Vous êtes sur la page 1sur 53

Natural Sciences Tripos Part II

MATERIALS SCIENCE
C7: Kinetics
Prof A. L. Greer
Lent Term 2010-11
II
CONTENTS
1. INTRODUCTION
2. ATOMIC AND MOLECULAR DIFFUSION
2.1 Ficks First and Second Laws
2.2 Interdiffusion
2.3 Thermodynamics of Diffusion
2.3.1 Chemical potential
2.3.2 Diffusion in an ideal solution
2.3.3 Diffusion in a non-ideal solution
2.4 Diffusion in Silicon
2.4.1 Self-diffusion
2.4.2 Dopant diffusion
2.4.3 Impurity diffusion
2.4.4 Diffusion at contacts
2.5 Diffusion in Ionic Solids
2.5.1 Frenkel and Schottky defects
2.5.2 Diffusion and doping in oxides
2.5.3 Diffusion in compound semiconductors
2.6 Electromigration
2.6.1 Effective charge
2.6.2 Electromigration damage
2.6.3 Critical length for electromigration damage
2.6.4 Improving the reliability of aluminium metallization
2.6.5 Use of copper for metallization
2.7 Atomic and Molecular Transport in Liquids
2.7.1 Temperature dependence of mobility
2.7.2 Free-volume model
3. INTERFACE MIGRATION
3.1 Effective Pressure on Interfaces
3.2 Grain Growth
3.2.1 Origins of grain structure
3.2.2 Normal grain growth
3.2.3 Abnormal grain growth
3.2.4 Grain growth in thin films
3.3 Kinetics of the Solid-Liquid Interface
3.3.1 Contributions to the total supercooling
3.3.2 Kinetic supercooling
3.3.3 Loss of local equilibrium: Solute trapping
4. DIFFUSIONAL CONTROL OF TRANSFORMATION RATES
4.1 Precipitation
4.2 Ostwald Ripening
4.3 Eutectic Growth
5. ORDERING TRANSFORMATIONS
5.1 First- and Second-Order Transformations
5.2 Anti-Phase Domains
6. OVERALL TRANSFORMATION KINETICS
LIST OF SYMBOLS

a constant ()
molecular diameter (m)
A fraction of sites ()
b constant (
1 1 1
m m S

O )
B constant (, or K)
mobility (
1 1
s Pa m

)
C concentration (
3
m

)

p
C spec. ht at const. press. (
1 1
mol K J

)
D (chemical) diffusivity (
1 2
s m

)

*
D tracer diffusivity (
1 2
s m

)
D
~
interdiffusivity (
1 2
s m

)

D average grain diameter (m)
e electronic charge (C)
E electric field (
1
m V

)

bond energy (J)
F volume fraction ()
force (N)
fluidity (
1 1
s Pa

)
g geometrical constant ()
G Gibbs free energy (
1
mol J

)
V
G Gibbs free energy/unit vol. (
3
m J

)
H enthalpy (
1
mol J

)
I nucleation rate (
1 3
s m

)
j electron current density (
2
m A

)
J atomic flux (
1 2
s m

)

k Boltzmann constant (
1
K J

)
vel.-dependent partition coefft ()
reaction rate constant (
1
s

)
0
k equilibrium partition coefficient ()
K equilibrium constant ()
grain-growth constant (
1 2
s m

)

transformation-rate const (
3
s

or
n
s )
1
K eutectic-growth constant (
1 1 2
s K m

)
2
K eutectic-growth constant ( m K )
L segment length (m)
M interfacial mobility (
1 1
s Pa m

)
n mole fraction of neg. carriers ()
number ()
N number ()
a
N Avogadro number (
1
mol

)
p mole fraction of positive carriers ()
probability ()
P pressure (Pa)
Q activation energy (eV/atom;
1
mol J

)
r radius (m)
r average grain radius (m)
R gas constant (
1 1
mol K J

)
s short-range order parameter ()
S entropy (
1 1
mol K J

)
t time (s)
T temperature (K)
liq
T liquidus temperature (K)
m
T melting temperature (K)
T A supercooling (K)
u gas kinetic velocity (
1
s m

)
U growth rate (
1
s m

)
v velocity (
1
s m

)
site volume (
3
m )
V volume (
3
m )
mol
V molar volume (
1 3
mol m

)
x linear co-ordinate (m)
X mole fraction ()
ss
X A supersaturation ()
z charge number ()
co-ordination number ()


o volume expansion coefft (
1 3
K m

)
geometrical constant ()
grain-bound./interfac. energy (
2
m J

)
u thermodynamic factor ()
q viscosity (Pa s)
long-range order parameter ()
jump distance (m)
eut
eutectic lamellar spacing (m)
chemical potential (
1
mol J

