-dependent
Schrodinger equation, including electron-phonon interaction. His results were later adopted by Kane to obtain
an expression for the tunneling current density per unit of
energy dJbbr/dE [14]. Both directly from the work of
Keldysh, and from Kane's work by using the relation

the following expression for Rbbtcan be obtained:

In [12]-[14] it can be found that U = 2 for direct transitions and U = 5 / 2 for indirect transitions, including electron-phonon interaction. Since silicon is an indirect semiconductor, we use U = 5/2. In (1 1) rl, is the electrostatic
potential, while in (12) 'E and 'E are the Fermi levels at
the neutral n and p side, respectively. In the above transformation from d Jbbt to Rbbrthe tunneling of an electron
at a certain energy, say El (see Fig. 4), from x1 to x2 is
represented by the generation of an electron-hole pair in

HURKX er a l . : A NEW MODEL FOR DEVICE SIMULATION INCLUDING TUNNELING

s-', respectively [4], [15]. The prefactor B is taken to be


temperature-independent. The quantity F,, which is proportional to
where Eg is the bandgap [12]-[14], depends on the temperature due to the temperature dependence of this bandgap. In order to have an idea of the
electric field strength above which band-to-band tunneling becomes important at room temperature, we compare
Rbbt with the ratio n r e / r ,which is a measure of the generation rate via traps. When we take for this ratio a realistic value (at room temperature) of IO" cmP3 * s-', we
find that band-to-band tunneling becomes important at a
field strength above 7 X lo5 V/cm.

neutral p

.---------

'4 I

neutral n
Et"

_ _ _ _ _ _ _' - - -A
x,

1
xp

Fig. 4 . Schematic energy-band diagram of a reverse-biased p-n junction.


The band-to-band tunneling mechanism is indicated. Band-to-band tunneling is only possible in the region x,, 5 x < x,.

the middle of the gap (xI + x2)/2. The function D(F, E,


Eb,, ESP)accounts for the relative position of the Fermi
levels Efi,and Efp in the neutral regions and for the influence of the motion of the electron perpendicular to the
electric field on the tunneling probability [ 141. In forward
bias, it accounts for the well-known peak in the tunneling
current observed in Esaki diodes. An expression for D(F,
E , Eb,, Eh), which is valid in zero and reverse bias and
which is suitable for implementation in a device simulator
can be obtained from [14]. This gives

1
exp [(-E, - q$)/kTl

1.

335

(13)

This function virtually equals zero for x < x, in Fig. 4,


because in this region there are no final states into which
an electron can tunnel. Forx > xp in Fig. 4, this function
also equals zero because there are no initial states from
which electrons can tunnel. For x, < x < xp or, equivalently, when the tunneling energy E l lies between Ef,and
ESP,this function equals unity. As in the case of trap-assisted tunneling, the quantities Efi,and Eh can be replaced
by -q&(x) and -q4p(x), respectively. However, from
numerical simulations it is found that when the current
density in reverse bias is very high, the above replacement gives incorrect results. The reason for this is that the
finite saturated drift velocity of the carriers causes an increase in the camer densities. When the generated electron and hole densities are in the order of the intrinsic
carrier density, the quasi-Fermi levels lie very close to the
intrinsic level E, (E, = -q$). In that case, the replacement of Ef, and Eh by -q&(x) and -q$p (x)gives a value
of D significantly less than its actual value 1. This can be
remedied simply by putting D = 1 at those mesh points
where the magnitude of the electron or hole current density has increased to a certain fraction (e.g., lop3) of
qnrcUs,where U , is the saturated drift velocity.
The quantities F, and B at room temperature are found
to be 1.9 X lo7 V/cm and 4 X l O I 4 cm-'l2 * Vp5/*

