Chapter 2 1

CHAPTER 2: Crystal Growth and Wafer Preparation

Silicon is the most important semiconductor for the microelectronics industry.

When compared to germanium, silicon excels for the following reasons:

(1) Si has a larger bandgap (1.1 eV for Si versus 0.66 eV for Ge).
(2) Si devices can operate at a higher temperature (150oC vs 100oC).
(3) Intrinsic resistivity is higher (2.3 x 105 Ω-cm vs 47 Ω-cm).
(4) SiO2 is more stable than GeO2 which is also water soluble.
(5) Si is less costly.

2.1 Crystal Structure

The processing characteristics and some material properties of silicon wafers

depend on its orientation. The <111> planes have the highest density of atoms
on the surface, so crystals grow most easily on these planes and oxidation occurs
at a higher pace when compared to other crystal planes. Traditionally, bipolar
devices are fabricated in <111> oriented crystals whereas <100> materials are
preferred for MOS devices. Real crystals are imperfect and contain point defects,
line defects or dislocations, area or plane defects, and volume defects.

Point defects manifest themselves in several forms, as shown in Figure 2.1. Any
non-silicon atoms incorporated into the lattice at either a substitutional or
interstitial site are considered point defects. Point defects are important in the
kinetics of diffusion and oxidation. Moreover, to be electrically active, dopants
must occupy substitutional sites in order to introduce an energy level in the
bandgap.

Dislocations are line defects. Figure 2.2 is a schematic representation of a line

dislocation in a cubic lattice. Dislocations in a lattice are dynamic defects. That
is, they can diffuse under applied stress, dissociate into two or more dislocations,
or combine with other dislocations. Dislocations in devices are generally
undesirable, because they act as sinks for metallic impurities and alter diffusion
profiles.
Chapter 2 2

Figure 2.1: Location and types of point defects in a simple lattice.

Figure 2.2: An edge dislocation in a cubic lattice created by an extra plane of

atoms. The line of the dislocation is perpendicular to the page.
Chapter 2 3

Two typical area or planar defects are twins and grain boundaries. Twinning
represents a change in the crystal orientation across a twin plane, such that a
mirror image exists across that plane. Grain boundaries are more disordered than
twins and separate grains of single crystals in polycrystalline silicon. Planar
defects appear during crystal growth, and crystals having such defects are not
considered usable for IC manufacture and are discarded.

Precipitates of impurity or dopant atoms constitute the fourth class of defects.

The solubility of dopants varies with temperature, as exhibited in Figure 2.3.
Thus, if an impurity is introduced at the maximum concentration allowed by its
solubility, a supersaturated condition will exist upon cooling. The crystal
achieves an equilibrium state by precipitating the impurity atoms in excess of the
solubility level as a second phase. Precipitates are generally undesirable as they
act as sites for dislocation generation. Dislocations result from the volume
mismatch between the precipitate and the lattice, inducing a strain that is relieved
by the formation of dislocations.

Figure 2.3: Solid solubility of impurity elements in silicon.

Chapter 2 4

2.2 Electronic-Grade Silicon

Electronic-grade silicon (EGS), a polycrystalline material of high purity, is the

starting material for the preparation of single crystal silicon. EGS is made from
metallurgical-grade silicon (MGS) which in turn is made from quartzite, which is
a relatively pure form of sand (Table 2.1). MGS is purified by the following
reaction:

Si (solid) + 3HCl (gas)  SiHCl3 (gas) + H2 (gas) + heat

The boiling point of trichlorosilane (SiHCl3) is 32oC and can be readily purified
using fractional distillation. EGS is formed by reacting trichlorosilane with
hydrogen:

2SiHCl3 (gas) + 2H2 (gas)  2Si (solid) + 6HCl (gas)

Chapter 2 5

Table 2.1: Comparison of typical impurity contents in various materials (values

in ppm except as noted).

Impurity Quartzite Carbon MGS  EGS  Crucible quartz

Al 620 5500 1570 ... ...

