Diffusion in Silicon

Written by: Scotten W. Jones

 i

Preface

The following document was written in 2000 as a chapter on Diffusion in Silicon for inclu-
sion in a highly technical - text book on Silicon Integrated Circuit Process Technology. For
marketing reasons we abandoned the book without completing it but this was one of several
chapters that are complete. We have recently been posting chapters from this book for free
download, we hope you find it useful. This particular chapter is an update of the previous diffu-
sion chapter post and now includes the appendix with the properties of the error function.
Please note that this material was written back when we planned to publish it as a hard
cover book in black and white and is not in color the way all of the current IC Knowledge
reports are.

Scotten W Jones
President
IC Knowledge LLC
April 25, 2008
 ii

Contents

1.1. Introduction 1
1.2. Basic concepts 1
1.3. Atomic mechanisms of diffusion 2
1.4. Mathematics of diffusion 4
1.5. Diffusion equations 7
1.6. Diffusivity 15
1.7. Solid solubility 37
1.8. Deviations from simple diffusion theory 41
1.9. Diffusion Equipment 56
1.10. Metrology 56
1.11. Properties of the error function 67
Copyright © 2000 IC Knowledge LLC, all rights reserved 1

Diffusion in Silicon

1.1. Introduction
1.2. Basic concepts
1.3. Atomic mechanisms of diffusion
1.4. Mathematics of diffusion
1.5. Diffusion equations
1.6. Diffusivity
1.7. Solid solubility
1.8. Deviations from simple diffusion theory
1.9. Diffusion Equipment
1.10. Metrology

1.1. Introduction
IC fabrication is accomplished by selectively changing the electrical properties of silicon
through the introduction of impurities commonly referred to as dopants. In the early years of
integrated circuit fabrication, deep semiconductor junctions required doping processes fol-
lowed by a “drive-in” step to diffuse the dopants to the desired depth, i.e. diffusion was
required to successfully fabricated devices. In modern state-of-the-art IC fabrication the
required junction depths have become so shallow that dopants are introduced into the silicon at
the desired depth by ion implantation and any diffusion of the dopants is unwanted, therefore
diffusion has become a problem as opposed to an asset. There are many non state-of-the-art
processes still in use throughout that industry where doping and diffusion are still in use and for
state-of-the-art processes diffusion must be understood in order to minimize undesired effects.
In this report the physical mechanisms of diffusion will be reviewed, the mathematics of diffu-
sion will be presented and diffusion data will be provided to allow diffusion effects to be calcu-
lated.

1.2. Basic concepts

Throughout this report the concepts of impurity profile, background concentration and
junction depth will be used.
• Impurity profile - the concentration of an impurity versus depth into the silicon.
• Background concentration - an impurity concentration existing in the silicon that an impu-
rity profile is formed into.
Copyright © 2000 IC Knowledge LLC, all rights reserved 2

• Junction depth - the depth at which the impurity profile concentration is equal to the back-
ground concentration.
Figure 1.1 illustrates the preceding three concepts. The impurity profile, C(x) varies with
depth into the silicon, x, the background concentration CB is shown at a constant level and the
junction depth, xj is the depth at which C(x) = CB.

C
Impurity
Impurity concentration -

profile C(x)
C (atoms/cm )
3

C(x) = CB

Background
concentration - CB

Junction x
depth, xj
Depth - x (␮m)
Figure 1.1: Impurity profile, background concentration and junction depth.

1.3. Atomic mechanisms of diffusion

Impurity atoms may occupy either substitutional or interstitial positions in the Si lattice
(see figure 1.2). Impurity atoms utilized as dopants such as boron (B), phosphorus (P) and
arsenic (As) occupy substitutional positions where the dopant atoms can contribute free elec-
trons or holes to the silicon lattice (dopant atoms introduced to silicon by ion implantation may
not occupy substitutional positions until the dopant is activated). During high temperature pro-
cessing impurity profiles are redistributed through a process known as diffusion. The redistri-
bution of impurities may be intentional as in a “drive-in” step or unintentional as a result of
high temperature oxidation, deposition or annealing processes.

a) Substitutional b) Interstitial
Figure 1.2: Substitutional and interstitial impurities in silicon.

• Diffusion - the redistribution of an initially localized substance throughout a background

medium due to random thermal motion.
Copyright © 2000 IC Knowledge LLC, all rights reserved 3

The movement of a substance due to diffusion is driven by the slope of the concentration
profile.
Impurity atom diffusion may be vacancy, interstitial or a combination mechanism known
as interstitialcy - see figure 1.3.

a) Vacancy b) Interstitial c) Interstitialcy

Figure 1.3: Vacancy, interstitial and interstitialcy diffusion mechanisms.

Vacancy diffusion occurs when a substitutional atom exchanges lattice positions with a
vacancy - requires the presence of a vacancy. Interstitial diffusion occurs when an interstitial
atom jumps to another interstitial position. Interstitialcy diffusion results from silicon self-
interstitials displacing substitutional impurities to an interstitial position - requires the presence
of silicon self-interstitials, the impurity interstitial may then knock a silicon lattice atom into a
self-interstitial position. It is important to observe here that since dopant atoms such as phos-
phorus, arsenic and boron occupancy substitutional positions once activated, dopant diffusion
is closely linked to and controlled by the presence of vacancy and interstitial point defects.
Whether an impurity atom occupies a substitutional or interstitial position in single crystal
silicon, the atom is trapped in a periodic potential defined by the lattice - see figure 1.4. The
probability of an atom jumping from one position to the next increases exponentially with
increasing temperature.

Figure 1.4: Impurity atom diffusing along a periodic potential.

In order to understand diffusion, consider the highly concentrated impurity profile illus-
trated in figure 1.5a, time = 0. If random thermal motion causes 20% of the atoms in each posi-
tion to jump to a new position in each unit of time, 10% of the atoms would be expected to
jump to the left and 10% of the atoms would be expected to jump to the right at time = 1. The
areas of higher concentration would have more atoms jump out of the area than jump in. Over
time the net result of the random thermal motion would be a gradual spreading of the profile -
figure 1.5a, time = 0 to 1.5e, time = 4. At infinite time the impurity distribution would be uni-
form throughout the background medium
Copyright © 2000 IC Knowledge LLC, all rights reserved 4

a) time=0 b) time=1 c) time=2 d) time=3 e) time=4

Figure 1.5: Impurity profile spreading due to random thermal motion - diffusion.

1.4. Mathematics of diffusion

From the basic principles outlined in the preceding section, the diffusion equations may be
derived as follows [1],[2].

1.4.1. Fick’s first law

Consider the two equal thickness slices of a solid illustrated in figure 1.6 where slice 1 and
slice 2 have different impurity concentrations, C1 and C2 given by
N
C 1 = ------1- (1.1)
Aλ
and
N
C 2 = ------2- (1.2)
Aλ
and, N1 and N2 are the number of impurity atoms in slice 1 and slice 2 respectively, A is the
cross sectional area of the slices and ␭ is the thickness of the slices.
Atoms throughout the solid are constantly vibrating from thermal energy. At a given tem-
perature each atom in the slice has a frequency of jumps to the adjacent slice, ␯, with equal
probability of jumping in either direction, therefore the frequency of jumps from one slice to
the next in a particular direction is ␯/2. For the plane separating slice 1 from slice 2, the flow of
atoms across the boundary is

+ΔN 1->2 N1 ν
-------------------- = --------- (1.3)
Δt 2
Copyright © 2000 IC Knowledge LLC, all rights reserved 5

and
-ΔN 2->1 N2 ν
------------------ = --------- (1.4)
Δt 2
where t is time. Note that left to right is the positive direction and right to left is the negative
direction of motion.
N1 atoms N2 atoms N3 atoms

Area A

␭ ␭ ␭
Slice 1 Slice 2 Slice 3
Figure 1.6: Diffusion from a concentration gradient.

Combining equations 1.3 and 1.4 yields the net flow across the boundary
ΔN ν
-------- = --- ( N 1 – N 2 ) (1.5)
Δt 2
and substituting concentration from equations 1.1 and 1.2 into equation 1.5, gives

ΔC ( C 1 – C 2 )Aλν
-------- = ---------------------------------- (1.6)
Δt 2
Since the concentration gradient is given by

ΔC C2 – C1
-------- = ------------------ (1.7)
Δx λ
equation 1.6 and 1.7 can be combined to give
ΔC -Aλ2ν ΔC
-------- = ---------------- -------- (1.8)
Δt 2 Δx
Copyright © 2000 IC Knowledge LLC, all rights reserved 6

By now introducing flux, j, defined as

• Flux - the number of units passing through a boundary area per unit time.
and dividing equation 1.8 by area, the expression for flux may be found as
2
1 ΔC λ ν ΔC
j = --- -------- = - --------- -------- (1.9)
A Δt 2 Δx
The diffusion constant D, is now introduced and defined as
2
λ ν
D ≡ --------- (1.10)
2
Combining equations 1.9 and 1.10 results in
ΔC
j = -D -------- (1.11)
Δx
which in partial differential terms is known as Fick’s first law
∂C
j = -D ------ (1.12)
∂x
Fick’s first law relates the flux of atoms across a boundary to the concentration gradient
and from Fick’s law it can be seen that regions with a large concentration gradient diffuse more
rapidly than regions with a small concentration gradient.

1.4.2. Fick’s second law

Fick’s first law allows the diffusive flux as a function of the concentration gradient to be
calculated, Fick’s second law allows the concentration function, C(x), to be calculated as a
function of time.
Consider the second slice in figure 1.5. Since matter is conserved the time change in con-
centration in slice 2 must be the sum of the fluxes across the slice 1 to slice 2 boundary and the
slice 2 to slice 3 boundary.
Which results in
ΔC- = j – j
------- (1.13)
Δt 1 2

and
j1 – j2 Δj
-------------- = ------ (1.14)
λ Δx
therefore
Δj-
ΔC- = - -----
------- (1.15)
Δt Δx
Copyright © 2000 IC Knowledge LLC, all rights reserved 7

In differential terms, equation 1.15 is the transport equation

∂j-
∂C = - ----
------ (1.16)
∂t ∂x
Substituting Fick’s first law - equation 1.12 into equation 1.16 gives Fick’s second law
2
C = D ∂---------
∂------ C-
(1.17)
∂t ∂x
2

Fick’s second law as derived in the preceding sections is identical in form to the equation
for heat conduction differing only in the constant D, and therefore the large body of work on
heat flow can be applied to the problems of impurity atom diffusion in silicon.

1.5. Diffusion equations

Fick’s laws can now be applied to solve diffusion problems of interest. As was the case
previously the solutions presented here assume a constant diffusivity.

1.5.1. Infinite source diffusion into a semi-infinite body - single step diffusion
Early in the development of integrated circuit fabrication technology, semiconductor dop-
ing was accomplished by exposing semiconductor substrates to a high concentration of the
desired impurity. In order to maintain good process control, the concentration of the impurity
dopant was maintained at a level in excess of the level required to achieve solid solubility of the
dopant at the semiconductor surface (solid solubility is the maximum doping level that is solu-
ble in silicon at a given temperature - see section 1.7). The high concentration of dopant
reduced the process control requirements to time and temperature. The doping source may be a
gas containing the dopant of interest such as arsine - AsH3 - an arsenic source, or a glass layer
such as P2O5 as a phosphorus source. The dopant in these cases may be considered an infinite
impurity source in contact with a semi-infinite medium (the semiconductor substrate thickness
is orders of magnitude greater than the diffusion depth of the dopant for most instances of prac-
tical interest). The surface concentration of the semiconductor will be CS set by the solid solu-
bility at any time during the diffusion process and the dopant concentration in the gas CG is >
CS - see figure 1.7
To solve for the impurity profile in the silicon versus time, Fick's second law - equation
1.17, must be solved for the boundary and initial conditions, where the boundary conditions are
given by
C = C 0 , x = 0, t > 0 (1.18)

and the initial conditions are

C = 0, x > 0, t = 0 (1.19)
Copyright © 2000 IC Knowledge LLC, all rights reserved 8

.
C

CG
CS

gas silicon
-x +x
x=0
Figure 1.7: Single step diffusion - initial conditions.

The solution to Fick’s law for single step diffusion that will be derived here will apply
Laplace transforms after reference [4]. There are several alternate methods that can also be uti-
lized to reach the same result. The laplace transform for a known function of t, for t > 0, is
∞

∫e
-pt
f (p) = f ( t ) dt (1.20)
0

Laplace transforms of common functions may be constructed by integrating equation 1.20.

Starting with Fick’s law - equation 1.17, multiplying both sides by e-pt and integrating with
respect to t from zero to infinity results in
∞ ∞
2
∂ C- 1 -pt ∂C
∫e dt – ---- ∫ e ------ dt = 0
-pt
--------- (1.21)
∂x
2 D ∂t
0 0

Assuming that the orders of differentiation and integration can be interchanged (this can be
justified for the functions of interest [4]), then
∞ ∞
2
∂ C-
2 2
∫e dt = ∂ ∂ C-
∫ Ce dt = ---------
-pt -pt
--------- (1.22)
2
∂x ∂x
2
∂x
2
0 0

Integrating by parts
∞ ∞
∂C
∫e
-pt
------ dt = [ Ce -pt ] ∞0 + p ∫ Ce -pt dt = pC (1.23)
∂t
0 0

Since the term in the square brackets vanishes at t = 0 for the initial conditions - equation
1.19 and at t = infinity through the exponential factor - equation 1.17 reduces to
Copyright © 2000 IC Knowledge LLC, all rights reserved 9

2
∂ C
D ---------
2
- = pC (1.24)
∂x
By treating the boundary conditions - equation 1.18 in the same way
∞
CS
∫ C0 e
-pt
C = dt = ------ , x=0 (1.25)
p
0

Thus the Laplace transform reduces from the partial differential equation 7.17 to an ordi-
nary differential equation 1.24.
The solution of equation 1.24 that satisfies equation 1.25 and for which C bar remains finite
as x approaches infinity is
C S -bx
C = ------ e (1.26)
p
where, b2 = p/D. The Laplace transform for equation 1.26 is [4]
x
C ( x, t ) = C S erfc ------------- (1.27)
2 Dt
where, erfc is the complementary error function, one of the most important functions in diffu-
sion theory. Error function properties are presented in appendix F.
The concentration of impurities versus depth into the silicon calculated using equation 7.27
is plotted as a linear plot versus 2 multiplied by the square root of DT in the top of figure 7.8
and as a log plot versus 2 root DT in the bottom of figure 1.8.
The total number of impurity atoms Q which enter a 1 cm2 section of silicon is defined by
∞

Q ( t ) ≡ ∫ C ( x, t ) dx (1.28)
0

Integrating equation 1.28 results in

2
Q ( t ) = ------- DtC S (1.29)
π
If a single step diffusion as outlined above is used to introduce impurities into silicon as
dopants, equation 1.29 can be utilized to calculate the total number of dopant atoms introduced.
Q will also be used later to calculate profiles resulting from two step diffusion processes.
The gradient or slope of the impurity profile can be calculated by

∂C C S –x 2 /4Dt
------ = – -------------- e (1.30)
∂x ( x, t ) πDt
Copyright © 2000 IC Knowledge LLC, all rights reserved 10

1.0
0.9
0.8
0.7
0.6
C/CS

0.5
0.4
0.3
0.2
0.1 0.1␮m 0.5␮m 2 Dt = 1.0␮m
0.0
0 1.0 2.0 3.0
x(␮)
0
1x10

-1
1x10

-2
1x10
C/CS

-3
1x10 2 Dt = 0.1␮m 0.5␮m 1.0␮m

-4
1x10

-5
1x10
0 1.0 2.0 3.0
x(␮)
Figure 1.8: Complementary error function profile evolution versus time [5].