IV. SIMULATION
RESULTSAND A COMPARISON
WITH
EXPERIMENTS
To give an impression of the model behavior, Fig. 5
shows 1D simulations of diodes in reverse and forward
bias. The diodes are step junctions with No = lo2' cmp3,
while N A is varied. In these simulations conventional
models for the mobilities, bandgap narrowing, recombination lifetimes, and impact-ionization rates are used, as
can be found, for instance, in [16]. Furthermore, we have
used E&) = E, (x) (i.e., "midgap" states) and temperature-independent lifetimes. From the reverse characteristics, shown in Fig. 5(a), we can observe that for dopant
concentrations above 5 X lOI7 cmP3or, equivalently, for
breakdown voltages below 4Eg/q-6Eg/q, the reverse
characteristics are dominated by band-to-band tunneling
(Zener tunneling). This is in agreement with the criteria
mentioned in standard textbooks (e.g., [17]). From the
forward characteristics given in Fig. 5(b) we see that the
nonideal current increases significantly for dopant concentrations above a few times 10'' cmp3, which is due to
trap-assisted tunneling. This is in agreement with experimental observations in [3], [5]-[7] and with our own experiments, as will be shown below.
Fig. 6 shows a comparison of simulation results with
measurements on different diodes having linearly graded
junctions. These diodes have a large junction area (204
x 204 pm2), and sidewall effects are eliminated by the
use of guard rings. The junction is formed by the diffusion
of boron into a heavily doped, homogeneous n-type substrate. The doping profiles are determined from C- I/ measurements and from the resistivity of the substrate. Diodes
A , B , and C have a zero-bias depletjon layer width of approximately 200, 270, and 400 A , respectively. The
magnitude of the calculated curves depends on the lifetimes, while both the slope (i.e., the nonideality factor)
and the temperature dependence are given by the effective
mass m*. Since the values of the lifetimes are unknown,
we have taken a constant value for T, = r,, = T for each
diode. For each diode the value of T is chosen such that
at T = 294 K the magnitude of the simulated curve, using
the new model, fits the measurements. The resulting lifetimes are 0.6, 2.5, and 20 ps for diodes A , B , and C ,
which have a substrate doping concentration of around 2
X 1019, 7 X lo1', and 1.9 x 10" ~ m - respectively.
~ ,

IEEE TRANSACTIONS ON ELECTRON DEVICES. VOL. 39, NO. 2 . FEBRUARY 1992

336

100

0.0

15

10

reverse voltage (VI

10-12
0.0

0.6

(a)

0.4

0.2

0.4

forward voltage (V)

(a)

0.0

0.2

0.6

forward voltage ( V I

0.8

1O ~ 2
0.0

0.4

forward voltage (V)

(b)

0.6

(b)

Fig. 5 . One-dimensional simulation results for step junctions (N,] = IO


cm-3 while N A is indicated in the figure) in reverse bias (a) and in forward
bias (b). Tunneling is included. The regions where band-to-band tunneling
predominates are indicated.

From this figure we observe that the nonideal current of


the highest doped diode ( A ) has a much weaker temperature dependence than predicted by the conventional SRH
model, while for the lowest doped diode this difference is
much less pronounced. For diode A the nonideality factor
(i.e., the slope of the low-bias I-V curve) is also, especially at low temperatures, much larger than 2, whereas
the conventional SRH model predicts a value of slightly
less than 2. In [4]this is discussed in greater detail. It is
important to notice that calculations with a different value
of the effective mass agree less well with both the measured nonideality factor and temperature dependence of
these diodes. This is illustrated in Fig. 7 where the calculated forward curve of diode A is given for three values
of the effective mass.
Fig. 8 shows the reverse characteristics of these diodes
at room temperature. The values of the lifetimes that are
used are the same as for the forward-bias calculations. For
these diodes band-to-band tunneling predominates in reverse bias. In order to show the extreme sensitivity of the
band-to-band tunneling current on the electric field, in
Fig. 9 the calculated reverse characteristics of diode B are
shown for three slightly different doping profiles. The
doping profiles differ in such a way that the corresponding
zero-bias depletion capacitance is 10% higher or 10%
lower than the value used to obtain Fig. 8.
Fig. 10 shows the reverse characteristics of a (lower
doped) diode with a linearly graded junction at three temperatures. The zero-bias depletion layer width of this

,I

0.2

0.0

0.2

0.4

forward voltage (V)

0.6

(C)
Fig. 6 . Measurements and simulation results for three diodes in forward
bias at two temperatures. The solid dots are measurements, while the lines
are simulation results with (solid lines) and without (dashed lines) the inclusion of tunneling effects.

0.0

0.2

0.4

forward voltage (V)

0.6

Fig. 7. Forward J-Vcurves of diode A calculated for three different values


of the effective mass and at two temperatures.

diode is 460 A , so the electric field is less than that in


diodes A-C. This can also be observed from the fact that
band-to-band tunneling dominates trap-assisted tunneling

HURKX

CI

al.: A NEW MODEL FOR DEVICE SIMULATION INCLUDING TUNNELING

102
100

NI

10-2

10~4

t
?