B 8 40 44 <1 ppb ...
Cu <5 14 ... 0.4 0.23
Au ... ... ... 0.07 ppb ...
Fe 75 1700 2070 4 5.9
P 10 140 28 <2 ppb ...
Ca ... ... ... ... ...
Cr ... ... 137 1 0.02
Co ... ... ... 0.2 0.01
Mn ... ... 70 0.7 ...
Sb ... ... ... 0.001 0.003
Ni ... ... 4 6 0.9
As ... ... ... 0.01 0.005
Ti ... ... 163 ... ...
La ... ... ... 1 ppb ...
V ... ... 100 ... ...
Mo ... ... ... 1.0 5.1
C ... ... 80 0.6 ...
W ... ... ... 0.02 0.048
O ... ... ... ... ...
Na ... ... ... 0.2 3.7

* Metallurgical-grade silicon
 Electronic-grade silicon
Chapter 2 6

2.3 Czochralski Crystal Growth

Crystal growth typically involves a phase change from a solid, liquid, or gas
phase to a crystalline solid phase. The Czochralski (CZ) process, which accounts
for 80% to 90% of worldwide silicon consumption, consists of dipping a small
single-crystal seed into molten silicon and slowly withdrawing the seed while
rotating it simultaneously. Figure 2.4 is a schematic of a Czochralski crystal
growing apparatus. The crucible is usually made of quartz or graphite with a
fused silica lining. After the seed is dipped into the EGS melt, the crystal is
pulled at a rate that minimizes defects and yields a constant ingot diameter.

Figure 2.4: The Czochralski crystal growth apparatus.

Impurities, both intentional and unintentional, are introduced into the silicon
ingot. Intentional dopants are mixed into the melt during crystal growth, while
unintentional impurities originate from the crucible, ambient, etc. All common
impurities have different solubilities in the solid and in the melt. An equilibrium
segregation coefficient ko can be defined to be the ratio of the equilibrium
Chapter 2 7

concentration of the impurity in the solid to that in the liquid at the interface, i.e.
ko = Cs/Cl. Table 2.2 lists the equilibrium segregation coefficients for some
common impurities and dopants. Note that all are below unity, implying that the
impurities preferentially segregate to the melt and the melt becomes
progressively enriched with these impurities as the crystal is being pulled.

Table 2.2: Segregation coefficients for common impurities in silicon.

Impurity Al As B C Cu Fe O P Sb
-6 -6
ko 0.002 0.3 0.8 0.07 4x10 8x10 0.25 0.35 0.023

The distribution of an impurity in the grown crystal can be described

mathematically by the normal freezing relation:

Cs  k o Co (1  X ) ko 1 (Equation 2.1)

where X is the fraction of the melt solidified, Co is the initial melt concentration,
Cs is the solid concentration, and ko is the segregation coefficient. Figure 2.5
illustrates the segregation behavior for three segregation coefficients.

CZ-Si crystals are grown from a silicon melt contained in a fused silica (SiO2)
crucible. Fused silica reacts with hot silicon and releases oxygen into the melt.
Hence, CZ-Si has an indigenous oxygen concentration of approximately 1018
atoms/cm3. Although the segregation coefficient of oxygen is <1, the axial
distribution of oxygen is governed by the amount of oxygen in the melt. Less
dissolution of the crucible material occurs as the melt volume diminishes, and
less oxygen is available for incorporation. Figure 2.6 shows the variation in the
oxygen content along a typical silicon ingot. Oxygen is an important element in
CZ-Si on account of the following reasons:
(1) Oxygen forms a thermal donor in silicon.
(2) Oxygen increases the mechanical strength of silicon.
(3) Oxygen precipitates provide gettering sites for unintentional impurities.

Thermal donors are formed by the polymerization of Si and O into complexes

such as SiO4 in interstitial sites at 400oC to 500oC. Careful quenching of the
crystal annihilates these donors.
Chapter 2 8

Under certain annealing cycles, oxygen atoms in the bulk of the crystal can be
precipitated as SiOx clusters that act as trapping sites to impurities. This process,
which is illustrated in Figure 2.7, is called internal gettering and is one of the
most effective means to remove unintentional impurities from the near surface
region where devices are fabricated.

