1

Chapter 8

CHAPTER 8: Diffusion
Diffusion and ion implantation are the two key processes to introduce a
controlled amount of dopants into semiconductors and to alter the conductivity
type. Figure 8.1 compares these two techniques and the resulting dopant
profiles. In the diffusion process, the dopant atoms are introduced from the gas
phase of by using doped-oxide sources. The doping concentration decreases
monotonically from the surface, and the in-depth distribution of the dopant is
determined mainly by the temperature and diffusion time. Figure 8.1b reveals
the ion implantation process, which will be discussed in Chapter 9. Generally
speaking, diffusion and ion implantation complement each other. For instance,
diffusion is used to form a deep junction, such as an n-tub in a CMOS device,
while ion implantation is utilized to form a shallow junction, like a source / drain
junction of a MOSFET.

Boron is the most common p-type impurity in silicon, whereas arsenic and
phosphorus are used extensively as n-type dopants. These three elements are
highly soluble in silicon with solubilities exceeding 5 x 1020 atoms / cm3 in the
diffusion temperature range (between 800oC and 1200oC). These dopants can be
introduced via several means, including solid sources (BN for B, As2O3 for As,
and P2O5 for P), liquid sources (BBr3, AsCl3, and POCl3), and gaseous sources
(B2H6, AsH3, and PH3). Usually, the gaseous source is transported to the
semiconductor surface by an inert gas (e.g. N2) and is then reduced at the surface.
 2
Chapter 8

Figure 8.1: Comparison of (a) diffusion and (b) ion implantation for the
selective introduction of dopants into a semiconductor substrate.
 3
Chapter 8

8.1 Diffusion Theory

Diffusion in a semiconductor can be envisaged as a series of atomic movement of

the diffusant (dopant) in the crystal lattice. Figure 8.2 illustrates the two basic
atomic diffusion mechanisms. The open circles represent the host atoms
occupying the equilibrium lattice positions. The solid dots represent impurity
atoms. At elevated temperature, the lattice atoms vibrate around the equilibrium
lattice sites. There is a finite probability that a host atom can acquire sufficient
energy to leave the lattice site and to become an interstitial atom thereby creating
a vacancy. When a neighboring impurity migrates to the vacancy site, as shown
in Figure 8.2a, the mechanism is called vacancy diffusion. If an interstitial atom
moves from one place to another without occupying a lattice site (Figure 8.2b),
the mechanism is interstitial diffusion.

Figure 8.2: Models of atomic diffusion mechanisms for a two-dimensional

lattice, with a being the lattice constant: (a) Vacancy mechanism. (b)
Interstitial mechanism.

The basic diffusion process of impurity atoms is similar to that of charge carriers.
Let F be the flux of dopant atoms traversing through a unit area in a unit time,
and

C
F  D (Equation 8.1)
x

where D is the diffusion coefficient, C is the dopant concentration, and x is the

distance in one dimension. The equation imparts that the main driving force of
 4
Chapter 8

C
the diffusion process is the concentration gradient, . In fact, the flux is
x
proportional to the concentration gradient, and the dopant atoms will diffuse
from a high-concentration region toward a low-concentration region. The
negative sign on the right-hand-side of Equation 8.1 states that matters flow in
the direction of decreasing dopant concentration, that is, the concentration
gradient is negative.

According to the law of conservation of matter, the change of the dopant

concentration with time must be equivalent to the local decrease of the diffusion
flux, in the absence of a source or a sink. Thus,

C F  C
  (D ) (Equation 8.2)
t x x x

When the concentration of the dopant is low, the diffusion constant at a given
temperature can be considered as a constant and Equation 8.2 can be written as:

C  2C
D 2 (Equation 8.3)
t x

Equation 8.3 is often referred to as Fick's Second Law of Diffusion.

Figure 8.3 displays the measured diffusion coefficients for low concentrations of
various dopant impurities in silicon and gallium arsenide. The logarithm of the
diffusion coefficients plotted against the reciprocal of the absolute temperature
yield a straight line in most of the cases, implying that over the temperature
range, the diffusion coefficients can be expressed as:
Ea

D  Do e kT
(Equation 8.4)

where Do denotes the diffusion coefficient extrapolated to infinite temperature

and Ea stands for the Arrhenius activation energy.