Using the asymptotic approximation for the complementary error function from appendix
F, equation 1.31 results that is valid for most practical situations [5].
∂C x
------ ≅ – --------- C ( x, t ) (1.31)
∂x ( x, t )
2Dt
Copyright © 2000 IC Knowledge LLC, all rights reserved 11

1.5.2. Plane source - two step diffusion

When a single step diffusion is used to introduce dopant into a semiconductor, a follow-on
second drive-in diffusion is also frequently used. During deposition an oxide-masking layer
maybe used to mask the dopant from some areas of the semiconductor. If single step diffusion
is performed with a long cycle, the dopant can diffuse through the masking oxide resulting in a
loss of the desired pattern. Alternately a short single step diffusion can be used to introduce the
desired Q or total dopant into the semiconductor. The single step diffusion is followed by an
etch back to remove the doped portion of the oxide and then a long drive-in cycle can be used
without fear of the dopant penetrating the masking oxide. Figure 1.9 illustrates the pre-deposi-
tion and drive in doping process.

Oxide
Mask and etch
Silicon

Dopant

Pre-deposition

Oxide etch-back

Drive-in and
grow oxide

Figure 1.9: Pre-deposition and drive-in process.

Following the pre-deposition step the impurity distribution can be determined by solving
the diffusion equation which satisfies the boundary conditions
∂------
C = 0 (1.32)
∂x ( 0, t )

and
C ( ∞, t ) = 0 (1.33)
The solution must also provide for a constant Q to be maintained in the semiconductor. The
initial condition is given by equation 1.27 using the Dt for the pre-deposition step. Solving
equation 1.27 with the boundary condition given by equation 1.32 and equation 1.33 is a diffi-
cult problem. In practice the Dt from the drive-in step is much greater than the Dt for the pre-
Copyright © 2000 IC Knowledge LLC, all rights reserved 12

deposition step. A practical approximation is to assume that the total dopant Q is located in a
sheet at the interface.
By differentiation it may be shown that
A –x2 /4Dt
C = ----- e (1.34)
t
is a solution to equation 1.17 where A is an arbitrary constant. Equation 1.34 is symmetric
about x, tends to zero as x approaches infinity positively or negatively for t>0, and at t=0 van-
ishes everywhere but x=0, where it becomes infinite.
The total amount of impurities Q, diffusing in a cylinder of infinite length and unit cross
section is given by
∞

Q = ∫ C dx (1.35)
–∞

If the concentration distribution is given by equation 1.34, then

2 2
x ⁄ 4Dt = ξ , dx = 2 Dt dξ (1.36)
and
∞
2
–ξ
Q = 2A D ∫e dξ = 2A πD (1.37)
–∞

Equation 1.37 shows that amount of impurity diffusing remains a constant and is equal to
the amount in the interface sheet at x=0. Therefore, substituting for A from equation 1.37 into
equation 1.34 results in
2
Q – x /4Dt
C ( x, t ) = ----------------- e (1.38)
2 πDt
In the actual pre-deposition - drive-in case being considered the gas-silicon interface trun-
cates one side of the distribution, (equation 1.35 assumes the distribution is continuous in both
directions). Equation 1.38 becomes
Q – x2 /4Dt
C ( x, t ) = -------------- e (1.39)
πDt
Which is the drive-in concentration following pre-deposition. Equation 1.39 is the well
known gaussian distribution. The top of figure 1.10 illustrates the progression of the impurity
profile with time as a linear plot and the bottom of figure 1.10 illustrates a log plot.
Copyright © 2000 IC Knowledge LLC, all rights reserved 13

1.2

1.0

0.8
C/Q (cm )
-1

2 Dt = 0.1␮m
0.6

0.4
0.5␮m
0.2
1.0␮m
0.0
0 1.0 2.0 3.0
x(␮)

1
1x10

0
1x10
C/Q (cm )
-1

-1
1x10

-2
1x10

2 Dt = 0.1␮m 0.5␮m 1.0␮m

-3
1x10
0 1.0 2.0 3.0
x(␮)
Figure 1.10: Gaussian distribution normalized concentration versus
distance for successive times [5].

Since Q is fixed, as the impurities diffuse into the silicon the surface concentration CS must
decrease - this is in contrast to the single step diffusion case where CS is a constant. From equa-
tion 1.39, the surface concentration may be calculated by
Copyright © 2000 IC Knowledge LLC, all rights reserved 14

Q
C S ( t ) = -------------- (1.40)
πDt
Thus the concentration distribution can also be represented by
2
– x /4Dt
C ( x, t ) = C S ( t )e (1.41)

The gradient of the concentration distribution can be obtained by differentiating equation

1.41 and is given by
∂C x
------ = – --------- C ( x, t ) (1.42)
∂x ( x, t )
2Dt

Figure 1.11 presents some values of the error function and Gaussian function versus z
where
x
z = ------------- (1.43)
2 Dt

0
1x10

-1
1x10

-2
1x10
Erfc Gaussian
-3
1x10
CB/CS

-4
1x10

-5
1x10

-6
1x10

-7
1x10

-8
1x10
0 1.0 2.0 3.0 4.0
z(␮)
Figure 1.11: Gaussian and error function values versus z.
Copyright © 2000 IC Knowledge LLC, all rights reserved 15

As will be seen in the problem section, figure 1.11 is very useful for performing diffusion
calculations.

1.6. Diffusivity
In equation 1.10 the concept of a diffusion constant - the diffusivity was introduced. In this
section the physical basis of diffusivity will be reviewed and some data on diffusivity values
will be presented.
In order for an impurity to diffuse through silicon, the impurity must either move around
silicon atoms or displace silicon atoms. During interstitial diffusion the diffusing atom jumps
from one interstitial position to another interstitial position, with relatively low barrier energy
and a relatively high number of interstitial sites. Substitutional atoms require the presence of a
vacancy or an interstitial to diffuse and must break lattice bonds. Vacancy and interstitial for-
mation are relatively high-energy processes and so are relatively rare in equilibrium (as will be
seen later certain processes can result in concentrations of interstitials and or vacancies well in
excess of equilibrium). Breaking bonds to the lattice is also a relatively high-energy process
and substitutional atoms tend to diffuse at a much lower rate than interstitial atoms.
The diffusion process can be characterized by a barrier with activation energy Ea, where Ea
is the energy required to jump from one site to the next site. The probability of an atom jump-
ing to another site is given by the product of two terms. The first term is the frequency with
which the atom collides with the barrier ␯0 (the frequency of atomic oscillation in silicon is
approximately 1013 to 1014 /sec.). The second term is the probability the atom will surmount
the barrier during a collision (given by the boltzman factor exp(-Ea/kT)) where k is boltzman’s
constant and T is the temperature in degrees Kelvin. The rate at which atoms jump to a new
position is given by
– E a /kT
ν = ν0 e (1.44)

Each atom can move to any adjacent site, so equation 1.44 should be multiplied by the
number of adjacent sites which is four for silicon for both the interstitial and substitutional case.
For the interstitial case equation 1.44 becomes
– E a /kT
ν inst = 4ν 0 e (1.45)

For substitutional atoms an additional term must be added to account for the probability of
a vacancy or interstitial existing in the adjacent site (vacancies for vacancy diffusion and inter-
stitials for interstitialcy diffusion). If Ed is the energy for point defect formation, the resulting
expression is
– ( E a + E d ) /kT
v subst = 4ν 0 e (1.46)

Experimentally determined activation energies for interstitial atoms are approximately

0.5eV and for substitutional atoms are approximately 3eV (including defect formation) [3]. It is
Copyright © 2000 IC Knowledge LLC, all rights reserved 16

now possible to rewrite the diffusion constant equation 1.10 in terms of the physical properties
of the materials.
Substituting equation 1.45 into equation 1.10 and taking into account that the spacing
between atoms ␭ is d divided by the square root of 3 for a diamond structure, interstitial diffu-
sivity is given by
2
4ν 0 d – Ea /kT
D = -------------- e (1.47)
6
Substituting equation 1.46 into equation 1.10 and taking into account the inter-atomic spac-
ing gives the diffusivity for substitutional atoms of
2
4ν 0 d – ( Ea + Ed ) /kT
D = -------------- e (1.48)
6
Equations 1.47 and 1.48 imply that the diffusivity for all interstitial atoms will be the same
and the diffusivity for all substitutional atoms will be the same, this is not the case. Electric
fields, impurity - lattice size mis-match induced strain and multiple diffusion mechanisms com-
plicate actual diffusivity values.
From the discussion of diffusion mechanisms presented in section 1.2 one would expect
that vacancy diffusion would depend on vacancy concentration, which is a function of temper-
ature, and any non-equilibrium vacancy generation or annihilation mechanisms. Conversely,
interstitialcy diffusion would be expected to be sensitive to silicon self-interstitial concentra-
tion which, also depend on temperature and non-equilibrium processes. There are a number of
IC fabrication processes which can generate non-equilibrium point defect concentrations. Table
1.1 summarizes process effects on point defects where V indicates vacancy generation and I
indicates interstitial generation.
Table 1.1: Process generated point defects

Process step Point defects

Ion implantation I-V pairs
Oxidation I
Nitridation V
Salicidation V

Ion implantation generates interstitial vacancy pairs by knocking lattice atoms from their
lattice position, oxidation is a partially complete reaction and injects interstitials into the under-
lying silicon, nitridation and silicide formation inject vacancies. Silicon surfaces act as recom-
bination sites for point defect concentrations so that the proximity of a surface must be taken
into account. Interstitials may also cluster together to form defects which grow and shrink
depending on time, temperature and interstitial concentrations with the result that defects may
act as an interstitial sink or source. Interstitials and vacancies may also recombine in the bulk.
Copyright © 2000 IC Knowledge LLC, all rights reserved 17

In order to accurately calculate diffusivity it is therefore important to understand the atomic

mechanism by which each dopant diffuses as well as the local point defect concentrations.
Effective diffusivities in the presence of point defects can be expressed as
D eff C CV
- = f I -------I- + ( 1 – f I ) ---------
-------- (1.49)
D∗ C∗ I C∗ V

where, Deff is the effective diffusivity, D* is the equilibrium diffusivity, fI is the fraction of dif-
fusion due to an interstitialcy mechanism, CI is the interstitial concentration, C*I is the equilib-
rium interstitial concentration, CV is the vacancy concentration and C*V is the equilibrium
vacancy concentration.
Equation 1.49 allows vacancy, interstitialcy or mixed vacancy-interstitialcy diffusivity in
the presence of point defects to be modeled.
One of the classic experiments for determining the effect of interstitials on diffusivity and
therefore the interstitialcy component of diffusion is illustrated in figure 7.12. The experiment
begins with the formation of a uniform doped layer. A layer of silicon nitride (Si3N4) is then
deposited and patterned to leave an opening where oxidation is to take place - Si3N4 of suffi-
cient thickness can block thermal oxidation. A silicon dioxide (SiO2) layer is then grown in the
Si3N4 opening. Oxidation is usually incomplete and approximately 1 in every 1,000 silicon
atoms are un-reacted. The un-reacted silicon atoms breaks free of the interface and becomes sil-
icon self-interstitials. The concentration of silicon self-interstitials is higher under the center
region where oxidation is occurring and if interstitials increase the diffusivity of the dopant
being measured then the dopant will diffuse deeper under the center section. Conversely if
interstitials retard diffusion, for example if interstitials recombine with vacancies limiting the
vacancy supply for vacancy diffusion, the dopant will diffuse to a shallower depth under the
center section.

Doped layer
Form uniform doped layer
Silicon

Silicon nitride
Deposit and pattern silicon
nitride layer (blocks oxidation)

Degree of
enhancement

Silicon dioxide Grow oxide - dopant diffuses

during oxidation

Figure 1.12: Experiment to measure oxidation enhanced diffusion.

Copyright © 2000 IC Knowledge LLC, all rights reserved 18

It is important to note here that the oxidation reaction is consuming silicon in the center
section. Without diffusivity enhancement the distance between the SiO2 - Si interface and dif-
fusion boundary under the center section would be smaller than the distance between the Si3N4
- Si interface and the diffusion boundary out at either side.
Many researchers have investigated the mechanisms of diffusion. Table 1.2 summarizes
the interstitialcy component

Table 1.2: Interstitialcy component of diffusion [20].