Figure 2.5: Impurity concentration profiles for different ko with Co = 1.

Chapter 2 9

Figure 2.6: Axial distribution of oxygen in a typical CZ-Si ingot.

Figure 2.7: Schematic of a denuded zone in a wafer cross section illustrating the
gettering sites (a and b are zones denuded of defects whereas c represents
the region of intrinsic gettering).
Chapter 2 10

Derivation of the normal freezing relation (Equation 2.1):

Weight = M

Ingot Weight = dM
Dopant conc. = Cs

Melt

S = dopant remaining in melt

Consider a crystal being grown from a melt having an initial weight Mo with an
initial doping concentration Co in the melt (i.e., the weight of the dopant per 1
gram melt). At a given point of growth when a crystal of weight M has been
grown, the amount of the dopant remaining in the melt (by weight) is S. For an
incremental amount of the crystal with weight dM, the corresponding reduction
of the dopant (-dS) from the melt is Cs dM, where Cs is the doping concentration
in the crystal (by weight):

 dS  C s dM (1)

Now, the remaining weight of the melt is Mo - M, and the doping concentration
in the liquid (by weight), Cl, is given by

S
Cl  (2)
Mo  M

Combining (1) and (2) and substituting Cs Cl  ko :

dS  dM 
 ko   (3)
S  Mo  M 

Given the initial weight of the dopant, Co M o , we can integrate (3):

Chapter 2 11

S dS M  dM
Co M o S
k o 
o M  M
o
(4)

Solving (4) and combing with (2) give

k o 1
 M 
C s  k o C o 1   (5)
 Mo 

Details on how to solve Eq. (4)

S dS  S 
(1) Left hand side of Eq. (4): Co M o S
 ln  
 Co M o 

(2) Right hand side of Eq. (4): Let m  M o  M , dm  dM ;

Mo M
M
 dM dm M  M 
ko   ko   k o ln  o 
0
Mo  M Mo
m  Mo 

 S  M  M 
Eq. (4) becomes: ln    k o ln  M
o

 o o
C M  o 

k0
S M  M 
and so  o
Co M o  M o 

From Eq. (2), S  Cl ( M o  M )

k0
 
Therefore, Cl ( M o  M )   M o  M 
Co M o  Mo 

Cs
(M o  M ) k0 k0
ko  Mo  M  Cs  M   M 
  or 1  M   1  M 
Co M o  Mo  k o Co  o  0

ko 1
 M 
Finally, C s  k o Co 1   (5)
 Mo 
Chapter 2 12

Example 2.1

Using Equation 2.1 and the segregation coefficient for oxygen from Table 2.2,
predict the concentration of oxygen in the crystal at a fraction solidified of 0.4 in
Figure 2.6.

Solution

First, find the concentration of oxygen in the melt at the top of the crystal (x =
0.05) using Equation 2.1:

1.3 x 1018 = 0.25 x Co(1-0.05)0.25-1

Co = 5.0 x 1018 atoms/cm3

Second, calculate the expected value at a fraction solidified of 0.4:

Cs = 0.25 x 5.0 x 1018 (1-0.4)0.25-1

Cs = 1.83 x 1018 atoms/cm3

Since this is higher than the observed value of 1.0 x 1018 atoms/cm3, it indicates a
loss of oxygen from the melt during growth and shows that Equation 2.1 is not
strictly applicable for this case.
Chapter 2 13

2.4 Float-Zone Process

The float-zone process has some advantages over the Czochralski process for the
growth of certain types of silicon crystals. The molten silicon in the float-zone
apparatus (Figure 2.8) is not contained in a crucible, and is thus not subject to the
oxygen contamination present in CZ-Si crystals. The float-zone process is also
necessary to obtain crystals with a high resistivity (>> 25 Ω-cm).