For interstitial diffusion, Ea is related to the energy required to move a dopant

atom from one interstitial site to another. The values of Ea are between 0.5 to 1.5
eV in both Si and GaAs. For vacancy diffusion, Ea is related to both the energies
of motion and formation of vacancies. Hence, Ea for vacancy diffusion is larger
than that for interstitial diffusion and is usually between 3 to 5 eV. For fast
diffusing species such as Cu, the measured activation energy is less than 2 eV,
implying that interstitial atomic movement is the dominant diffusion mechanism.
 5
Chapter 8

For slow diffusing species like As, Ea is higher than 3 eV, and vacancy diffusion
is naturally the dominant mechanism.

Figure 8.3: Diffusion coefficient (also called diffusivity) as a function of the

reciprocal of temperature for (a) silicon and (b) gallium arsenide.
 6
Chapter 8

8.2 Diffusion Profiles

The diffusion profile of dopant atoms is dependent on the initial and boundary
conditions. Solutions for Equation 8.3 have been obtained for various simple
conditions, including constant-surface-concentration diffusion and constant-total-
dopant diffusion. In the first scenario, impurity atoms are transported from a
vapor source onto the semiconductor surface and diffuse into the semiconductor
wafer. The vapor source maintains a constant level of surface concentration
during the entire diffusion period. In the second situation, a fixed amount of
dopant is deposited onto the semiconductor surface and is subsequently diffused
into the wafer.
 7
Chapter 8

8.3 Constant-Surface-Concentration Diffusion

The initial condition at t = 0 is C(x, 0) = 0 which states that the dopant

concentration in the host semiconductor is initially zero. The boundary
conditions are:

C(0, t) = Cs and C(, t) = 0

where Cs is the surface concentration (at x = 0) which is independent of time.

The second boundary condition states that at large distances from the surface,
there are no impurity atoms. The solution of the differential equation that
satisfies the initial and boundary conditions is given by:

 x 
C ( x, t )  C s erfc   (Equation 8.5)
 2 Dt 

where erfc stands for the complementary error function, Dt is the diffusion
length, x is the distance, D is the diffusion coefficient, and t is the diffusion time.
Some properties of the erfc are summarized in Table 8.1.

The diffusion profile for the constant-surface-concentration condition is

exhibited in Figure 8.4(a) on both linear and logarithmic scales. The total
number of dopants per unit area of the semiconductor, Q(t), is given by
integrating C(x, t) from x = 0 to x = :

2
Q (t )  C s Dt  (1.13)C s Dt (Equation 8.6)


dC
The gradient of the diffusion, , can be obtained by differentiating Equation
dx
8.5, and the result is:

dC Cs   x2 
  exp  (Equation 8.7)
dx Dt  4Dt 
 8
Chapter 8

Table 8.1: Error Function Algebra.

2 x
  e y dy
2
erf (x)
 0

erfc (x)  1 - erf(x)

erf (0) = 0
erf () = 1
2
erf (x)  x for x  1

1 e x
2

erfc (x)  for x  1

 x
d 2 x2
erf (x) = e
dx 
d2 4
=  xe x
2
erf (x)
dx 2 
x 1
 erfc( y') dy' = x erfc( x)  (1  e x )
2

0

 1
 erfc( x) dx
0
=

 9
Chapter 8

Figure 8.4: Diffusion profiles. (a) Normalized complementary error function

(erfc) versus distance for successive diffusion times. (b) Normalized
Gaussian function versus distance for successive times.
 10
Chapter 8

8.4 Constant-Total-Dopant Diffusion

For this case, a fixed (or constant) amount of dopant is deposited onto the
semiconductor surface in a thin layer, and the dopant is subsequently diffused
into the semiconductor. The initial condition at t = 0 is again C(x, 0) = 0. The
boundary conditions are:

 C( x, t )
0
dx  S and C( , t )  0

where S is the total amount of dopant per unit area. The solution of the diffusion
equation satisfying the above conditions is:

S   x2 
C( x, t )  exp  (Equation 8.8)
Dt  4Dt 

This expression is the Gaussian distribution, and the dopant profile is displayed
in Figure 8.4b. By substituting x = 0 into Equation 8.8:

S
Cs ( t )  (Equation 8.9)
Dt

The dopant surface concentration therefore decreases with time, since the dopant
will move into the semiconductor as time increases. The gradient of the diffusion
profile is obtained by differentiating Equation 8.8:

dC x
  C( x , t ) (Equation 8.10)
dx 2Dt

The gradient is zero at x = 0 and x = , and the maximum gradient occurs at x =

2 Dt .

Both the complementary error function and the Gaussian distribution are
x
functions of a normalized distance, . Hence, if we normalize the dopant
2 Dt
concentration with the surface concentration, each distribution can be represented
by a single curve valid for all diffusion times, as shown in Figure 8.5.
 11
Chapter 8

Figure 8.5: Normalized concentration versus normalized distance for the erfc
and Gaussian functions.
 12
Chapter 8

Example 8.1

Phosphorus is added to a silicon wafer from a gaseous source at 975 oC for 30

minutes. Determine the junction depth for: (a) 0.3 -cm p-type substrate and
(b) 20 -cm p-type substrate. Assume that the diffusion coefficient of
phosphorus is 10-13 cm2 s-1 and that its solid solubility is 1021 cm-3 at 975oC.

Solution

The diffusion mechanism is governed by the complementary error function:

 x 
C ( x, t )  C S erfc  
 2 Dt 

CS = 1021 cm-3; D = 10-13 cm2 s-1 at 975oC

(a)
Resistivity = 0.3 -cm p-type or NA = 6.3 x 1016 cm-3 from Fig. 1.11
C / CS = 6.3 x 1016 / 1021 = 6.3 x 10-5
x
= 2.83 from Fig. 8.5
2 Dt
xj = (2.83)(2)[(10-13)(1800)]1/2 = 7.6 x 10-5 cm = 0.76 m

(b)
Resistivity = 20 -cm p-type or NA = 6.8 x 1014 cm-3 from Fig. 1.11
C / CS = 6.8 x 1014 / 1021 = 6.8 x 10-7
x
= 3.5 from Fig. 8.5 (Extending the erfc line linearly on a log scale is
2 Dt
reasonable in this range)

xj = (3.5)(2)[(10-13)(1800)]1/2 = 9.4 x 10-5 cm = 0.94 m

 13
Chapter 8

8.5 Dual Diffusion Process

In VLSI processing, a two-step diffusion sequence is commonly used, in which a

predeposition diffusion layer is formed under a constant-surface-concentration
condition and is followed by a drive-in diffusion or redistribution under a
constant-total-dopant condition. For most practical cases, the diffusion length
Dt for the predeposition diffusion is much smaller than that for the drive-in
condition. Hence, the predeposition profile can be treated as a delta function at
the surface.
 14
Chapter 8

Example 8.2

(a) A predeposition process is carried out for 15 minutes on an n-type silicon

wafer with a phosphorus dopant concentration of 10 17 atoms/cm3 at 950oC
using diborane gas. Assuming intrinsic diffusion, determine the junction depth.
At 950oC the boron saturation concentration is 3.8 x 10 20 atoms/cm3 and the
boron diffusion constant is 1.5 x 10-15 cm2/s.

(b) After the initial predeposition process described in (a), the sample
undergoes a drive-in diffusion for 1 hour at 1250 oC. What is the final junction
depth? At 1250oC, the boron diffusion constant is 1.2 x 10-12 cm2/s.

Solution

(a)
The junction depth is determined by the point of transition from p-type to
n-type silicon. For a predeposition process, the diffusion behavior is given
by the complementary error function. Therefore, the junction depth, xj, is
determined by:

 x 
C ( x j , t )  C s erfc  j   1017
 2 Dt 
 xj  1017
erfc    20
 2.63 x10 4
 2 Dt  3.8 x10
c x
From Fig. 8.5, for  2.63x10 4 ,  2.6
cs 2 Dt
Thus, x j  2.6 x2 x (1.5x10 15 )(900) = 6.04x10-6 cm or 0.06 m.