Temperature (oC)
Element Reference
950 1,000 1,050 1,100
Aluminum 0.6-0.7 11
Antimony 0.02 9
0.02 11
Arsenic 0.10 0.12 0.15 0.19 12
0.2-0.4 9
0.22 0.25 0.30 0.35 19
0.09 0.09 0.09 0.09 8
0.25 0.30 0.35 0.42 10
0.2-0.5 11
Boron 0.16 0.17 0.18 0.19 12
0.24- 9
0.22 0.34 0.42 0.52 19
0.17 0.32 0.17 0.17 8
0.38 0.17 0.60 0.80 10
0.45 0.8-1.0 11
Gallium 0.6-0.7 11
Indium 0.30 9
Phosphorus 0.14 0.16 0.19 0.23 12
0.38 0.55 9
0.27 0.34 0.42 0.51 19
0.12 0.12 0.12 0.12 8
0.30 0.35 0.45 0.55 10
0.5-1.0 11

From table 1.2 it is clear that there is a great deal of disagreement about the interstitialcy
component of diffusion. Recently Grossman [67] has pointed out that experimental measure-
ments of the interstitialcy component of diffusion for a single species require the solution to a
single equation, equation 1.49 with three unknowns, the percent of interstitialcy diffusion and
the ratios of interstitial and vacancy concentrations to equilibrium. If however, two different
Copyright © 2000 IC Knowledge LLC, all rights reserved 19

diffusing species are measured under identical conditions a system of two equation in two
unknowns can be formulated and solved without resulting to assumptions about the vacancy
and interstitial ratios. Grossman has performed this experiment for boron and antimony and
found

f B ( 860 C ) ≥ 0.98 −
o
+ 0.01 (1.50)

and

f Sb ( 790 C ) ≤ 0.01 −
o
+ 0.01 (1.51)

Or in other words, B diffuses through an almost exclusively interstitialcy mechanism and

Sb diffuses through an almost exclusively vacancy mechanism. Grossman further suggests that
any substitutional dopant in silicon should diffuse through either a pure vacancy or a pure inter-
stitialcy mechanism.
A complete diffusion model needs to account for point defect interactions with the diffus-
ing species and also formations of defect clusters. For example, a complete set of interactions
for boron would need to account for the following eight interactions [68]:
A neutral interstitial interacts with a cluster to either grow or shrink the cluster
0
I +C↔C (1.52)
An interstitial and vacancy either annihilate each other or are generated as a Frenkel pair
0 0
I +V ↔0 (1.53)
An interstitial and a hole interact to create a positively charged interstitial from a neutral
interstitial or a positively charged interstitial releases a hole and becomes neutral.
0 + +
I +h ↔I (1.54)
A negatively charged boron atom combines with a neutral interstitial to form a neutral
boron - interstitial pair and release an electron or a neutral boron - interstitial pair captures an
electron and breaks up into a negatively charged boron atom and a neutral interstitial.
- 0 0 -
B + I ↔ BI + e (1.55)
A negatively charged boron atom combines with a neutral interstitial and a hole to form a
neutral boron - interstitial pair or a natural boron - interstitial pair releases a hole and breaks up
into a negatively charged boron atom and a neutral interstitial.
- 0 +
B + I + h ↔ BI (1.56)
A neutral boron - interstitial pair combine with a vacancy annihilating the vacancy - inter-
stitial pair and releasing a negatively charged boron atom and a hole or a hole and a negatively
charged boron atom combine and generate a neutral boron - interstitial pair and release a vac-
ncy.
Copyright © 2000 IC Knowledge LLC, all rights reserved 20

0 0 - +
BI + V ↔ B + h (1.57)
A positively charged boron - interstitial pair combine with a vacancy annihilating the
vacancy - interstitial pair and releasing a negatively charged boron atom and two holes or two
holes and a negatively charged boron atom combine and generate a positively charged boron -
interstitial pair and release a vacancy.
+ 0 - +
BI + V ↔ B + 2h (1.58)
A negatively charged boron atom and a positively charged boron - interstitial pair combine
to form a neutral boron cluster or a neutral boron cluster breaks apart into a negatively charged
boron atom and a positively charged boron - interstitial pair.
- +
B + BI ↔ BB cls (1.59)

The equation set can be implemented through a set of reaction diffusion equations of the
form [69]:
∂C 0 ∂ ∂C 0
1 1 - 2 2
D 0 ⋅ ------------- + K F C B – C 0 – K R C 0 n – K F C BI + C 0 + K R C B – p
BI BI
------------
- = (1.60)
∂t ∂ x BI ∂ x I BI V

This system of equation contains a large number of parameters that must be know in order
to evaluate the model predictions. Recently general purpose differential equation solvers have
become available which allows systems like this to be evaluated [80]. It seems reasonable to
expect that commercial process simulators will include such models in the future.
In order to enable the reader to make diffusion calculations of reasonable accuracy Fair’s
diffusivity model will be presented in the next few sections. Fair has developed a model for dif-
fusivity based on impurity interactions with charged vacancy states that provides a reasonably
good fit to most observed diffusion results [6],[7]. Fair's model calculates diffusivity as the sum
of the individual charged vacancy - impurity diffusivities.
The intrinsic model is given by
0 + - =
Di = D + D + D + D + … for n or p << n i (1.61)

where: Di is the intrinsic diffusivity, D0 is the neutral vacancy - impurity diffusion, D+ is the
positively charged vacancy - impurity diffusion, D- is the negatively charged vacancy - impu-
rity diffusion and D= is the doubly negatively charged vacancy - impurity diffusivity and ni is
the intrinsic carrier concentration - see chapter 1.
For extrinsic silicon
0 + p - n = n 2
D x = D + D ---- + D ---- + D ---- +… for n or p >> n i (1.62)
ni ni ni

where, Dx is the extrinsic diffusivity. For specific impurities not all vacancy charge state -
impurity combinations will participate in the diffusivity.
Copyright © 2000 IC Knowledge LLC, all rights reserved 21

Even though it is now known that substitutional diffusion is not strictly vacancy domi-
nated, Fair’s model is useful for the reasonable results it obtains without resorting to complex
simulation.
The temperature dependence of the diffusivity values presented in the next several sections
will take the arhenius form
– E a /kT
D = D0 e (1.63)

where D0 is a pre exponential constant.

Arhenius equations form a straight line when ln(D) is plotted versus 1/T.

1.6.1. Diffusivity of antimony in silicon

Antimony is believed to diffuse by a purely vacancy mechanism with the Sb+V- dominat-
ing. The intrinsic diffusivity of antimony is given by [6]
– 3.65 /kT 2
D i = 0.214e cm /s (1.64)

The extrinsic diffusivity of antimony is given by [6]

0 - n
D x = D + D ---- (1.65)
ni

and D+ and D= = 0, D0 and D- are given by

0 – 3.65 /kT 2
D = 0.214e cm /s (1.66)
and
- – 4.08 /kT 2
D = 15.0e cm /s (1.67)
The diffusivity of antimony versus temperature is illustrated in figure 1.13
Antimony has a large tetrahedral radius of 0.136nms versus a radius of 0.118nms for sili-
con giving a 1.15 mismatch factor and creating strain in the silicon lattice [3]. Antimony diffu-
sion is retarded by oxidation due to injected silicon self-interstitials annihilating vacancies.
Table 1.3 presents the ratio of antimony diffusivity under oxidizing (dry O2)and non-oxidizing
conditions for a variety of times and temperatures [71].
Table 1.3: Oxidation retardation of antimony diffusivity [71]

Temperature (oC) Time (mins.) DOX/D*

1,000 500 0.20

1,000 1,500 0.18
1,100 10 0.25
1,100 60 0.30
1,100 600 0.44
Copyright © 2000 IC Knowledge LLC, all rights reserved 22

-11
1x10

-12
1x10

21
-13 10
1x10
Diffusivity (cm /s)

20
2

-14 10
1x10

19
10
-15
1x10

-16
1x10 Intrinsic

-17
1x10

-18
1x10
800 900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.13: Diffusivity of antimony in silicon versus temperature and
antimony concentration. Calculated from equations 1.64 through 1.67.

1.6.2. Diffusivity of arsenic in silicon

Arsenic is believed to diffuse primarily through a vacancy mechanism with an interstitialcy
component [7]. Estimates of the interstitialcy component of arsenic diffusion are presented in
table 1.2. The arsenic interstitial formation energy is estimated to be a relatively high 2.5eV.
Above approximately 1,050oC diffusion is dominated by As+V- vacancy pairs with As+V=
being relatively rare due to the small pair binding energy. Oxidation has little effect on arsenic
diffusivity due to the high arsenic interstitial formation energy.
The intrinsic diffusivity of arsenic is given by [6]
– 4.1 /kT 2
D i = 22.9e cm /s (1.68)

At high concentration (>1020 atoms/cm3) arsenic diffusivity is complicated by clustering.

Arsenic atoms are believed to form clusters where 3 arsenic atoms bond with an electron (see
section 1.7 - solid solubility). Clusters are virtually immobile at T <1,000oC. Below 1,000oC
equations 7.68 through equation 1.71 gives the diffusivity for un-clustered arsenic. Above
~1,000oC arsenic un-clusters and equation 1.68 through equation 1.71 again gives the diffusiv-
ity
Copyright © 2000 IC Knowledge LLC, all rights reserved 23

0 - n
D x = D + D ---- (1.69)
ni

and, D+ and D= = 0, D0 and D- are given by

0 – 3.44 /kT 2
D = 0.066e cm /s (1.70)
- – 4.05 /kT 2
D = 12.0e cm /s (1.71)
Diffusivity of arsenic versus doping is plotted in figure 1.14.
-11
1x10

-12
1x10

21
-13 10
1x10
Diffusivity (cm /s)

20
10
2

-14
1x10
19
10
-15
1x10

1x10
-16 Intrinsic

-17
1x10

-18
1x10
800 900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.14: Diffusivity of arsenic in silicon versus temperature and
arsenic concentration. Calculated form equations 7.68 through 7.71.

Arsenic has exactly the same tetrahedral diameter as silicon and so arsenic does not strain
the silicon lattice or induce enhanced diffusivity in other dopants for T > ~700oC.

1.6.3. Diffusivity of boron in silicon

In Fair's model of diffusion, boron is assumed to diffuse exclusively by a vacancy mecha-
nism under non-oxidizing conditions. Estimates of the interstitialcy component of boron diffu-
sion are presented in table 1.2 and have recently been estimated at >98% [67].
Copyright © 2000 IC Knowledge LLC, all rights reserved 24

In Fair’s model boron diffuses by a B+V- vacancy pair with a migration energy approxi-
mately 0.5eV lower than other vacancy-ion pairs. Boron diffusivity is enhanced by p dopants
when p > ni and reduced for p < ni, boron diffusion is actually retarded in N type silicon where
n > ni.
The intrinsic diffusivity of boron is given by
– 3.46 /kT 2
D i = 0.76e cm /s (1.72)

The extrinsic diffusivity of boron is given by

0 + p
D x = D + D ---- (1.73)
ni

and D- and D= = 0, D0 and D+ are

0 – 3.46 /kT 2
D = 0.037e cm /s (1.74)
and
+ – 3.46 /kT 2
D = 0.76e cm /s (1.75)
The diffusivity of boron versus doping is plotted in figure 1.15
Boron is a faster diffuser than either phosphorus or arsenic. Boron has a tetrahedral radius
of 0.82nms versus 1.18nms for silicon or a 0.75 mismatch ratio [3]. The relatively large size
mismatch for boron versus silicon produces lattice strain that can lead to dislocation formation
and reduced diffusivity. The strain introduced by boron into the silicon lattice can help to getter
impurities by a trapping mechanism. High concentrations of boron are particularly good at get-
tering iron. Boron has a relatively low energy for interstitial formation of 2.26eV, therefore
boron diffusivity is enhanced by oxidation. Figure 1.16 presents some data on oxidation
enhanced boron diffusion.
The left side of figure 1.16 - 1.16a illustrates the increase in boron junction depth due to
oxidation enhanced diffusion versus time for wet oxidation at 1,100oC. The right side of figure
1.16 - 1.16b illustrates the diffusivity of boron during dry oxygen oxidation versus an inert
atmosphere versus temperature. Boron diffusion is enhanced during oxidation for both the
intrinsic and extrinsic cases [15].
Boron forms stable oxides such as B2O5 and HBO2 and readily segregates into the growing
oxide during oxidation processes.

1.6.4. Diffusivity of phosphorus in silicon

The diffusivity of phosphorus is again explained as a vacancy dominated diffusion. Esti-
mates of the interstitialcy component of phosphorus diffusion are presented in table 1.2. Phos-
phorus diffusion exhibits three distinct regions of behavior illustrated in figure 1.17
Copyright © 2000 IC Knowledge LLC, all rights reserved 25

-10
1x10

-11
1x10

-12 21
1x10 10
Diffusivity (cm /s)

-13
1x10 10
20
2

-14
1x10 10
19

-15
1x10

Intrinsic
-16
1x10

-17
1x10

-18
1x10
700 800 900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.15: Diffusivity of boron in silicon versus temperature and boron concentration.
Calculated from equations 1.72 through 1.75.

-11
1.4 10

1.2 Dry O2 Oxidation

-12
10
Diffusivity (cm2/s)

1.0
⌬Xj (␮m)

0.8
-13
10
0.6
Non Oxidizing

0.4 -14
10
0.2 0
Wet Oxygen - 1,100 C
-15
0 10
0 1.0 2.0 3.0 4.0 1,000 1,100 1,200
Oxidation Time (hrs.) Temperature (oC)

Figure 1.16: Oxidation enhancement of boron diffusivity [13],[14].