Figure 2.8: Float-zone crystal growth apparatus.

Chapter 2 14

Example 2.2

Show that in the float-zone process, the impurity concentration at a distance x,

Cs(x), is given by the following relationship:

ko x
C s ( x)  C o  C o (1  k ) exp[  ]
l

where Co is the original impurity concentration in the ingot, ko is the impurity

segregation coefficient, and l is the length of the molten zone.

Solution

Assume that the zone in the above figure moves a distance x to the right. Write
the balance that the quantity of impurity in the zone after the shift of x is equal
to that before the shift, minus the quantity solidified at the left, and plus that
melted into the zone at the right. Obtain the standard derivative form and let x
approach zero. Solve the differential equation for CL(x) using the boundary
condition that Cl (x = 0) = Co.

Therefore,

lAC L ( x  x)  lAC L ( x)  C s ( x) Ax  C o Ax

Since C s ( x)  k o C L ( x) ,

C L ( x  x)  C L ( x) C o k o
  C L ( x)
x l l

Let x  0,
Chapter 2 15

dC L ( x) C 0 k o
  C L ( x)
dx l l

The solution of this differential equation is:

Co
C L ( x)   Be ko x / l
ko

At x = 0, CL(0) = Co, and B = Co (1-1/ko). Thus,

ko x
Co 1 
CL ( x)   Co (1  )e l
ko ko

ko x
Cs ( x)  C0  Co (1  ko ) exp[  ]
l
Chapter 2 16

2.5 Characterization

Routine evaluation of ingots or boules involves measuring the resistivity,

evaluating their crystal perfection, and examining their mechanical properties,
such as size and mass. Other less routine tests include the measurement of
oxygen, carbon, and heavy metal contents.

Resistivity measurements are made on the flat ends of the crystal by the four-
point probe technique. As depicted in Figure 2.9, a current I passes through the
outer probes and the voltage V is measured between the inner probes. The
measured resistance (V/I) is converted to resistivity (Ω-cm) using the relationship

ρ = (V/I)2πS (Equation 2.2)

where S is the probe spacing in centimeters. The calculated resistivity can be

correlated with dopant concentration using a chart such as the one displayed in
Figure 2.10.

Gross crystalline imperfections are detected visually and defective crystals are
cut from the boule. More subtle defects such as dislocations can be disclosed by
preferential chemical etching. Chemical information can be acquired employing
wet analytical techniques or more sophisticated solid-state and surface analytical
methods.

Figure 2.9: Four-point probe measurement on crystal end.

Chapter 2 17

Figure 2.10: Conversion between resistivity and dopant density in silicon.

Chapter 2 18

Example 2.3

A boron-doped crystal is measured at its seed end with a four-point probe of

spacing 1 mm. The (V/I) reading is 10 ohms. What is the seed end doping and
the expected reading at 0.95 fraction solidified?

Solution

First use Equation 2.2 to find the seed end resistivity.

 = 10 x 2 x  x 0.1

 = 6.3 -cm

Figure 2.10 shows that this equates to a doping of 2 x 1015 atoms/cm3.

Next use Equation 2.1 and Table 2.2 to find the melt concentration:

2 x 1015 = 0.8 Co (1-0)0.8-1

Co = 2.5 x 1015 atoms/cm3

At 0.95 fraction solidified, the solid concentration is:

Cs = 0.8 x 2.5 x 1015 (1-0.95)0.8-1

Cs = 3.6 x 1015 atoms/cm3

or about 5 -cm
Chapter 2 19

Example 2.4

Resistivity is usually inversely proportional to dopant concentration in a silicon

wafer. Suppose that in a Czochralski process, a chunk of undoped polysilicon
is mixed with a small amount of heavily doped silicon with resistivity of 0.01
Ω-cm and the crystal is subsequently pulled. For one run, the resistivity of the
grown crystal at the midsection is 1 Ω-cm. The amount of heavily doped
silicon is to be changed in the next run to obtain a resistivity of 0.5 Ω-cm at the
midsection of the new ingot. Determine the increase in the amount of heavily
doped silicon that is needed. Assume that all operating conditions remain the
same and the amount of the added heavily-doped polysilicon is small compared
to the undoped polysilicon charge. Discuss whether your answer is correct only
at the midpoint of the ingot or is universally true for all other axial positions on
the ingot.