10-6

108

reverse voltage (V)

Fig. 8. Measurements (dots) and simulation results with (solid lines) and
without (dashed lines) tunneling for the three diodes of Fig. 6 in reverse
bias and at room temperature.

t
I

10 l o

reverse voltage (V)

Fig. 9. Measurements and simulation results for diode B in reverse bias


for three slightly different doping profiles. The solid dots are measurements. The solid line denotes the same simulation results as given in Fig.
8. The doping profiles corresponding to the dashed lines differ in such a
way that the corresponding zero-bias depletion capacitance is 10% higher
or 10% lower than the value used to obtain the curve in Fig. 8.

loot

10 2

1:T=294K
2: T = 338K
3:T=383K

10.~

t
YI
C

10-6

U
a+

10-8

rn-10

reverse voltage (V)

Fig. 10. Reverse characteristics of a diode with a linearly graded junction


at three temperatures. The dots are measurements, while the lines are simulation results with (solid lines) and without (dashed lines) tunneling. For
this diode impact ionization is not included.

only above 3 V reverse bias. Notice the different temperature dependence of the two regimes.
Finally, we return to Fig. 1, which shows a comparison
between measurements (from [3], [5]-[7] and own measurements) and simulations. In this plot the forward current density at 0.3 V and at room temperature is plotted
versus the zero-bias depletion width. For the data in [3]
and for our own data the values of the zero-bias depletion
layer width are obtained from the zero-bias depletion capacitance. For the other data we have estimated the de-

337

pletion layer width from simulations on junctions with a


similar doping profile. Furthermore, although the junction areas are rather large, it is not explicitly mentioned
in [5]-[7] that sidewall effects do not play a significant
role. This means that, since the current density is obtained
by dividing the current by the junction area, for these data
the values of the current density are somewhat uncertain.
Neverthelessb we can clearly observe that below approximately 300 A zero-bias depletion layer width or, equivalently, above a dopant concentration of a few times lo'*
cmP3 for a steep junction, the nonideal current increases
significantly due to tunneling. When tunneling is included
in the recombination model, this increase is also given by
the simulation results. In Fig. 1 this is shown by solid line
1 , which represents results for a step function (similar to
the results in Fig. 5(b)) and by solid line 2 which are results for an emitter-base profile of a high-frequency process. The dashed line is obtained for step junctions by
using the conventional recombination model without tunneling.
V. SUMMARY
AND CONCLUSIONS
In this paper we have presented a new recombination
model for device simulation which includes both trap-assisted tunneling and band-to-band tunneling (Zener tunneling). The model is formulated in terms of analytical
functions of local variables, which makes it easy to implement in a numerical device simulator. The trap-assisted tunneling effect is described by an expression that
for weak electric fields reduces to the conventional SRH
expression for recombination via traps. Compared with
the conventional SRH expression, the proposed model has
one extra physical parameter, viz. the effective mass m*.
For m* = 0.25mo, which is a quite plausible value, the
model correctly describes the following experimental observations:
1) The weak temperature dependence of the nonideal
forward current in heavily doped junctions.
2) The nonideality factor of such a junction which, especially at low temperatures, has a value significantly
larger than two.
3) The significant increase in the nonideal current for
a diode wit! a zero-bias depletion layer width less than
about 300 A , or, equivalently, above a dopant concentration of a few times 10'' ~ 1 1 1 ~ ~ .
The band-to-band tunneling contribution is found to be
important at room temperature for electric fields larger
than 7 x lo5 V/cm. We have seen that for dopant concentrations above 5 X 10'' cmP3 or, equivalently, for
breakdown voltages below 4Eg/q-6Eg / q , the reverse
characteristics are dominated by band-to-band tunneling.
This is in agreement with the criteria given in standard
textbooks.
APPENDIX
In order to obtain an analytical approximation for the
we must distinguish between
field-effect functions
two situations:

338

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 39, NO. 2, FEBRUARY 1992

1) For

i.e., for not too large values of the electric field (e.g., at
room temperature and for A E n , p = 0.5 eV this criterion
corresponds to F < 9 X lo5 V/cm) the maximum contribution to the integral in (7) comes from U = U,, where
0 < U, < 1. In this case the integral can be approximated
by a second-order series expansion of the function of U in
the exponent of (7) around its maximum at U,. After setting the integration boundaries to - CO and a,integration
yields

which, by the substitution of (8) into (Al), reduces to

with

Fr

JiiGji$

(43)
9h
So, in the situation where the maximum contribution to
the tunneling effect comes from energy levels above the
minimum level at which electrons can tunnel, the integration interval is irrelevant. Obviously, the same reasoning
holds for the tunneling of holes. If this situation holds for
both electrons and holes, the field-effect functions are
equal, i.e., r, = rp = I.
2) For

i.e., at strong electric fields, the maximum contribution


to the integral in (7) comes from U = 1, i.e., the lowest
energy level at which tunneling is possible. In that case
we expand the term in the exponent of (7) to second order
around U = 1. After setting the lower integration limit
from 0 to --03 and subsequent integration, we arrive at
the following expression:

where a = 0.375 Kn,p,b = 0.5 AE,,,/kT - 0.75Kn,,,


and c = Kn,p - AE,,,/kT. Using the approximate expression for the complementary error function, as can be found
in [18], the following expression is obtained:

1-1

+ P

and p = 0.61685, a l = 0.3480242, a2 = -0.0948798,


and u3 = 0.7478556. The values of a l , u2, and u3 are from
[18], while the value of p is found from the correct behavior of
for Kn,p+ 0.