Copyright © 2000 IC Knowledge LLC, all rights reserved 26

Total Phosphorus Concentration

- s
(PV) ⬀n3
2
ns D⬀n Electron Concentration, n

+ =
P V Pair Dissociation Region
Log (Concentration)

- 0 -
ne (PV) -> (PV) + e
0 + -
(PV) <-> P + V

Kink

Excess Vacancy
Concentration

Tail Region
3
D = const. x n s Emitter Dip
Effect
-
(V)

0 Xo X
Figure 1.17: Phosphorus diffusion profile and vacancy model [6].

The phosphorus profile illustrated in figure 1.17 has three distinct regions:
1. The high concentration region where the total phosphorus concentration exceeds the free
carrier concentration.
2. A kink in the profile
3. A tail region of enhanced diffusivity.
In the high concentration region a fraction of P+ ions pair with V= vacancies as P+V= pairs
noted as (PV)-. The concentration of (PV)- pairs is proportional to the surface electron concen-
tration cubed (ns3), which has to be determined experimentally.
The intrinsic diffusivity of phosphorus is given by [6]
– 3.66 /kT 2
D i = 3.85e cm /s (1.76)

In the high concentration region the extrinsic diffusivity of phosphorus is given by [6]
0 = n 2
D x = D + D ---- (1.77)
ni

and D- ~ 0, D+ = 0, D0 and D= are given by

0 – 3.66 /kT 2
D = 3.85e cm /s (1.78)
Copyright © 2000 IC Knowledge LLC, all rights reserved 27

and
= – 4.37 /kT 2
D = 44.2e cm /s (1.79)
Phosphorus has a tetrahedral radius of 0.110nms versus 0.118nms for silicon resulting in a
mismatch ratio of 0.93 [3], and at high concentrations - phosphorus lattice strain can lead to
defect formation. Fair and Tsai have proposed that tail formation in the phosphorus profile is
due to dissociation of P+V= pairs when n drops below 1020 /cm3 at the diffusion front (the
Fermi level is ~0.11eV from the conduction band), the V= vacancy changes state and the bind-
ing energy decreases enhancing the probability of disassociation and increasing the vacancy
flux. Other researchers have found evidence of interstitialcy diffusion for phosphorus calling
into question the physical correctness of the vacancy model [17],[18].
The electron concentration in the transition or “kink” region is given by
21 – 0.39 /kT 3
n e = 4.65 × 10 e n/cm (1.80)

The disassociation of the (PV)- pairs increase the vacancy concentration tail by
- x -
( PV ) → ( PV ) + e (1.81)
and
x + -
( PV ) ↔ P + V (1.82)
The diffusivity in the tail region increase as the V- concentration increases and is given by
[6]
3
0 ns -
D tail - [ 1 + e 0.3/kT ]
= D + D --------- cm /s
2
(1.83)
2
ne ni

and
- – 4 /kT 2
D = 4.44e cm /s (1.84)
Figure 1.18 illustrates the diffusivity of phosphorus versus temperature for the intrinsic and
extrinsic cases. Figure 1.19 illustrates the diffusivity of phosphorus in the tail region.
The same (PV)- dissociation mechanism that creates enhanced diffusivity in the phospho-
rus “tail” can increase the diffusivity of dopants in junction pre-existing under phosphorus dif-
fusions - see figure 1.20.
The effect of heavy phosphorus doping levels on co-diffusing species was first noted dur-
ing emitter diffusion for bipolar transistors. The heavy phosphorus emitter was observed to
increase the junction depth of the underlying boron doped base region relative to surrounding
areas. The enhanced diffusion under the phosphorus emitters came to be known as “emitter
push”.
Copyright © 2000 IC Knowledge LLC, all rights reserved 28

-10
1x10

-11
1x10

21
-12 10
1x10
Diffusivity (cm /s)

-13
1x10
2

20
-14
10
1x10

-15
1x10
19
-16
10
1x10
Intrinsic

-17
1x10

-18
1x10
700 800 900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.18: Phosphorus diffusivity in silicon versus phosphorus concentration and
temperature - high concentration region.
Calculated from equations 1.76 through 1.79.

Above 5 x 1020 phosphorus atoms/cm3 the misfit lattice strain from the phosphorus intro-
duces lattice strain (band gap narrowing) and reduces (PV)- pair dissociation. Figure 1.20 illus-
trates the increase in boron doped base junction depth versus phosphorus content for a 1,000C
phosphorus emitter diffusion. The solid line is the increase in junction depth excluding the
strain effect and the dotted line includes the strain effect, the round circles are experimental
results.
The increase in base junction depth directly under the phosphorus diffusion is given by [21]
1/2
⎛ 2D in t⎞
B B 1/2
– ⎛ 1 + --------------⎞
2Di t
δ ( t ) = W 0 ⎜ 1 + ------------
-⎟ (1.85)
⎝ ⎝ W0 ⎠
W0 ⎠
2

where ␦(t) is the increase in junction depth under the emitter diffusion and W0 is a quantity
related to the profile of the base diffusion prior to emitter diffusion.
W0 is given by
Copyright © 2000 IC Knowledge LLC, all rights reserved 29

Q0
W 0 = 0.4 -------- (1.86)
C p0

where, Q0 is the integrated doping of the base, and Cp0 is the peak concentration of the base
dopant prior to emitter diffusion.
The diffusivity of boron under the emitter region is given by
0 -
B B DP + DP
D in = D i --------------------
0
(1.87)
DP

where, DinB is the diffusivity of the boron under the emitter, D0P is the diffusivity of the phos-
phorus - neutral vacancy pair, and D-P is the diffusivity of the phosphorus - negative vacancy
pair.

-9
1x10

-10
1x10 21
10
-11
1x10

-12
1x10
Diffusivity (cm /s)
2

-13
1x10 20
10
-14
1x10

-15
1x10

-16
1x10
19
10
-17
1x10 Intrinsic

-18
1x10
700 800 900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.19: Phosphorus Diffusivity in silicon versus surface electron concentration and
temperature - tail region. Calculated from equations 1.78, 1.83 and 1.84.

Phosphorus diffusivity (both intrinsic and extrinsic) is enhanced by oxidation (similar to

boron) - see figure 1.22.
Copyright © 2000 IC Knowledge LLC, all rights reserved 30

Heavy
phosphorus
emitter

Region of
enhanced
boron Region of
diffusion intrinsic
boron
diffusion
Figure 1.20: Enhanced diffusivity under heavy phosphorus doped regions -
“emitter push”.

12

Strain
10 effect
included

8
Cs(10 cm )
-3

Intrinsic
2 0

6 theory

0
0 0.4 0.8 1.2 1.6 2.0
⌬ (␮m)

Figure 1.21: High concentration phosphorus induced strain effect on

enhanced diffusion [21].

Figure 1.22 plots the ratio of diffusivity during oxidation to diffusivity under an inert ambi-
ent versus oxidation rate at four different temperatures. The data in figure 1.22 was generated
by varying the oxygen partial-pressure and the initial oxide thickness to produce varying oxida-
tion rates. For a given temperature the degree of diffusivity enhancement increases with
increasing oxidation rate.
Copyright © 2000 IC Knowledge LLC, all rights reserved 31

3.0
o
(100) Si, dry O2 1,050 C,
270mins
o
1,100 C,
o
1,000 C, 90mins
990mins
DOX/D

1.0

o
1,150 C,
30mins

0.5
-7 -6 -5 -4 -3
10 10 10 10 10
Oxidation Rate (␮m/min)
Figure 1.22: Phosphorus diffusivity enhancement versus oxidation rate.
Adapted from [22]/

1.6.5. Diffusivity of miscellaneous dopants in silicon

In the preceding four sections the diffusivity of the four most common silicon dopants was
discussed in detail. There are a a variety of other impurities whose diffusivity may be of inter-
est. Table 1.4 presents a compilation of diffusivity values.

Table 1.4: Diffusivity of miscellaneous impurities in silicon.

Temperature
Impurity D0 (cm2/s) Ea (eV) Mechanism Reference
range (oC)
Ag 1,100-1,300 2x10-3 1.6 Interstitial 23
Al 0.5 3.0 28
1,350 4.1 6
As (intrinsic) 22.9 4.1 Vacancy 6
-3
Au 800-1,200 1.1x10 1.12 Interstitial 23
Au (interstitial) 700-1,300 2.4x10-4 0.39 23
Au (substitutional) 700-1,300 2.8x10-3 2.04 23
B (intrinsic) 0.76 3.46 Vacancy 6
Bi 396 4.12 6
C 1.9 3.1 28
Co 900-1,200 9.2x10-4 2.8 25
Copyright © 2000 IC Knowledge LLC, all rights reserved 32

Table 1.4: Diffusivity of miscellaneous impurities in silicon.

Temperature
Impurity D0 (cm2/s) Ea (eV) Mechanism Reference
range (oC)
Cr 1,100-1,250 0.01 1.0 24
-2
Cu 800-1,100 4.0x10 1.0 Interstitial 23
-3
Cu (interstitial) 4.7x10 0.43 Interstitial 23
-3
Fe 1,100-1,250 6.2x10 0.87 Interstitial 23
Ga 225 4.12 28
5
Ge 6.26x10 5.28 28
H2 9.4x10-3 0.48 Interstitial 23
He 0.11 1.26 28
In or Tl 16.5 3.9 28
269 4.19 6
K 800-1,000 1.1x10-3 0.76 Interstitial 23
Li 25-1,350 2.3x10-3 0.63 Interstitial 23
to 9.4x10-4 to 0.78
N 0.05 3.65 6
Na 800-1,100 1.6x10-3 0.76 Interstitial 23
Ni 450-800 0.1 1.9 Interstitial 23
O2 700-1,240 7x10-2 2.44 26
O (interstitial) 330-1,250 0.17 2.54 Interstitial 26
P (intrinsic) 3.85 3.66 Vacancy 6
Pt 800-1,100 1.5x102 2.22 Interstitial 27
to 1.7x102 to 2.15
S 0.92 2.2 Interstitial 28
Sb (intrinsic) 0.214 3.65 Vacancy 6
Si (interstitial) 0.015 3.89 Vacancy 6
Sn 32 4.25 28
Zn 0.1 1.4 Interstitial 28

Figure 1.23 presents the values from table 1.4 in graphical form versus temperature, the
graph is useful for comparing relative diffusivities.
As mentioned in previous sections, the size mismatch between a diffusing specie and the
silicon lattice can result in lattice strain which may enhance the diffusivity of co-diffusing
specie. Table 1.5 presents the relative sizes of selected impurities versus silicon.
Copyright © 2000 IC Knowledge LLC, all rights reserved 33

-3
1x10
H2 Co
-5
Cu Fe Na
1x10 Cr K Li
He

-7
1x10 Au
Zn Ni
Pt
-9
1x10
Ag

-11
1x10 O2
S
Diffusivity (cm /s)

C
2

1x10
-13 Al
Sb
In
-15 Ga As
1x10
Sn
P
-17 N
1x10 B
Bi
Ge Si
-19
1x10

-21
1x10

-23
1x10

-25
1x10
600 700 800 900 1,000 1,100 1,200 1,300
o
Temperature ( C)
Figure 1.23: Diffusivities of selected impurities in silicon versus temperature -
see table 1.4.

1.6.6. Diffusivity of impurities in polysilicon

Polysilicon is made up of grains of single crystal silicon randomly oriented relative to each
other. Within the crystalline grains diffusion has characteristics similar to bulk single crystal
silicon. In the grain boundaries diffusion can proceed at 100 times the rate of diffusion within
the grains. Because the grain boundaries dominate polysilicon diffusion, grain size and hence
the grain boundaries determine the diffusivity of impurities in the polysilicon film.
Copyright © 2000 IC Knowledge LLC, all rights reserved 34

Table 1.5: Size of impurity atoms relative to silicon [3].

Tetrahedral Size relative

Element radius (nm) to silicon
Aluminum 0.126 1.07
Gallium 0.126 1.07
Gold 0.150 1.27
Indium 0.144 1.22
Silicon 0.118 1.00

The grain size for polysilicon can range from 0.1 to 20µms in size and depends on the pol-
ysilicon deposition conditions and thickness, the underlying film, and the subsequent thermal
history of the film. Due to the number of variables that effect grain size and the dominant role
of grain boundaries in polysilicon diffusion, diffusivities reported by various researchers vary
widely and it is difficult to make definitive predictions. Table 1.6 presents some reported val-
ues for impurity diffusion in polysilicon.
Table 1.6: Diffusivity for miscellaneous impurities in polysilicon [41].

Temperature
Impurity D0 (cm2/s) Ea (eV) Mechanism System Reference
range (oC)
As 700-1,050 0.28 2.8 ----- RTP 30
800-1,000 8.4x104 3.8 Tail region RTP 31
800-1,000 1,700 3.8 Peak region RTP 31
750-950 8.6x104 3.9 Grain-bound Furnace 37
700-850 10 3.36 Grain-bound Furnace 38
950-1,100 0.63 3.22 Combined Furnace 39
700-1,000 1.66 3.22 Combined Furnace 36
B 1,000-1,200 6.6x10-4 1.87 Grain-bound Furnace 32
P 800-1,000 1.9x10-2 1.76 Grain-bound RTP 29
1,100-1,200 4.0x10-3 1.71 Grain-bound Furnace 33,32
900-1,100 5.3x10-5 1.95 Grain-bound Furnace 34
700-850 44.8 3.4 Combined Furnace 35
700-1,100 1.0 2.75 Combined Furnace 36
700-1,050 1.81 3.14 ----- Furnace 30
Sb 1,050-1,150 13.6 3.9 Bulk Furnace 40
1,050-1,150 812 2.9 Grain-bound Furnace 40
Copyright © 2000 IC Knowledge LLC, all rights reserved 35

Figure 1.24 illustrate the diffusivity of selected impurity in polysilicon versus temperature
from table 1.4 in graphical form to allow easy comparisons of relative diffusivities.

-7
1x10
Sb
P
-9
1x10
B
Diffusivity (cm /s)

As P
2

-11
1x10
P

-13 P
1x10
P As
As Sb
-15
1x10 As

-17
1x10
700 800 900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.24: Diffusivities of selected impurities in polysilicon - see table 7.4.

1.6.7. Diffusivity of impurities in silicon dioxide

Silicon dioxide is often used to mask against impurity diffusion during pre-deposition - see
figure 1.8. Silicon dioxide may also become inadvertently doped by ion implantation during
processing. It is therefore important to be able to calculate the diffusion of impurities through
silicon dioxide. Table 1.7 presents the diffusivity of selected impurities in silicon dioxide.
Table 1.7: Diffusivity of selected impurities in silicon dioxide.

Impurity D0 (cm2/s) Ea (eV) Source and ambient Reference

As 67.25 4.7 Ion implant, N2 43

3.7x10-2 3.7 Ion implant, O2 43
-10
Au 8.2x10 0.8 44
1.52x10-7 2.1 44
B 7.23x10-6 2.38 B2O3 vapor, N2+O2 42
1.23x10-4 3.39 B2O3 vapor, Ar 42
3.16x10-4 3.53 Borosilicate 42
-4
D2 5.01x10 0.455 46
5
Ga 1.04x10 4.17 Ga2O3 vapor, H2+N2+H2O 42
H+ 1 0.73 46
H2 5.65x10-4 0.446 44
Copyright © 2000 IC Knowledge LLC, all rights reserved 36

Table 1.7: Diffusivity of selected impurities in silicon dioxide.