Solution

Based on the normal freezing relationship: Cs  k o Co (1  X ) k 1

o

At the midsection of the first ingot: C0.5,1  ko C1 (0.5) k 1

o

At the midsection of the second ingot: C0.5,2  ko C2 (0.5) k 1

o

C0.5,1 C1 1
Therefore,  
C0.5, 2 C2 2

The concentration in the midsection depends only on the initial dopant

concentration in the melt. If the volume change is neglected, one can simply
double the amount of the heavily-doped polysilicon to achieve a two-fold
decrease in the resistivity everywhere.
Chapter 2 20

2.6 Wafer Preparation

Silicon, albeit brittle, is a hard material. The most suitable material for shaping
and cutting silicon is industrial-grade diamond. Conversion of silicon ingots into
polished wafers requires several machining, chemical, and polishing operations.

After a grinding operation to fix the diameter of the material (Figure 2.11), one
or more flats are grounded along the length of the ingot. The largest flat, called
the "major" or "primary" flat, is usually relative to a specific crystal orientation.
The flat is located by x-ray diffraction techniques. The primary flat serves as a
mechanical locator in automated processing equipment to position the wafer, and
also serves to orient the IC device relative to the crystal. Other smaller flats are
called "secondary" flats that serve to identify the orientation and conductivity
type of the wafer. Secondary flats thus provide a means to quickly sort and
identify wafers should mixing occur. The flat locations for the four common
types of silicon wafers are exhibited in Figure 2.12.

Figure 2.11: Schematic of the grinding process.

Chapter 2 21

Figure 2.12: Identifying flats on silicon wafers.

The drawback of these flats is that the usable area on the wafer, i.e. the area on
which microelectronic devices can be fabricated, is reduced. For some 200 mm
and 300 mm diameter wafers, only a small notch is cut from the wafer to enable
lithographic alignment but no dopant type or crystal orientation information is
conveyed.

Once these operations have been completed, the ingot is ready to be sliced by
diamond saw into wafers. Slicing determines four wafer parameters: surface
orientation (e.g., <111> or <100>); thickness (e.g., 0.5 – 0.7 mm, depending on
wafer diameter); taper, which is the wafer thickness variations from one end to
anther; and bow, which is the surface curvature of the wafer measured from the
center of the wafer to its edge. After slicing, the wafers undergo lapping
operation that is performed under pressure using a mixture of Al2O3 and
glycerine. Subsequent chemical etching removes any remaining damaged and
contaminated regions. Historically, mixtures of hydrofluoric, nitric, and acetic
acids have been employed, but alkaline etching, using potassium or sodium
hydroxide, is also common.

Polishing is the final step. Its purpose is to provide a smooth, specular surface on
which device features can be photoengraved. Figure 2.13 depicts the schematic
of a typical polishing machine and the process. Figure 2.14 displays finished
Chapter 2 22

silicon wafers of various dimensions. A finished wafer is subject to a myriad of

physical tolerances, and examples of the specifications are shown in Table 2.3.

Figure 2.13: Schematic of the polishing process.

Figure 2.14: Finished silicon wafers of various sizes.

Chapter 2 23

Table 2.3: Typical specifications for monocrystalline silicon wafers.

Parameter 125 mm 150 mm 200 mm 300 mm

Diameter (mm) 125+1 150+1 200+1 300+1
Thickness (mm) 0.6-0.65 0.65-0.7 0.715-0.735 0.755-0.775
Bow (μm) 70 60 30 <30
Total thickness variation 65 50 10 <10
(μm)
Surface orientation +1o +1o +1o +1o