REFERENCES
[ l ] E. 0. Kane, Zener tunneling in semiconductors, J . Phys. Chem.
Solids, vol. 12, pp. 181-188, 1959.
[2] A. G. Chynoweth, W. L. Feldman, and R. A. Logan, Excess tunnel
current in silicon Esaki functions, Phys. Rev., vol. 121, no. 3, pp.
684-694, 1961.
[3] J. A. Del Alamo and R. M. Swanson, Forward-bias tunneling: A
limitation to bipolar device scaling, IEEE Electron Device Lett..
vol. EDL-7, no. 11, pp. 629-631, 1986; also in Proc. 18th Con$ on
Solid Stare Devices and Materials (Tokyo, Japan), 1986, pp. 283286.
[4] G. A. M. Hurkx, D. B. M. Klaassen, M. P. G . Knuvers, and F. G.
OHara, A new recombination model describing heavy-doping effects and low-temperature behaviour, in Proc. Int. Electron Device
Meeting (Washington, DC), 1989, pp. 307-310.
[ 5 ] H. Schaber, J. Bieger, B. Benna, and T. Meister, Vertical scaling
considerations for polysilicon-emitter bipolar transistors, in Proc.
European Solid-State Device Res. Conf. (Bologna, Italy), 1987, pp.
365-368.
[6] P.-F. Lu, T.-C. Chen, and M. J. Saccamango, Modeling of currents
in a vertical p-n-p transistor with extremely shallow emitter, IEEE
Electron Device Lett., vol. 10, no. 5 , pp. 232-235, 1989.
[7] J. C. Sturm, E. J . Prinz, P. V. Schwartz, P. M. Garone, and Z.
Matutinovic, Growth and transistor applications of Si, - ,GE, structures by rapid thermal chemical vapor deposition, in Proc. 37th Nar.
American Vacuum Soc. Symp (Toronto, Ont., Canada), 1990, pp. 510.
[8] G. A. M. Hurkx, F. G. OHara, and M. P. G. Knuvers, Modeling
forward-biased tunnelling, in Proc. European Solid-State Device
Res. Conf. (Berlin, Germany), 1989, pp. 793-396.
[9] A. G. Milnes, Deep Impurities in Semiconductors. New York:
Wiley, 1973.
[lo] G. Vincent, A. Chantre, and D. Bois, Electric field effect on the
thermal emission of traps in semiconductor junktions, J. Appl. Phys.,
vol. 50, no. 8, pp. 5484-5487, 1979.
Auckland, New Zealand:
[ l l ] L. I. Schiff, Quantum Mechanics.
McGraw-Hill, 1968, pp. 268-279.
[12] L. V. Keldysh, Behavior of non-metallic crystals in strong electric
fields, Sov. Phys.-JETP, vol. 33, no. 4, pp. 763-770, 1958.
Influence of the lattice vibrations of a crystal on the production
[13] -,
of electron-hole pairs in strong electric fields, Sov. Phys. -JETP,
vol. 34, no. 4, pp. 665-668, 1958.
[ 141 E. 0. Kane, Theory of tunneling, J . Appl. Phys., vol. 32, no. 1, pp.
83-89, 1961.
[ 151 G. A. M. Hurkx, On the modelling of tunnelling currents in reversebiased p-n junctions, Solid-State Elecrron., vol. 32, no. 8, pp. 665668, 1989.
[16] H. C. de Graaff and F. M. Klaassen, Compact Transistor Modelling
for Circuit Design. Vienna, Austria: Springer, 1990.
[17] S . M. Sze, Physics of Semiconductor Devices. New York: Wiley,
1981, p. 98.
[18] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions. New York: Dover, 1970, p. 299.

G . A. M. Hurkx, photograph and biography not available at the time of

- -

publication.

k~ 4 3 ~ n , p
(A5)

D. B. M. Klaassen, photograph and biography not available at the time of


publication.

M. P. G. Knuvers, photograph and biography not available at the time of


publication.

Das könnte Ihnen auch gefallen