Impurity D0 (cm2/s) Ea (eV) Source and ambient Reference

He 3x10-4 0.24 44
H2O 10-6 0.79 44
-4
O2 2.7x10 1.16 44
-4
OH 9.5x10 0.68 46
Na 6.9 1.3 44
P 5.73x10-5 2.20 P2O5 vapor, N2 42
1.86x10-1 4.03 Phosphosilicate glass, N2 45
Pt 1.2x10-13 0.75 44
Sb 1.31x1016 8.75 Sb2O5 vapor, O2+N2 42

Figure 1.25 presents the diffusivity values from table 1.6 in graph form for easier compari-
son. The diffusivities presented in table 1.6 for arsenic, boron and phosphorus are nitrogen
ambient diffusion values. Boron and phosphorus are vapor sources and arsenic is an ion
implanted source.

1.6.7.1. Diffusivity of boron in silicon dioxide

As IC minimum feature sizes have shrunk the diffusion of boron through silicon dioxide
has taken on an increasingly critical aspect. When MOSFET minimum features were above
approximately 0.5␮ms, the gate polysilicon was doped N-type with phosphorus. As minimum
features have shrunk below approximately 0.5␮ms, adequate PMOS performance cannot be
achieved without doping the PMOS gate polysilicon P-type. For <0.5␮m minimum feature size
devices, the NMOS polysilicon is phosphorus doped and the PMOS polysilicon is boron doped.
Boron is readily taken up by oxide (see section 1.81), and boron can diffuse through the thin
gate oxide layer and dope the channel resulting in uncontrolled shifts in the PMOS device
threshold voltage. The diffusion of boron through silicon dioxide may be increased through
interactions with other species.
One technique for introducing boron into device structures is ion implantation of BF2 ions.
Fluorine from the BF2 ions increases the diffusivity of boron in silicon dioxide. The diffusivity
of boron in the presence of fluorine is given by [66]
–3 –6 1/2 – 3.56 /kT
D ( BF 2 ) = 1.1x10 ( 0.33 + 1.1x10 φF 2 )e (1.88)

where, ␾F2 is the BF2 dose in atoms/cm2.

Hydrogen also increases the diffusivity of boron in silicon dioxide given by [66]
– 4 – 3.56 /kT 1/2 – 5 – 3.0 /kT
D ( H 2 ) = 3.63x10 e + [ H2 ] ⋅ 2.65x10 e (1.89)

where, H2 is the percent of H2 in the diffusing ambient.

Copyright © 2000 IC Knowledge LLC, all rights reserved 37

-2
1x10
+
H Na
He H2
-5
1x10
OH
D2
-8 O2
1x10

H2O Ga
-11
1x10
Au

1x10
-14 Pt
Diffusivity (cm /s)

R
2

-17 As
1x10

B
-20
1x10 Sb

-23
1x10

-26
1x10

-29
1x10

-32
1x10

-35
1x10
600 700 800 900 1,000 1,100 1,200 1,300
o
Temperature ( C)
Figure 1.25: Diffusivity of selected impurities in silicon dioxide versus temperature -
data from table 1.7.

Equation 1.88 for fluorine effects and equation 1.89 for H2 effects on boron diffusion in sil-
icon dioxide are presented in the left and right side of figure 1.26 respectively.
Boron diffusivity in silicon dioxide is also believed to be increased by SiO content and
therefore by decreasing silicon dioxide thickness [66].

1.7. Solid solubility

Steady state solid solubility is defined as:
Copyright © 2000 IC Knowledge LLC, all rights reserved 38

• The maximum concentration of a substance that can be dissolved in a solid at a given tem-
perature.

-13 -14
10 10

-14 -15
10 1,100 C
o 10 o
1,100 C
Boron diffusivity (cm /s)

Boron diffusivity (cm /s)

2

2
-15 -16
10 1,000 C
o 10 o
1,000 C

-16 -17
10 10 o
900 C
o 900 C

-17 -18
10 10
o
800 C
o
-18 800 C -19
10 10

-19 -20
10 10
14 15 16
10 10 10 0.001 0.01 0.1 1 10 100
2
BF2 dose - ␾F2 (atoms/cm ) H2 (%)
Figure 1.26: Effect of fluorine and hydrogen on boron diffusivity in silicon dioxide [66].

The first attempt to summarize experimental data for the solid solubility of impurities in
silicon was performed by Trumbore in 1960 [47]. Recently Borisenko and Yudin reexamined
the solid solubility curves for antimony, arsenic, boron and phosphorus utilizing newer more
accurate data [48] - see figure 1.27.
The data presented in figure 1.27 represents the solid solubility, not the electrically active
dopant. Dopants introduced into silicon may occupy interstitial or substitutional positions in the
silicon lattice, or form clusters, i.e. arsenic - see the next section. The method of introducing a
dopant into silicon and subsequent heat treatments determine the amount of dopant that con-
tributes free carriers, i.e., is electrically active. Prior to annealing a significant percentage of a
dopant may be electrically inactive.

1.7.1. Solid solubility of arsenic in silicon

Arsenic forms clusters when concentrations in silicon are in excess of 1020 atoms/cm3. It
is believed that 3 arsenic atoms and one electron form a cluster that is electrically active at high
temperature and is neutral at room temperature [49].
The model for clustering is

+ - high temp 2+ room temp

3As + e ↔ As 3 ↔ As 3 (1.90)
Copyright © 2000 IC Knowledge LLC, all rights reserved 39

As
Trumbore [47]
P
21 Borisenko and Yudin [48]
10

P
Solubility (atoms/cm )
3

B
As

20
10 Sb

Sb

19
10
400 500 600 700 800 900 1,000 1,100 1,200 1,300 1,400
o
Temperature ( C)
Figure 1.27: Solid solubility of common dopants in silicon.
Summarized from [47],[48].

A generalized model of arsenic clustering has been derived [50], and the expression for
Cmax versus temperature is given by
22 – 0.453 /kT 3
C max = 1.896x10 e atoms/cm (1.91)

1.7.2. Solid solubility of boron in silicon

There is no physical model for electrically active boron concentrations - the values have to
be found experimentally.

1.7.3. Solid solubility of phosphorus in silicon

At high concentration the expression for the total electrically active phosphorus concentra-
tion is given by [44]
– 43 3 3
C T = n + 2.4x10 n atoms/cm (1.92)
Copyright © 2000 IC Knowledge LLC, all rights reserved 40

where, n is the electron concentrations or electrically active phosphorus concentration. Equa-

tion 1.92 is valid from 900 to 1,050oC.
For arsenic, boron, and phosphorus in silicon the maximum electrically active dopants
match up well with the solid solubility curves of Borisenko and Yudin presented in figure 1.25.

1.7.4. Solid solubility of miscellaneous dopants in silicon.

Figure 1.28 present additional solid solubility data covering impurities not included in fig-
ure 1.27. The data in figure 1.28 comes from Trumbore [47], updated curves for these impuri-
ties are not available.

20
10
Sn

Ga
Al
Li
19
10

18
10
3
Atoms/cm

O
Cu

17 Au
10

Zn

10
16 Co

S Ag
Fe

15
10
600 700 800 900 1,000 1,100 1,200 1,300
o
Temperature ( C)
Figure 1.28: Solid solubility of selected impurities in silicon [47].
Copyright © 2000 IC Knowledge LLC, all rights reserved 41

1.8. Deviations from simple diffusion theory

In the preceding sections the basics of diffusion theory have been presented, in this section
the complications of selected real world diffusion problems will be examined.

1.8.1. Redistribution of impurities during thermal oxidation

In section 1.5.2 the solution for two step or drive-in diffusions was presented. One of the
assumptions was constant Q during diffusion. It is very unusual for a drive-in diffusion to be
performed without simultaneous oxidation during at least some part of the cycle. It has been
found that impurities are redistributed in silicon near a growing thermal oxide [55]. Redistribu-
tion is due to several factors. If any two phases - solid, liquid or gas are brought into contact an
impurity found in either of the two phase will redistribute between the two phase until equilib-
rium is reached. In equilibrium the ratio of the concentrations of the impurity in the two phases
will be constant.
• Segregation Coefficient - the ratio of the concentration of an impurity in one phase to the
concentration of the impurity in a second under equilibrium.
The segregation coefficient m, for the Si - SiO2 system is defined as
Equalibrium concentration of impurity in silicon -
m = --------------------------------------------------------------------------------------------------------------------------------------- (1.93)
Equalibrium concentration of impurity in silicon dixoide
Another factor effecting impurity redistribution during oxidation is the diffusivity of the
impurity in silicon dioxide. If the impurity has a high diffusivity in silicon dioxide the impurity
can “escape” through the oxide layer. If the impurity has a low diffusivity in silicon dioxide
then the impurity will be trapped in the silicon. Figure 1.29 illustrates four possible impurity
profiles due to oxidation.
The third factor in redistribution is the moving oxide-silicon interface. Even if an impurity
has a segregation coefficient of 1, the increase in volume occupied by oxide over silicon will
result in a decrease in the impurity concentration in the oxide layer.
Figure 1.29 illustrates four cases of impurity segregation. 1.29a and b illustrate the cases
where oxide takes up the impurity with low diffusivity in the oxide 1.29a, and fast diffusivity in
the oxide 1.29b. 1.29c and d illustrate the cases where the oxide rejects the impurity with low
diffusivity in the oxide 1.29c and fast diffusivity in the oxide 1.29d.
A theory on the redistribution process has been proposed [56]. The assumptions are:
1. The concentration of the impurity at the oxide-gas boundary is constant.
2. Deep inside the silicon the impurity concentration approaches the background concentra-
tion of the silicon.
3. The segregation coefficient ratio is maintained.
4. The impurity is preserved at the oxide-silicon interface.
5. The oxidation rate is parabolic and given by:

xo = Bt (1.94)

The resulting expression for the concentration on the silicon side of the interface is
Copyright © 2000 IC Knowledge LLC, all rights reserved 42

C 1 + ( C O ⁄ C B )λ
------S = ------------------------------------------------------------------------------------------------------------------------------------------------------------
- (1.95)
CB 2
1 + ( 1 ⁄ m – α ) π exp ( α B ⁄ 4D )erfc ( α B ⁄ 4D ) B ⁄ 4D + λ ⁄ m
and
2 2 2
r exp [ ( α r – 1 )B ⁄ 4Dr ]erfc ( α B ⁄ 4D )
λ ≡ ------------------------------------------------------------------------------------------------------ (1.96)
erf ( B ⁄ 4D O )

and

r ≡ DO ⁄ D (1.97)

where, D is the diffusivity in silicon, DO is the diffusivity in oxide, ␣ is the ratio of the thick-
ness of silicon consumed during oxidation to the thickness of oxide produced (0.45). One inter-
esting feature of the theory is that the interface will reach a steady state concentration
irrespective of oxidation time.

Oxide takes up impurity (m < 1)

Silicon Silicon
1.0 1.0
Oxide

Oxide

a) diffusion in C/C b) diffusion in

C/CB B
oxide slow - oxide fast -
Boron Boron with H2

0 x(␮) 1.0 0 x(␮) 1.0

Oxide rejects impurity (m > 1)

Silicon Silicon
1.0 1.0
c) diffusion in
Oxide

Oxide

C/CB oxide slow - C/CB d) diffusion in

Antimony, Arsenic, oxide fast -
Phosphorus Gallium

0 x(␮) 1.0 0 x(␮) 1.0

Figure 1.29: Impurity redistribution in silicon due to thermal oxidation [5].

Table 1.7 presents some values for segregation coefficient.

Based on equations 1.95 through 1.97 and the values in table 1.7 the ratio of the surface
concentration of impurity to the background concentration can be calculated.
Copyright © 2000 IC Knowledge LLC, all rights reserved 43

Table 1.8: Segregation coefficient of impurities at the silicon-silicon dioxide interface.

Impurity Segregation coefficient Reference

As ~10 56
~800 58
~1,000-,2000 59
B ~0.3 56
0.03exp(0.52/kT) for (100) 57
0.05exp(0.52/kT) for (111) 58
~0.1-0.3 59
P ~10 56
~2,000-3,000 59
Sb ~10 56

Figures 1.30 through 1.32 illustrate the segregation phenomena for arsenic, boron and
phosphorus under oxygen and steam oxidation conditions. In each case steam creates a greater
segregation phenomena due to the greater oxidation rate of steam over oxygen. Figure 1.30
through 1.32 are calculated from equation 1.95 using Grove's values for segregation for arsenic
and phosphorus [56] and Kolby and Katz value for boron (100) [57].

5.0

4.0

H2O
3.0
CS/CB

2.0 O2
SiO2 CS
CB

1.0 Original
interface

0.0
900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.30: Surface concentration versus temperature for arsenic following oxidation.
Copyright © 2000 IC Knowledge LLC, all rights reserved 44

0.5

CB
0.4 SiO2 CS
O2

Original
0.3 interface
CS/CB

0.2 H 2O

0.1

0.0
900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.31: Surface concentration versus temperature for boron following oxidation.

5.0

4.0 SiO2 CS
H2O
CB

3.0 Original
CS/CB

interface

2.0

O2
1.0

0.0
900 1,000 1,100 1,200
o
Temperature ( C)
Figure 1.32: Surface concentration versus temperature for phosphorus
following oxidation.
Copyright © 2000 IC Knowledge LLC, all rights reserved 45

The distance over which the segregation effect influences the impurity profile is the diffu-
sion length given by 2 times the square root of Dt. Figure 1.33 illustrates boron segregation ver-
sus oxidation temperature [60].

1.0

920 C
0C
1,00 o
o

C
0 o
o
C
00

10
1,2

1,
0.8

0.6
C/CB

0.4

0.2

0
0 0.5 1.0 1.5 2.0
Distance from the Si-SiO2 interface (␮m)
Figure 1.33: Calculated boron concentration versus distance after dry oxygen oxidation
to produce a 0.2␮m thick oxide at various temperatures [60].