(b)
The total integrated dose, S, in the predeposition process is:

2 2
S Cs Dt = (3.8 x10 20 ) (1.5x10 15 )(900) = 5x1014
 1.77
 15
Chapter 8

Since the drive-in time and temperature (D1250C >> D950C) is much larger
than those in the predeposition process, the boron distribution resulting
from the predeposition can be assumed to be a delta function. For the
drive-in process, the dopant profile is given by the Gaussian distribution:

S  x2 
C ( x, t )  exp  .
Dt  4 Dt 

5 x1014  xj
2

Thus, 10  17
exp  12 
 (1.2 x10 12 )(3600)  4(1.2 x10 )(3600) 
  x j 2 
10  4.3x10 exp 
17 18
8 
1.7 x10 

2
  x j 2 
2.3x10  exp  8 
1.7 x10 

 xj
2

 3.76 
1.7 x10 8

Thus, xj = 2.5 x 10-4 cm or 2.5 m.

 16
Chapter 8

Example 8.3

(a) A predeposition process is carried out for 15 minutes on an n-type silicon

wafer with a bulk doping concentration of 10 17 atoms/cm3 at 950oC using
diborane gas. Determine the p-n junction depth given that the surface solubility
of boron at 950oC is 3.8 x 1020 atoms/cm3, the intrinsic diffusivity (Do) is 0.76
cm2/s, and the Arrhenius activation energy (Ea) is 3.46 eV.

(b) A drive-in process is subsequently performed on this sample after the

deposition process. In order to produce a p-n junction 1.28 μm below the wafer
surface, estimate the required Dt, the product of the diffusion coefficient and
time. Based on your result, determine the time required if the drive-in
temperature is 1250oC.

Solution

(a)
Using Equation 8.4,
3, 46

( 8.36 x105 )(1223)
D  0.76e  1.527 x10 15 cm2/s

For the predeposition, the distribution is given by the erfc function.

Cs = 3.8x1020 atoms/cm3
At the junction depth, C = 1017 atoms/cm3
C 1017
Therefore,  20
 2.63x10 4
C s 3.8 x10

x
Using the erfc curve in Fig. 8.5, the corresponding value of is 2.58.
2 Dt
xj xj
Thus,  2.58  .
2 Dt 2 (1.527 x10 15 )(15 x60)

The junction depth, xj = 6.05x10-6 cm = 0.0605 μm.

 17
Chapter 8

(b)
The drive-in process yields the Gaussian distribution and the amount of
dopant introduced during the predeposition process is given by Equation
8.6:
Q  (1.13)C s Dt  (1.13)(3.8 x10 20 ) (1.527 x10 15 )(15 x60)  5.03x1014 cm-2

Substituting this value of Q or S into Equation 8.8,

S  x j 2  5.03 x1014  1.28 x10 4 
C( x j , t)  exp    exp    10
17

Dt  4 Dt  Dt  4 Dt 

Using trial and error gives Dt = 9 x 10-9 cm2

3, 46

( 8.36 x105 )(1523)
D  0.76e  1.193x10 12 .
9 x10 9
Hence, t   7.54 x10 3 seconds or 126 minutes.
D
 18
Chapter 8

8.6 Extrinsic Diffusion

Diffusion that occurs when the doping concentration is lower than the intrinsic
carrier concentration, ni, at the diffusion temperature is called intrinsic diffusion.
In this region, the resulting dopant profiles of sequential or simultaneous
diffusion of n-type or p-type impurities can be determined by superposition, that
is, the diffusion processes can be treated independently. However, when the
dopant concentration exceeds ni (e.g. at 1000oC, ni = 5 x 1018 atoms/cm3), the
process becomes extrinsic, and the diffusion coefficients become concentration
dependent, as shown in Figure 8.6.

Figure 8.6: Donor impurity diffusion coefficient versus electron concentration

showing regions of intrinsic and extrinsic diffusion.

When a host atom acquires sufficient energy and leaves its lattice site, a vacancy
is created. Depending on the charges associated with the vacancy, we can have:
(1) a neutral vacancy, Vo,
(2) an acceptor vacancy, V-,
(3) a doubly-charged acceptor vacancy, V2-,
(4) a donor vacancy, V+,
(5) and others.