1.8.2. Diffusion through an oxide mask

The diffusion discussions presented in preceding sections are all based on one-dimensional
diffusion. During pre-deposition - the deposition process is made selective by utilizing an oxide
layer to block pre-deposition in unwanted areas and to diffuse the dopant into the silicon
through windows in the oxide where the dopant is desired (see figure 1.34).
During pre-deposition the dopant enters the silicon where the oxide is missing and is
blocked by the oxide where the oxide is present. The dopant diffusing into the silicon verti-
cally, also laterally diffuses under the oxide window. Solving the diffusion equations for the
two dimensional problem reveals that dopant diffuses more slowly parallel to the surface than
normal to the surface (the parallel diffusion is 75 to 85% of the normal diffusion). Figure 1.35
illustrates equal concentrations contours for a pre-deposition (complementary error function)
diffusion - figure 1.35a and a drive-in (gaussian) diffusion - figure 1.35b.

1.8.3. Oxide thickness to mask against diffusion

If a silicon dioxide layer is utilized to create a selective doping process it is necessary to
determine the thickness of oxide required to block the pre-deposition process. In order to calcu-
late the oxide thickness required to block deposition, consider pre-deposition through an oxide
layer schematically illustrated in figure 1.36.
Copyright © 2000 IC Knowledge LLC, all rights reserved 46

Lateral Lateral
diffusion diffusion
Window
Dopant
gas
D D D D D
D
D D D D D
D D D
D D D D
D D D D
D
D D
D
D
D D
Silicon Dioxide D Silicon Dioxide

Diffused
Silicon dopant

Figure 1.34: Pre-deposition through a silicon dioxide window.

Cs=const Total Dopants = const

0 0 C/Cs
C/Cs
0.5 0.9
0.5 0.5 0.75
0.3 0.6
1.0 1.0 0.5
0.1 0.3
y/2 Dt
y/2 Dt

1.5 1.5 0.1

0.03
2.0 0.01 2.0 0.03
0.003 0.01 0.001
2.5 0.001 2.5 0.003
0.0003
0.0001 3.0 0.0003
3.0

3.5 0.0001
3.5
2.5 2.0 1.5 1.0 0.5 0 0.5 1.0 1.5 2.0 2.5 2.5 2.0 1.5 1.0 0.5 0 0.5 1.0 1.5 2.0 2.5

x/2 Dt x/2 Dt

a) Complementary error function b) Gaussian

Figure 1.35: Lateral diffusion profiles [52].

The assumptions used to solve the problem are:

1. The concentration at the surface of the oxide layer Co and the oxide thickness xo are both
constant throughout the pre-deposition process.
2. The concentration goes to zero deep inside the silicon.
3. The ratio of the concentrations on either side of the oxide-silicon interface are determined
by the segregation coefficient m.
4. The flux of impurities through the interface is continuos.
Copyright © 2000 IC Knowledge LLC, all rights reserved 47

C(x, t)

Oxide Silicon
xO xj

CO

CB

x
x=0
Figure 1.36: Impurity distribution from pre-deposition through an oxide [5].

The solution to this problem is given by an infinite series [53]. For small values of x such
that

x « D ⁄ DO xO (1.98)

then the solution may be approximated by the first term in the series [5]

C ( x, t -) = ------------
---------------
2mr xO
- erfc ---------------- x
- + ------------- (1.99)
CO m+r 2 D O t 2 Dt

The solution for the junction depth xj is given by the conditions which when substituted
into equation 1.99 result in
x 1 xO
-----j = – --- -----
-+I (1.100)
t r t

and
( m + r )C B
I ≡ 2 D arg ------------------------- (1.101)
2mrC O

The oxide thickness required to mask against pre-deposition can now be found by setting xj
= 0 in equation 1.100 Figure 1.36 presents empirical data on oxide thickness to mask against
pre-deposition [54].
Copyright © 2000 IC Knowledge LLC, all rights reserved 48

1.8.4. Diffusion from a uniformly doped oxide layer

A doped oxide layer can be deposited on silicon and used as a doping source. Figure 1.38
presents a schematic representation of diffusion from a doped oxide layer.
For relatively short diffusion time such that

xO > 4 DO t (1.102)

then

CO DO ⁄ D
C ( x, t ) ≅ ---------------------------- erfc ⎛ -------------⎞
x
(1.103)
(1 + k) ⎝ 2 Dt⎠

and
1
k = ---- D O ⁄ D (1.104)
m

10
o
Phosphorus - 1,200 C
o
Phosphorus - 1,100 C
o
Phosphorus - 1,000 C
o
Boron - 1,200 C
Oxide Mask Thickness (␮m)

1 o
Phosphorus - 900 C
o
Boron - 1,100 C
o
Boron - 1,000 C

o
0.1 Boron - 900 C

0.01

0.001
10 100 1,000
Time (minutes)
Figure 1.37: Oxide thickness to mask against deposition [54].

For longer diffusion times or thin oxide [61]

C O rm
C ( 0, t ) ≈ -------------- [ 1 – 2erfc ( φ O ) + 2erfc ( 2φ O ) ] (1.105)
m+r
Copyright © 2000 IC Knowledge LLC, all rights reserved 49

and
C O rm
C ( x, t ) ≈ -------------- [ erfc ( φ ) – 2erfc ( φ O + φ ) + erfc ( 2φ O φ ) ] (1.106)
m+r
where
xO
φ O = ----------------
- (1.107)
2 DO t

and
x
φ = ------------- (1.108)
2 Dt

C(x, t)

Oxide Silicon
xO xj

CO

COx (0, t)
C (0, t)

COX, DO

C, D
CB

x
x=0
Figure 1.38: Schematic of diffusion from a uniformly doped oxide layer [61].

1.8.5. Solving for junction depth

It is often useful to define a diffused impurity profile by the junction depth. The junction
depth for pre-deposition is given by
–1
x j = 2 Dt erfc ( C S ⁄ C B ) (1.109)

and for drive-in by

1/2
x j = [ 4Dt ln ( Q ⁄ C S πDt ) ] (1.110)
Copyright © 2000 IC Knowledge LLC, all rights reserved 50

1.8.6. Successive diffusions

During integrated circuit fabrication multiple high temperature processes are used in suc-
cession. Each subsequent thermal process further diffuses any impurities present in the silicon
substrate. The effect of successive diffusions is calculated by summing the Dt as follows

Dt tot = ∑ D1 t1 + D2 t2 + D3 t3 + … (1.111)

1.8.7. Ramping
In order to prevent crystallographic damage and wafer warpage due to thermal gradients,
wafers are most commonly inserted into diffusion furnaces at a relatively low temperature and
then the furnace temperature is increased to the process temperature at a controlled rate -
referred to as ramping the furnace. At the completion of processing the furnace temperature is
once again ramped down to a lower temperature for the wafers to be removed from the furnace.
Most commonly a temperature of 750oC is used for inserting and removing wafers from a dif-
fusion furnace. 750oC provides manageable thermal stresses, is a temperature from which a
furnace can reach a process temperature and return in a reasonable amount of time, and is
above the transition temperature for devitrification of the quartware commonly used in furn-
caes. The ramp rate employed in different types of thermal processors are presented in table
1.7.
Table 1.9: Thermal processor ramp rates

Ramp up Ramp down

Processor rate rate
Conventional batch furnace 5-10oC/min. 2-3oC/min.
Rapid ramp batch furnace 80-100oC/min. 60oC/min.
Single wafer rapid thermal processor 25-100oC/sec. 30-50oC/sec.

For a conventional furnace the time to ramp from 750oC to a process temperature hundreds
of degrees higher can result in significant Dt during ramping, the Dt resulting from furnace
ramping may be calculated as follows [2] (note here that the much faster ramp rates of rapid
ramp furnaces and rapid thermal processor systems allow Dt during ramping to be minimized,
the importance of low Dt ramping will be discussed in chapter 8)
Assume a furnace is being ramped down at a linear rate C such that
T ( t ) = T proc – Ct (1.112)

where, Tproc is the processing temperature from which the ramp occurs.
The effective Dt is therefore
t0

( Dt ) eff = ∫ D ( t ) dt (1.113)
0
Copyright © 2000 IC Knowledge LLC, all rights reserved 51

where the ramp down occurs for a time t0. Given that ramping is typically carried out until the
diffusivity is negligibly small, the integral can be approximated by taking the upper limit as
infinity, since

--- = ------------------------ ≅ ------------ ⎛ 1 + ------------ + …⎞

1 1 1 Ct
(1.114)
T T proc – Ct T proc ⎝ T proc ⎠

and substituting equation 7.114 into the arhenius equation for diffusivity, equation 1.63, gives
2
Ea – ( CE a /kT proc )t
D ( t ) = D 0 exp – --------------- ⎛ 1 + -----------
Ct - + …⎞ = D ( T proc )e (1.115)
kT proc ⎝ T proc ⎠

where, D(Tproc) is the diffusivity at Tproc. Substituting into equation 1.113 and using infinity as
the upper integration limit gives
2
kT proc
( Dt ) eff ≅ D ( T proc ) --------------- (1.116)
CE a

1.8.8. Deposition from a gas phase

Current state-of-the-art integrated circuit fabrication utilizes ion implantation to introduce
impurities into silicon, however trailing edge processes may still utilize gas phase deposition to
dope polysilicon. Legacy processes still in use may also utilize gas phase deposition processes
for general doping.
Henry's law relates the partial pressure of the impurity in the gas phase p to the surface con-
centration, as
C S = Hp (1.117)

where H is Henry’s gas law constant.

The surface concentration therefore depends on the partial pressure of the dopant in the gas
(until the surface reaches solid solubility). Same of the candidate doping sources for gas phase
deposition are listed in table 1.9.
The gaseous sources listed in table 1.9, arsine, diborane and phosphine are generally
avoided for open furnace use due to their high toxicity (these gases are used in closed deposi-
tion systems for in-situ doping - see chapter 10). the most commonly used sources are spin-on
arsenic for buried layers, boron tribromide and phosphorus oxychloride for junction formation.
Boron and phosphorus solid source discs are also widely used.
Boron tribromide and phosphorus oxychloride are both liquids at room temperature. In
order to introduce these sources into a diffusion furnace as a gaseous source an inert gas such as
nitrogen is bubbled through the liquids at a controlled flow rate and temperature. The mole
fraction of dopant entering the gas is given by
pl VC
M = -----------------------------
- (1.118)
p C × 22, 400
Copyright © 2000 IC Knowledge LLC, all rights reserved 52

where, M is the molar flow rate in moles/min., pl is the vapor pressure of the liquid at the bub-
bler temperature in mm of Hg, pC is the pressure of the carrier gas in mm of Hg (mm of Hg
equals PSI multiplied by 51.715), VC is the flow rate of the carrier gas in cm3/min., and 22,400
is the cm3 per mole at standard temperature and pressure.
Table 1.10: Pre-deposition sources

Temperature
Usable State at
of source
Dopant Dopant source Formula concentration room
during
range temperature
deposition
As Arsine AsH3 High Room temp Gas
Arsenic trioxide As2O3 High 200 to 400oC Solid
Spin-on Medium Room temp Liquid
Discs Low to high 850 to Solid
1,200oC
B Boron tribromide BBr3 Low to high 10 to 30oC Liquid
Boron trichloride BCl3 Low to high Room temp Gas
Diborane B2H6 Low to high Room temp Gas
Spin-on Low to high Room temp Liquid
Boron nitride Low to high 850 to Solid
1,200oC
P Phosphorus pentoxide P2O5 Very high 200 to 300oC Solid
Phosphorus Oxychloride POCl3 Low to high 2 to 40oC Liquid
Phosphorus tribromide PBr3 Low to high 170oC Liquid
Phosphine PH3 Low to high Room temp Gas
Spin-on Low to high Room temp Liquid
Discs
Low to high Low to high Solid

Figures 1.39 presents the vapor pressure curves for boron tribromide, phosphorus oxychlo-
ride and phosphorus tribromide.
From equations 1.117 and 1.118 and the data presented in figure 1.39 the appropriate con-
ditions for pre-deposition can be calculated.
The preliminary deposition reaction for phosphorus oxychloride is
4POCl 3 + 3O 2 → 2P 2 O 5 + 6Cl 2 (1.119)

and the preliminary deposition reaction for boron tribromide is

4BBr 3 + 3O 2 → 2B 2 O 5 + 6Br 2 (1.120)

In both reactions a glass layer is formed on the wafer surface and a volatile gas product is
produced. Both chlorine and bromine gas can etch silicon if sufficient oxygen is not present.
The oxygen causes oxide to grow on the wafer surface both incorporating phosphorus or boron
Copyright © 2000 IC Knowledge LLC, all rights reserved 53

and liberating phosphorus or boron to diffuse into the wafer. Following pre-deposition the
wafer surface is covered with a highly doped glass layer that must be removed by a hydrofluo-
ric acid based de-glaze prior to drive-in. During pre-deposition the inside of the furnace tube is
also coated with dopant containing glass. Furnace tubes will typically be saturated with dopant
prior to use for pre-depositions as saturating the tube helps to insure that the dopant is present at
the silicon surface in quantities high enough to insure the silicon is doped to solid solubility.
1,000
atmospheric pressure

BBr3
100
Pressure (mm Hg)

POCl3

PBr3

10

1
0 25 50 75 100 125 150
o
Temperature ( C)
Figure 1.39: Vapor pressure curves for liquid pre-deposition sources [62].

The equations for pre-deposition presented in section 1.6.1 assume that the surface of the
semiconductor is at some concentration CS at time zero. In practice the dopant must be trans-
ported to the semiconductor surface and diffuse into the semiconductor before the initial condi-
tion is met. The time to reach the initial surface doping level is approximately 1.2 minutes
which for short deposition times can be significant [5]. It can be shown that the solution
accounting for the initial transport period is [5].

( x, t -) = erf -------------
2
C x [ ht/ Dt ] ( ht/ Dt ) ( x2/ Dt ) x
ht - + -------------
--------------- +e e erfc --------- (1.121)
CS 2 Dt Dt 2 Dt
where
h = h G ⁄ HkT (1.122)

and, hG is the gas phase mass transfer coefficient.

Copyright © 2000 IC Knowledge LLC, all rights reserved 54

For

ht ⁄ Dt > 10 (1.123)
equation 1.121 converges to a simple error function.

1.8.9. Redistribution of impurities during epitaxial growth

Bipolar and BiCMOS technologies frequently employ a buried layers, i.e. diffused regions
are formed in the substrate followed by epitaxial layer growth (epitaxial layer growth is dis-
cussed in chapter 10). During the high temperature epitaxial growth, the impurities diffuse up
from the substrate into the epitaxial film. Figure 1.40 schematically illustrates impurity redistri-
bution during epitaxial growth.

Substrate Film Gas

Log C

Metallurgical
junction
Initial
distribution Film surface (moving
Csu boundary)
b

C1(x, t) impurity diffusing

from the surface

C2(x, t) external
Cf
doping impurity

x=0 x = xf
Figure 1.40: Impurity redistribution during epitaxial growth [5].

The solution is [5]

C 1 ( x, t ) x ht Vt
------------------ = f 1 -------------, ----------, ---------- (1.124)
C sub 2 Dt Dt Dt
where, V is the film growth rate.
For

Vt ⁄ Dt » 1 (1.125)
equation 1.124 reduces to
Copyright © 2000 IC Knowledge LLC, all rights reserved 55

C 1 ( x, t ) 1 x
------------------ = --- erfc ------------- (1.126)
C sub 2 2 Dt
regardless of the value of h.
The solution to the external impurity profile is given by
C 2 ( x, t ) x Vt
------------------ = f 2 -------------, ---------- (1.127)
Cf 2 Dt Dt
For

Vt ⁄ Dt » 1 and x » 2 Dt (1.128)
then equation 1.127 can be approximated by
C 2 ( x, t ) = C f (1.129)

and the combined solution for the net impurity level is given by
C ( x, t ) = C 1 ( x, t ) −
+ C 2 ( x, t ) (1.130)

where the positive sign is used for impurities of the same type and the minus sign is used for
impurities of opposite type.

1.8.10. Diffusion in an electric field

The preceding derivations of Fick’s law assumed that diffusivity is independent of concen-
tration and that thermal energy is the only driving force for diffusion under the conditions of
interest. In fact diffusing charged dopants atoms can be influenced by electric fields in the sili-
con.
A charged particle in an electric field experiences a force F, given by
F = qE (1.131)
where, q is the charge on the particle and E is the electric field acting on the particle.
A particle of mass, m, undergoes an acceleration
dv qE
------ = ------- (1.132)
dt m
where, v is velocity.
In a solid, the accelerating atom will quickly encounter another atom and undergo elastic
scattering randomizing the direction of motion. Scattering can be represented as a restraining
force Fr, given by [3]
F r = mαv (1.133)

where, ␣ is a proportionality factor related to the collision frequency (collisions/sec.). This

gives
Copyright © 2000 IC Knowledge LLC, all rights reserved 56

dv 1
------ = ---- ( F – F r ) = qE
------- – αv (1.134)
dt m m
setting dv/dt = 0, the steady state or drift velocity, -vd, can be calculated as
qE-
v d = ------- (1.135)
mα
The mobility ␮ is defined as
q
μ ≡ -------- (1.136)
mα
and therefore, combining equations 1.135 and 1.136, yields
v d = μE (1.137)

The flux jd due to the uniform motion of impurity atoms is the velocity multiplied by the
density, so
j d = μCE (1.138)

and jd represents the flux due to diffusion induced by the electric field. Combining the flux due
to the concentration gradient - equation 1.12 with the flux due to the electric field - equation
1.139 results in the total flux

j tot = -D ∂------
C + μCE (1.139)
∂x

1.9. Diffusion Equipment

Generally, diffusion is accomplished in resistance heated furnaces covered elsewhere in the
original book and not repeated here. Arguably some diffusion also occurs in rapid thermal pro-
cessors - at least unintentionally.

1.10. Metrology
In this section the metrology techniques suitable for diffusion characterization are briefly
reviewed. Table 1.11 summarizes some of the more commonly used techniques.

1.10.1. Bevel and Stain

Bevel and stain is a simple, easy to perform technique used to determine junction depth.
Either a cylindrical grinding wheel is used to create a groove through the junction to be mea-
sured or an angled bevel is ground through the junction. A stain solution is then applied which
stains the diffused junction. By knowing the bevel angle or grinding wheel diameter the junc-
tion depth may be calculated by measuring the width of the stained area (see figure 1.41).
Copyright © 2000 IC Knowledge LLC, all rights reserved 57

Table 1.11: Diffusion metrology techniques

Sensitivity Spot size

Technique Quantity measured
(atoms/cm3) (␮m)

Bevel and stain ~1012 ~1mm Junction depth

12
Capacitance voltage ~10 ~0.5mm Electrically active dopant levels
plotting (CV) versus depth
Four point probe ~1012 ~0.5mm Sheet resistance - can be con-
verted to average resistivity if the
junction depth is known
Secondary ion mass ~1016 ~1␮m Atomic concentration and type -
spectroscopy (SIMS) can depth profile
Spreading resistance ~1012 0.1mm Electrically active dopant versus
probe (SRP) depth

The junction depth is calculated by measuring x and applying the appropriate geometry
corrections to determine xj.
x x
Stained junction Stained junction

xj
xj

(a) Bevel and stain (b) Groove and stain

Figure 1.41: Bevel and stain and groove and stain junction depth measurements.

1.10.2. Capacitance voltage plotting (CV)

The reverse bias capacitance of a pn junction can be utilized to measure impurity levels.
The technique requires that a shallow highly doped junction be formed above the impurity pro-
file to be measured. The highly doped junction must be of opposite conductivity type to the
impurity being measured (n+ on p or p+ on n). The voltage on the junction is ramped and the
instantaneous capacitance is measured. The capacitance can be converted to electrically active
impurity concentration by [64]
qε S C B 1/2 1/2
C ( V ) = --------------- V 0 ± ⎛ V R – 2kT
---------⎞ (1.140)
2 ⎝ q ⎠
Copyright © 2000 IC Knowledge LLC, all rights reserved 58

where, ⑀S is the permittivity of silicon, CB is the background concentration, V0 is the junction

potential, and VR is the reverse bias on the junction.
The limitations to the technique are, the diffusion being measured must be deeper than the
shallow diffusion used for measurement plus the zero bias depletion width.
More commonly this technique is applied on an MOS capacitor structure removing the
shallow junction restrictions.

1.10.3. Four Point Probe (FPP)

The four-point-probe technique is utilized to measures the sheet resistance of a diffused junc-
tion. The FPP technique is illustrated in figure 1.42.

Constant Current Supply

Volt Meter

Sample

Figure 1.42: Four point probe - sheet resistance measurement system.

In the system illustrated in figure 1.42 a constant current is supplied to the outer two probes
and the voltage drop is measured across the inner two probes. The sheet resistance is given by
π V V
ρ = -------- --- x j = 4.532 --- x j = R S x j (1.141)
ln 2 I I
where, I is a forced current, V is the measured voltage drop and RS is the sheet resistance.
For a diffused junction the concentration varies with depth so the average resistivity is
given by
xj
xj 1
ρ = ---- ∫ ----------------------- dx (1.142)
q C ( x )μ ( x )
0

where, q is the charge on an electron and ␮(x) is the mobility.Equation 1.142 can be numeri-
cally evaluated for diffused gaussian and error function profiles with varying background con-
centrations. The resulting curves of surface concentration versus average resistivity are known
as Irvin curves after the first researcher to present the technique [51].
Copyright © 2000 IC Knowledge LLC, all rights reserved 59

Figure 1.43 through 1.46 present Irvin curves numerically evaluated by the author utilizing
the mobility equations recommended by NIST.
22
10

21
10

20
10
20
10 14
10 CB
16
CS 10
18
19
10
10

18
10

17
10
-5 -4 -3 -2 -1 0
10 10 10 10 10 10
␳ (⍀-cm)
Figure 1.43: Surface concentration versus average resistivity - n type - erfc profile.
22
10

21
10

20
10
20
10 14 CB
10
16
CS 10
18
10
19
10

18
10

17
10
-5 -4 -3 -2 -1 0
10 10 10 10 10 10
␳ (⍀-cm)

Figure 1.44: Surface concentration versus average resistivity - p type - erfc profile.
Copyright © 2000 IC Knowledge LLC, all rights reserved 60

22
10

21
10

20
10
20
10 14 CB
10
CS 10
16
18
19
10
10

18
10

17
10
-5 -4 -3 -2 -1 0
10 10 10 10 10 10
␳ (⍀-cm)
Figure 1.45: Surface concentration versus average resistivity - n type - gaussian profile.

22
10

21
10

20
10
20
10 14 CB
10
16
CS 10
18
19
10
10

18
10

17
10
-5 -4 -3 -2 -1 0
10 10 10 10 10 10
␳ (⍀-cm)

Figure 1.46: Surface concentration versus average resistivity - p type - gaussian profile.
Copyright © 2000 IC Knowledge LLC, all rights reserved 61

1.10.4. Secondary Ion Mass Spectroscopy (SIMS)

SIMS bombards the section of the wafer to be analyzed with a beam of energetic inert ions.
The ion beam sputters ions from the wafer surface that may be mass analyzed to determine the
atomic composition.
Continuous bombardment of a surface by the ion beam results in etching so that a depth
profile may be determined. The minimum spot size for SIMS is relatively small at 1µm. The
drawbacks to SIMS are relatively low sensitivity - approximately 5 x 1016 atoms/cm3, the
equipment is expensive and difficult to use - only the largest semiconductor companies can
have the technique in-house, and SIMS is destructive. SIMS does however provide atomic con-
centration versus depth and allow atomic species to be identified. Because of SIMS unique
abilities the technique is widely used to analyze doping profile and also to look for contamina-
tion.

1.10.5. Spreading Resistance Probe

To perform an SRP analysis the wafer is first bevel ground and polished along a shallow
angle. A tiny set of conductive needles are dragged along the beveled surface - see figure 1.47.

Probe path

SRP probes

Sample

Figure 1.47: Spreading resistance probe analysis.

The needles have a known current applied and the voltage drop across the needles is mea-
sured. The resistivity in a small volume under the needle is given by
ρ = 2R SR a (1.143)

where, RSR is the spreading resistance value and a is a geometric factor determined by measur-
ing a sample of known resistivity.
From the relationship between concentration and resistivity presented in chapter 1 the car-
rier concentration can be calculated. If the motion of the probes across the bevel is well con-
trolled and the bevel angle is known, the profile versus depth can be calculated. SRP has a
couple of limitations, one, only the electrically active dopant is measured so dopants must be
activated prior to SRP (this is in contrast to SIMS that measures atomic concentration regard-
Copyright © 2000 IC Knowledge LLC, all rights reserved 62

less of electronic state), secondly, SRPs require a fairly large measurement area, and third SRP
is destructive and very technique sensitive. Most smaller semiconductor companies send out
for SRP adding longer turn-around time to the draw backs.

References
[1] W. Scott Ruska, “Microelectronic Processing - An Introduction to the Manufacture of
Integrated Circuits,” McGraw-Hill Inc. (1987).
[2] Sorab K. Ghandhi, “VLSI Fabrication Principle: Silicon and Gallium Arsenide,” John
Wiley, (1983).
[3] Sorab K. Ghandhi, “The Theory and Practice of Microelectronics,” John Wiley, (1968).
[4] J. Crank, “The Mathematics of Diffusion,” Oxford University Press, (1956).
[5] A.S. Grove, “Physics and Technology of Semiconductor Devices,” John Wiley (1967).
[6] R.B. Fair, “Concentration Profiles of Diffuse Dopants in Silicon,” in F.Y.Y. Yang, Ed.
“Impurity Dopant Processes in Silicon,” North Holland (1981).
[7] R.B. Fair, “Diffusion and Ion Implantation,” in Gary E. McGuire Ed. “Semiconductor
Materials and Process Technology,” Noyes Publications (1988).
[8] R.B. Fair, J. Appl. Phys., 51, 5828 (1980).
[9] D.A. Antoniadis and I. Moskowitz, J. Appl. Phys., 53, 9214 (1982).
[10] S. Matumoto, Y Ishikawa and T. Niimi, J. Appl. Phys. (1983).
[11] U. Goesle, T.Y. Tan, “Defects in Semiconductors II,” S. Mahajan and J.W Corbett Ed.
North-Holland Press (1983).
[12] D. Mathiot and J.C. Pfister, J. Appl. Phys., 55, 3518 (1984).
[13] Taniguchi, et.al.
[14] A.M. Lin, R.W. Dutton, and D.A. Antoniadis, Paper 133 presented at the Electrochemi-
cal Society Meeting, Boston, Massachusetts, May 6-11 (1979).
[15] Fahey and Dutton.
[16] R.B. Fair and J.C.C. Tsai, J. Electrochem. Soc., 124, 1107 (1977).
[17] P. Fahey, R.W. Dutton and S.M. Hu, Appl. Phys. Lett., 44, 777 (1984).
[18] R.M. Harris and D.A. Antoniadis, Appl. Phys. Lett., 43, 937 (1983).
[19] E. Guerrero, Ph.D. Thesis, University of Vienna (1984).
[20] U. Gosele and T.Y. Tan, “The Influence of Point Defects on Diffusion and Gettering in
Silicon,” Mat. Res. Soc. Symp. Proc. Vol. 36 (1985).
[21] R.B. Fair, “The Effect of Strain-Induced Bandgap Narrowing on High Concentration
Phosphorus in Silicon,” J. Appl. Phys., 50, 860 (1979).
[22] Yoshiaki Shibata, Seiichi Hashimoto, Kenji Taniguchi and Chihiro Hamaguchi, “Oxida-
tion Enhanced Diffusion of Phosphorus over a Wide Range of Oxidation Rates,” J. Elec-
trochem. Soc. 139, 231 (1992).
[23] B.L. Sharma, “Diffusion in Semiconductors,” Trans. Tech. Pub. Germany, 87 (1970).
[24] W. Wurker, K. Roy, and J. Hesse, “Diffusion and Solid Solubility of Chromium in Sili-
con,” Mater. Res. Bull. (U.S.A.), 9, 971 (1974).
[25] H. Kitagano and K. Hashimoto, “Diffusion Coefficient of Cobalt in Silicon,” J. Appl.
Phys. Jpn., 16, 173 (1977).
Copyright © 2000 IC Knowledge LLC, all rights reserved 63