The vacancy density of a given charge state (i.e., the number of vacancies per
unit volume) has a temperature dependence similar to that of the carrier density:
 19
Chapter 8
E F  Ei

Cv  Ci e kT
(Equation 8.11)

where Cv is the vacancy density, Ci is the intrinsic vacancy density, EF is the

Fermi level, and Ei is the intrinsic Fermi level.

If the dopant diffusion is dominated by the vacancy mechanism, the diffusion

coefficient is expected to be proportional to the vacancy density. At low doping
concentrations (n < ni), the Fermi level coincides with the intrinsic Fermi level
(i.e., EF = Ei). The vacancy density is equal to Ci and independent of the dopant
concentration. The diffusion coefficient, which is proportional to Ci, will also be
independent of doping concentration. At high doping concentrations (n > ni), the
Fermi level will move toward the conduction band edge for donor-type
EF  Ei
vacancies, and the term e becomes larger than unity. This causes Cv to
kT

increase, which in turn gives rise to enhanced diffusion, as exhibited in Figure

8.6.

When the diffusion coefficient varies with the dopant concentration, we should
use Equation 8.2 in lieu of Equation 8.3. We shall also consider the situation
where the diffusion coefficient can be written as:

 C
D  Ds   (Equation 8.12)
 Cs 

where Ds is the diffusion coefficient at the surface, Cs is the surface

concentration, and  is a positive integer. Therefore,


C    C  C 
 D  (Equation 8.13)
tt x  s  Cs  x 
 

Equation 8.13 can be solved numerically, and the solutions for constant-surface-
concentration diffusion are shown in Figure 8.7, along with the result for a
constant D (i.e.,  = 0). For concentration-dependent diffusion, the diffusion
profiles are much steeper at low concentrations (C << CS). Thus, highly abrupt
junctions can be formed when diffusion is made into a background of an opposite
impurity type. In fact, the abruptness of the doping profile results in a junction
depth is virtually independent of the background concentration.
 20
Chapter 8

Figure 8.7: Normalized diffusion profiles for extrinsic diffusion where the
diffusion coefficient becomes concentration dependent.
 21
Chapter 8

8.7 Diffusion in Silicon

The measured diffusion coefficients of arsenic and phosphorus as a function of

dopant concentration are displayed in Figure 8.8. The diffusion of arsenic in
silicon is associated with the acceptor-type vacancy, V-, and the diffusion
coefficient for n > ni can be written as:

E'
n  kTa
D  D ( )e
'
o (Equation 8.14)
ni

where Do' = 45.8 cm2/s, Ea' = 4.05 eV, and n is the carrier concentration (or
dopant concentration, assuming 100% activation). The junction depth is
essentially independent of the indigenous p-type concentration, and is given by:

1/ 2

 '  kTa
E'  C 
x j  1.6 Dst  1.6 Doe  s t 
   ni 

(Equation 8.15)
  

Figure 8.8: Extrinsic diffusivities of arsenic and phosphorus in silicon as a

function of dopant concentration.
 22
Chapter 8

The measured diffusion profile for arsenic is shown as curve (a) in Figure 8.9
xj
where Yj is equal to and xj is given by Equation 8.15. Owing to its abrupt
4 Ds t
doping profile, arsenic is used extensively to form shallow junctions such as the
source and drain regions in n-channel MOSFETs.

Figure 8.9: Normalized diffusion profiles for arsenic and boron in silicon. The
erfc distribution is shown for comparison.

The diffusion of boron in silicon is associated with donor-type vacancy, V+, and
the diffusion coefficient varies approximately linearly with dopant concentration.
The expression for D has the same form as Equation 8.14, except that Do' is 1.52
cm2/s, Ea' is 3.46 eV and n is replaced by p. The measured boron profile is
exhibited as curve (b) in Figure 8.9 and is less abrupt than the arsenic profile.
Note that both the boron and arsenic profiles are steeper than the erfc case shown
as curve (c) in Figure 8.9.