[26] J.C. Mikkelsen, Jr., “Diffusivity of Oxygen in Silicon During Steam Oxidation,” J. Appl.
Phys. Jpn., 16, 173 (1977).
[27] R.F. Bailey and T.G. Mills, “Diffusion Parameters of Platinum in Silicon,” in R.R.
Habarecht and E.L. Kern, Eds., Semiconductor Silicon 1969, Electrochem. Soc., (1969).
[28] D.L. Kendall and D.B. DeVries, “Diffusion in Silicon,” ibid.
[29] V.E. Borisenko, L.F.Gorskaya, A.G. Dutov and V.A. Samuilov, “Povedenie implantiro-
vannogo fosfora v polikristallicheskom kremnii pri impulsnoii termoobrabotke,” Eleck-
tronnaya Tekhnika, Ser. 2, Poluprovodnikovie Pribori 2, 53 (1987).
[30] M. Takai, M. Izumi, K. Matnunga, K. gamo, S. Namba, T. Minamisono, M. Miyauchi,
and T. Hirao, “Backscattering study of implanted arsenic distribution in poly-silicon on
insulator,” Nucl. Instrum. Methods Phys. Res. B 19/20(1), 603 (1987).
[31] M. Takai, M. Izumi, T. Yamamoto, S. Namba, and T. Minamisono, “Rapid thermal
annealing of arsenic-implanted poly-Si layers on insulator,” Nucl. Instrum. Methods
Phys. Res. B 39(1-4), 352 (1989).
[32] A.D. Buonaquisti, W. Carter, and P.H. Holloway, “Diffusion characteristics of boron
and phosphorus in polycrystalline silicon,” Thin Solid Films, 100(3), 235 (1983).
[33] T.I. Kamins, J. Manolin, and R.N. Tucker, “Diffusion of impurities in polycrystalline sil-
icon,” J. Appl. Phys., 43(1), 83 (1972).
[34] H. Bomgart, H.J. Leamy, G.K. Cellar, and L.E. Trimble, “Grain boundary diffusion in
polycrystalline silicon film on SiO2,” J. Phys. (Paris) 43(10), c1/363 (1982).
[35] D.L. Losee, J.P. Lavine, E.A. Trabka, S.T. Lee, and C.M. Jarman, “Phosphorus diffu-
sion in polycrystalline silicon,” J. Appl. Phys., 55(4), 1218 (1984).
[36] H. Ryssel, H. Iberl, M. Bleier, G. Prinke, K. Haberger, and H. Kranz, “Arsenic -
implanted polysilicon layers,” Appl. Phys. 24(3), 197 (1981).
[37] B. Swaminathan, K.C. Saraswat, R.W. Dutton, and T.I. Kamins, “Diffusion of arsenic in
polycrystalline silicon,” J. Appl. Phys. Lett., 40(9), 795 (1982).
[38] M. Arienzo, Y. Komen, and A.E. Michel, “Diffusion of arsenic in bilayer polycrystal-
line silicon films,” J. Appl. Phys., 55(2), 365 (1984).
[39] K. Tsukamoto, Y. Akasaka, and K. Horie, “Arsenic implantation into polycrystalline sil-
icon and diffusion to silicon substrate,” J. Appl. Phys. 48(5), 1815 (1977).
[40] F.H.M. Spit, H. Albers, A. Lubbes, Q.J.A. Rijke, L.J. Van Ruijven, J.P.A. Westerveld,
H. Bakker, and S. Radelaar, “Diffusion of antimony (125Sb) in polycrystalline silicon,”
Phys. Status Solidi A, 89(1) 105 (1985).
[41] Victor E. Borisenko and Peter J. Hesketh, “Rapid Thermal Processing of Semiconduc-
tor” Plenum Press (1997).
[42] M. Ghezzo and D.M. Brown, “Diffusivity Summary of B, Ga, P, As, and Sb in SiO2,” J.
Electrochem. Soc., 120, 146 (1973).
[43] Y. Wada and D.A. Antoniadis, “Anomalous Arsenic Diffusion in Silicon Dioxide,” J.
Electrochem. Soc., 128, 1317 (1981).
[44] J.C.C. Tsai, “Diffusion,” S.M. Sze Ed. “VLSI Technology,” McGraw Hill (1983).
[45] R.N. Ghoshtagore, “Silicon Dioxide Masking of Phosphorus Diffusion in Silicon,” Solid
State Electron., 18, 399 (1975).
Copyright © 2000 IC Knowledge LLC, all rights reserved 64

[46] C.Y. Bartholomew “private communications” as reported in J.C.C. Tsai, “Diffusion,”

W.E. Beadle, J.C.C. Tsai and R.D. Plummer Eds. “Quick Reference Manual for Silicon
Integrated Circuit Technology,” John Wiley (1985).
[47] F.A. Trumbore, “Solid solubilities of impurity elements in germanium and silicon,”,
Bell Syst. tech. J., 39, 205 (1960).
[48] V.E. Borisenko and S.G. Yudin, “Steady-state solubility of substitutional impurities in
silicon,” Phys. Status Solidi, A 101(1) 123 (1987).
[49] M.Y. Tsai, F.F. Morehead, and J.E.E. Baglin, “Shallow Junctions by High Dose As
Implants in Si: Experiments and Modeling,” J. Appl. Phys., 51, 3230 (1980).
[50] E. Guerrero, H. Potzl, R. Tielert, M. Grasserbauer, and G. Stingeder, “Generalized
Model for the Clustering of As Dopant in Si,” J. Electrochem. Soc., 129, 1826 (1982).
[51] J.C. Irvin, “Resistivity of Bulk Silicon and of Diffused Layers in Silicon,” Bell System
Tech. J., 41, 387 (1962).
[52] D.P. Kennedy and R.R. O'Brien, “Analysis of the Impurity Atom Distribution Near the
Diffusion Mask for a Planar p-n Junction,” IBM J. Res. Dev., 9(3), 179 (1965).
[53] C.T. Shaw, H. Sello, and D.A. Tremere, “Diffusion of Phosphorus in Silicon Dioxide
Film,” J. Phys. Chem. Solids, 11, 288 (1959).
[54] O.D. Trapp, “Semiconductor Technology Handbook,” Technology Associates (1993).
[55] M.M. Atalla and E. Tannenbaum, “Impurity Redistribution and Junction Formation in
Silicon by Thermal Oxidation,” Bell System Tech. J., 39, 933 (1960).
[56] A.S. Grove, O.Leistiko, and C.T. Sah, “Redistribution of Acceptor and Donor Impurities
During Thermal Oxidation of Silicon,” J. Appl. Phys., 35, 2695 (1964).
[57] J.W. Colby and L.E. Katz, “Boron Segregation at Si-SiO2 Interface as a function of
Temperature and Orientation,” J. Electrochem. Soc., 123, 409 (1976).
[58] R.B. Fair and J.C.C. Tsai, J. Electrochem. Soc. 122, 1689 (1975).
[59] K. Sakamoto, K. Nishi, F. Ichikawa and S. Ushio, “Segregation and transport coeffi-
cients of impurities at the Si/SiO2 interface,” J. Appl. Phys., 61, 1553 (1987).
[60] B.E. Grove, A.S. Grove, E.H. Snow, and C.T. Sah, “Observation of Impurity Redistribu-
tion During Thermal Oxidation of Silicon Using the MOS Structure,” J. Electrochem.
Soc., 112, 308 (1965).
[61] J.C.C. Tsai, “Diffusion,” W.E. beadle, J.C.C. Tsai and R.D. Plummer, Eds. “Quick Ref-
erence Manual for Silicon Integrated Circuit Technology,” John Wiley (1985).
[62] J.C. Schumacher Company applications notes.
[63] Scotten W. Jones, “Process technology for the 21st Century,” Semiconductor Consult-
ing Services (1999).
[64] S. Wolf and R.N. Tauber, “Silicon processing for the VLSI Era: Volume 1 - Process
Technology,” Lattice Press (1986).
[65] R.L. Burden, J.D. Faires and, A.C. Reynolds, “Numerical Analysis,” 2nd ed. Prindle
Weber and Schmidt (1981).
[66] Richard B. Fair, “Unified Model of Boron Diffusion in Thin Gate Oxides: Effects of F,
H2, N, Oxide Thickness and Injected Si Interstitials,” IEDM 95, 85, (1995).
Copyright © 2000 IC Knowledge LLC, all rights reserved 65

[67] H.J. Grossman, “Dopants and Intrinsic Point-Defects During Si Device Processing,” in
H.R. Huff, U. Gosele and H. Tsuya, eds. “Semiconductor Silicon 1998,” Electrochem.
Soc. Proc., 98-1, (1998).
[68] M.D. Giles, Apld. Phys. Lett., 62, 1940 (1993).
[69] T. Diaz de la Rubia and G.H. Gilmer, Phys. Rev. Lett. 74, 2507 (1995).
[70] James D. Plummer, “Point Defect Based Modeling of Dopant Diffusion and Transient
Enhanced Diffusion in Silicon,” in H.R. Huff, U. Gosele and H. Tsuya, eds. “Semicon-
ductor Silicon 1998,” Electrochem. Soc. Proc., 98-1, (1998).
[71] E. Guerrero, W. Jungling, H. Potzl, U. Gosele, L. Mader, M. Grasserbauer, and G.
Stingeder, “Determination of the Retarded Diffusion of Antimony by SIMS Measurement
and Numerical Simulation,” J. Electrochem. Soc., 133, 2181 (1986).

Example Problems
1. A pre-deposition cycle is desired to produce a phosphorus diffusion with a surface concen-
tration of 2 x 1020 atoms/cm3 and a junction depth of 0.5␮ms when diffused into a back-
ground concentration of 1 x 1018 atoms/cm3.
• From figure 7.25 find the temperature where phosphorous has a solid solubility of 3 x 1020
atoms/cm2 - ~ 900oC.
• To determine the pre-deposition time required to produce a 0.2␮m junction depth at the
temperature determined in step 1. First divide the background concentration by the surface
concentration, 1 x 1018/3 x 1020 = 3.3 x 10-3. Next using figure 7.11 find the value of z for
an error function that corresponds to 3.3 x 10-3, ~2.15. Convert z to the required Dt, (x/2z)2
= Dt -> 2.16 x 10-11 cm2. Find the diffusivity of phosphorus for 900oC (this is complicated
by the concentration dependence of phosphorus diffusivity, for simplicity assume 5 x 1019
as an average concentration). From figure 7.18 the diffusivity is ~2 x 10-15. Dividing 2.16 x
10-11 by 2 x 10-15 gives 10,817 secs or ~180mins.
• To determine the oxide thickness required to mask against a 180min phosphorus pre-depo-
sition at 900oC, from figure 7.37, ~0.3µms.
2. A drive-in cycle is desired to produce a surface concentration of 1 x 1018 and a junction
depth of 1µm when diffused into a background concentration of 1 x 1014.
• To determine the required Dt, divide the background concentration by the surface concen-
tration, 1 x 1014/ 1 x 1018 -> 1 x 10-4. Look up z for a gaussian profile for 1 x 10-4, from
figure 7.11, ~3.1. Converting z to Dt (see above) gives 2.6 x 10-10. Using figure 7.15 for
the intrinsic diffusivity of boron the time and temperature can be contrasted to produce an
acceptable cycle length, ~1,060 gives a diffusivity of ~5 x 10-14 and a cycle time of
~87mins.
• To determine the required pre-deposition doping level for a 1 x 1018 surface concentration
after a Dt of 2.6 x 10-10. Rearranging equation 7.40 gives CS(␲Dt)1/2 = Q -> 2.86 x 1013.
Copyright © 2000 IC Knowledge LLC, all rights reserved 66

Problems
1. Calculate a pre-deposition cycle to produce a boron diffusion with a surface concentration
of 1 x 1018 atoms/cm3, and a junction depth of 0.5␮ms into a background concentration of
1 x 1014 atoms/cm3.
2. Calculate the junction depth for the pre-deposition calculated in 1) following a 5 hour diffu-
sion at 1,050oC in N2.
3. Calculate the junction depth for the pre-deposition calculated in 1) following a 5 hour diffu-
sion at 1,050oC in dry O2.
4. Calculate the surface concentration of boron under the conditions outlined in 1) and in 2).
Describe why the surface concentration is different.
5. Calculate the minimum carrier gas flow rate for BBr3 to meet the pre-deposition conditions
outline in 1).
6. Calculate the lateral diffusion of a junction formed under the conditions outlined in 2).
7. Calculate the effective Dt for an arsenic diffusion ramped from 750oC to 1,050oC at 5oC/
min.
 67

Appendix

Properties of the error function

∞
2 3 5 7
2 2 x x x
∫
-a
erf ( x ) ≡ ------- e da = ------
- x – -----------
- + -----------
- – -+…
----------- (F.1)
π π 3 ⋅ 1! 5 ⋅ 2! 7 ⋅ 3!
0

erfc ( x ) ≡ 1 – erf ( x ) (F.2)

erf ( 0 ) = 0 (F.3)

erf ( ∞ ) = 1 (F.4)
2
erf ( x ) ≅ ------- x for x << 1 (F.5)
π
2
-x
1 e
erfc ( x ) ≅ ------- -------- for x >> 1 (F.6)
π x

d erf ( x )- = ------
-------------------
2 - -x 2
e (F.7)
dx π
∞
2
1
∫
-x
erfc ( x' ) d x' = x erfc x + ------
- ( 1 – e ) (F.8)
π
0
∞
1
∫ erfc ( x ) dx = -------
π
(F.9)
0
68 Properties of the Error Function

Table E.1: Values of the complementary error function (erfc)

z erfc(z) z erfc(z) z erfc(z)

0.0 1.00000000 1.3 0.06599207 2.6 0.00023603
0.1 0.88753708 1.4 0.04771489 2.7 0.00013433
0.2 0.77729741 1.5 0.03389486 2.8 0.00007501
0.3 0.67137324 1.6 0.02365162 2.9 0.00004110
0.4 0.57160765 1.7 0.01620954 3.0 0.00002209
0.5 0.47950012 1.8 0.01090950 3.1 0.00001165
0.6 0.39614391 1.9 0.00720957 3.2 0.00000603
0.7 0.32219881 2.0 0.00467774 3.3 0.00000306
0.8 0.25789921 2.1 0.00297947 3.4 0.00000152
0.9 0.20309189 2.2 0.00186285 3.5 0.00000074
1.0 0.15729926 2.3 0.00114318 3.6 0.00000036
1.1 0.11979496 2.4 0.00068851 3.7 0.00000017
1.2 0.08968604 2.5 0.00040695 3.8 0.00000008