The diffusion of phosphorus in silicon is associated with the doubly-charged

acceptor vacancy, V2-, and the diffusion coefficient at high concentrations varies
as C2 (Figure 8.8). Figure 8.10 depicts the phosphorus diffusion profiles into
 23
Chapter 8

silicon for 1 hour at 1000oC. When the surface concentration is low (i.e.,
intrinsic diffusion region), the diffusion profile is given by the erfc [curve (a)].
As the concentration increases, the profile begins to deviate from the simple
expression [curves (b) and (c)]. At a very high concentration [curve (d)], the
profile near the surface is indeed similar to that shown as curve (b) in Figure 8.7.
However, at concentration ne, a kink occurs followed by a rapid diffusion
(broader in-depth distribution) in the tail region. The concentration ne
corresponds to a Fermi level 0.11 eV below the conduction band. At this energy
level, the coupled impurity-vacancy pair (P+V2-) dissociates to P+, V-, and an
electron. A large number of singly-charged acceptor vacancies V- are generated
to enhance diffusion in the tail region of the profile. The diffusivity in the tail
region is over 10-12 cm2/s, which is about 2 orders of magnitude larger than the
intrinsic diffusivity at 1000oC. Therefore, phosphorus is commonly used to form
deep junctions such as n-tubs in a CMOS device.

Figure 8.10: Phosphorus diffusion profiles for various surface concentrations

after diffusion into silicon for 1 hour at 1000oC.
 24
Chapter 8

8.8 Emitter Push Effect

In silicon n-p-n bipolar transistors employing a phosphorus-diffused emitter and

a boron-diffused base, the base region under the emitter region (inner base) is
deeper by up to 0.6 µm than that outside the emitter region (outer base). This
phenomenon is called the emitter push effect, as illustrated in Figure 8.11. The
dissociation of phosphorus vacancy (P+V2-) pairs at the kink region provides a
mechanism for the enhanced diffusion of phosphorus in the tail region. The
diffusivity of boron under the emitter region (inner base) is also enhanced by the
dissociation of P+V2- pairs.

Figure 8.11: Calculated and measured boron and phosphorus n-p-n transistor
profile showing the emitter push effect. Emitter diffusion is at 1000oC for
1 hour followed by a 900oC, 45-minute steam oxidation.
 25
Chapter 8

8.9 Measurement Techniques

The results of a diffusion process can be evaluated by three parameters:

(1) Junction depth.
(2) Sheet resistance.
(3) Dopant profile.

8.9.1 Staining

Junction depths are commonly measured on an angle-lapped (1o to 5o) sample

chemically stained by a mixture of 100 c.c. HF (49%) and a few drops of HNO3.
If the sample is subjected to strong illumination for one to two minutes, the p-
type region will be stained darker than the n-type region, as a result of a
reflectivity difference of the two etched surfaces. The location of the stained
junction depends on the p-type concentration level and sometimes on the
concentration gradient. In general, the stain boundary corresponds to a
concentration level in the range of mid-1017 atoms/cm3.

Alternatively, junction depths can be delineated by cutting a groove into the

semiconductor and etching the surface (Figure 8.12). If Ro is the radius of the
tool used to form the groove, the junction depth, xj, is given by:

xj = (a2 - b2) / 2Ro (Equation 8.17)

Figure 8.12: Junction depth measurement by grooving and staining.

 26
Chapter 8

8.9.2 Four-Point Probe

The sheet resistivity of a diffused layer can be measured by a four point probe:

V
Rs 
I
C. F. (Equation 8.18)

where Rs is the sheet resistance, V is the measured voltage across the voltage
probes, I is the constant dc current passing through the current probes, and C. F.
is the correction factor to account for differences in sample size, geometry, and
probe positions.

8.9.3 Spreading Resistance Profiling

In spreading resistance profiling (SRP), two probes are used. The total spreading
resistance, Rsr, is given by:

Rsr = ρ / 2a (Equation 8.20)

where ρ is the average resistivity near the probe points and a is the probe radius.
If the two probes are stepped simultaneously in discrete intervals along a beveled
edge as illustrated in Figure 8.13, a high depth resolution profile of the electrical
carrier concentration can be acquired.

8.9.4 Secondary Ion Mass Spectrometry

Secondary ion mass spectrometry (SIMS) is very useful in elucidating impurity

diffusion profiles with excellent depth resolution (1 nm – 20 nm) and high
sensitivity (“parts per million” to “parts per billion” detection limits). An oxygen
or cesium ion beam of energy 1 to 20 keV is used to sputter a sample and the
sputtered ions are separated according to their mass-to-charge ratios (m/e) by a
mass spectrometer typically composed of a magnetic sector or quadrupole. SIMS
provides elemental (atomic) information whereas electrical techniques such as
spreading resistance profiling reveal electrical carrier concentration. If a dopant
is 100% activated (ionized), the SIMS and SRP results should theoretically agree.
 27
Chapter 8

Figure 8.13: The spreading resistance profile of an npn transistor structure: xeb =
emitter – base junction depth = 1.7 m; xcb = collector – base junction
depth = 3.2 m.
 28
Chapter 8

8.10 Oxide Masking

The diffusivities of common dopants are considerably smaller in silicon dioxide

than in silicon. Therefore, silicon dioxide is an effective mask against impurities.
The diffusion process in SiO2 occurs in two steps. During the first step, the
dopant reacts with silicon dioxide to form a glass that grows until the entire
silicon dioxide film is consumed (forming PSG, for instance). After the glass
forms, the dopant diffuses into the silicon substrate. Hence, during the first step,
silicon dioxide is completely effective in masking the silicon substrate against
dopants in the gas phase. The thickness of silicon dioxide required for an
effective masking is determined by the rate of formation of the glass, which in
turn is governed by the diffusion of the impurity into the silicon dioxide. Figure
8.14 shows the minimum thickness, d, of dry-oxygen-grown silicon dioxide
required to mask against phosphorus and boron as a function of temperature and
time. For a given temperature, d varies as t , as the diffusion length is given by
Dt .

Figure 8.14: Minimum thickness of dry-oxygen-grown SiO2 required to mask

against phosphorus and boron as a function of diffusion time with the
diffusion temperature as a parameter.
 29
Chapter 8

8.11 Lateral Diffusion

The one-dimensional diffusion equation discussed previously cannot adequately

describe the process at the edge of the mask window, at which the impurities will
diffuse both downward and laterally. In this case, a two-dimensional diffusion
equation must be used and solved numerically. Figure 8.15 depicts the contours
for constant-surface-concentration diffusion assuming that the diffusivity is
independent of concentration. As shown, the vertical penetration is about 2.8
µm, whereas the lateral one is about 2.3 µm. The lateral penetration is thus about
80% of the vertical penetration. In the case of constant-total-dopant diffusion,
this ratio is about 70%. For concentration-dependent diffusivities, this ratio is
reduced slightly to about 65% to 70%.

Figure 8.15: Diffusion contours at the edge of an oxide window, where rj is the
radius of curvature.
 30
Chapter 8

8.12 Fast Diffusants

Some elements are fast diffusants in silicon. They include the groups I and VII
elements, and some heavy metals, such as Au, Cu, Pt, and so on. They are
therefore undesirable contaminants in VLSI and are usually gettered away from
the active device regions by internal gettering techniques, that is, using SiOx
clusters in the bulk of the wafer to trap impurities.
 31
Chapter 8

8.13 Diffusion in Polysilicon

Polysilicon films are typically used in VLSI as a gate or as an intermediate

conductor in two-level systems. Since the gate electrode is over a thin oxide (15
nm to 150 nm thick), it is imperative that dopant atoms in the polysilicon do not
diffuse through the gate oxide. Polysilicon films are usually deposited at a low
temperature without doping elements. After the gate region is defined, the
polysilicon film is doped by diffusion (from a doped-oxide source or gas source)
or by ion implantation.

Impurity diffusion in polysilicon film can be explained qualitatively by a grain-

boundary model. A polysilicon film is composed of single crystallites of varying
sizes that are separated by grain boundaries. The diffusivity of impurity atoms
that migrate along grain boundaries can be up to 100 times larger than that in a
single crystal lattice. In addition, experimental results indicate that impurity
atoms inside each crystallite have diffusivities either comparable to or a factor of
10 larger than those found in the single crystal. The diffusivity in a polysilicon
film therefore depends strongly upon the structure (grain size, etc.) and texture.
These are in turn functions of the film deposition temperature, rate of deposition,
thickness, and composition of the substrate. Hence, it is difficult to predict
diffusion profiles in polysilicon. Diffusivities are typically estimated from
junction depths and surface concentrations are determined experimentally.