)
v jump/vibration/attempt freq. (
1
s

)

resistivity (O m)
fraction of sites occupied ()
o elec. conductivity (
1 1 1
m m S

O )
H
o hydrostatic stress (Pa)
O regular solution parameter (
1
mol J

)
atomic volume (
3
m )



BOOK LIST
(library codes are indicated; many, but not all, of these books are in the Tripos section)


Main reference

D. A. Porter and K. E. Easterling, Phase Transformations in Metals and Alloys Ln73a


Other useful references

J. W. Christian, The Theory of Transformations in Metals and Alloys Ln65

C. R. M. Grovenor, Microelectronic Materials LcB106

W. Kurz and D. J. Fisher, Fundamentals of Solidification Ng128

E. S. Machlin,
Introduction to Aspects of Thermodynamics and Kinetics relevant to Materials Science Pj83

J. W. Martin and R. D. Doherty, Stability of Microstructure in Metallic Systems Mb313

M. Ohring, Reliability and Failure of Electronic Materials and Devices Kw34

P. Shewmon, Diffusion in Solids Lr62a

W. A. Tiller, The Science of Crystallization NbA76

In addition, the following lecture handouts from earlier years may be useful
Part IA Materials Science, Course C on Microstructure
Part IB Materials Science, Course A on Phase Transformations
Part IB Materials Science, Course B on Environmental Stability of Materials


Relevant DoITPoMS TLPs include:
Diffusion
Solid Solutions
Electromigration
Grain Growth, in development, available at:
http://www.doitpoms.ac.uk/grainGrowth-dev/grainGrowth/index.php


1. INTRODUCTION
2. ATOMIC & MOLECULAR DIFFUSION

2.1 Ficks First and Second Laws

The motion of individual atoms or molecules has to be the starting point for any analysis of
materials kinetics.

Atomic diffusion in the solid state takes place through well defined jumps, so that:

2
6
1
v = D
where v is the jump frequency (
1
s

)
is the jump distance (m)
D is the diffusivity (
1 2
s m

).

Ficks First Law (for steady state diffusion in a uniform concentration gradient):

x
C
D J
c
c
=
where J is the atomic flux (
1 2
s m

)
C is the concentration of the diffusing species (
3
m

)
x is a linear co-ordinate (m).

Ficks Second Law (for non-steady state diffusion):

2
2
x
C
D
t
C
c
c
=
c
c

where t is the diffusion time. (This is the diffusion equation.)

Both laws are stated here in their simple forms for 1-D diffusion. The form given for the
second law assumes that D is independent of C.

In Part IB Course A, two solutions to the diffusion equation were discussed

For a source at fixed concentration
( ) ( ) |
.
|

\
|
=
Dt
x
C C C t x C
2
erf ,
0 s s

where ( )
s
, 0 C t x C = = and ( )
0
0 , C t x C = = .

For a source of fixed amount (thin-film solution)
( )
|
|
.
|

\
|
=
Dt
x
DT
B
t x C
4
exp

,
2

where ( ) 0 0 , = = t x C and ( ) B x t x C =
}

d ,
0
.

Consider mixing of species A (on the left) and B (on the right) in a diffusion couple (each
side being in effect semi-infinite). We presume that the couple retains its external dimensions
and does not develop internal porosity. Suppose the diffusivities are asymmetric,
B A
D D >> .
Relative to the lattice, the flux
A
J (left to right) is then greater in magnitude than
B
J

(right to
left). But relative to the laboratory, there can be no net flux, so that there must be a lattice
drift (in this case, from right to left). The drift can be detected by the movement of inert
markers (Kirkendall effect).

The analysis of this problem by Darken shows that the lattice drift velocity, v, is given by
( )
x
C
D D
C
v
c
c
=
A
B A
0
1

where
0
C is the concentration of atoms throughout the system (assumed constant).

Then, relative to the laboratory frame,

x
C
D J
c
c
=
A
A
~
and
x
C
D J
c
c
=
B
B
~

where D
~
is the interdiffusivity (
1 2
s m

), given by

B A A B
~
D X D X D + =
where
A
X and
B
X are the mole fractions of the two species.

The expressions for v and D
~
are the Darken relations, and they apply when:

there is efficient operation of sources and sinks for vacancies
diffusion is sufficiently slow (i.e. over great enough distance) that the stresses resulting from
any volume changes can be effectively relaxed.

When there are not efficient sinks for vacancies, porosity develops. When diffusion distances
are short, as in a deposited multilayer, significant stresses build up and we are in the Nernst-
Planck regime, in which:


B
A
A
B
1 1
~
1
D
X
D
X
D
+ =





In the Darken regime, the interdiffusion
is analogous to parallel conduction, and
D
~
is controlled by the faster component.
In the Nernst-Planck regime, the
interdiffusion is analogous to series
conduction, and D
~
is controlled by the
slower component.

The change of regime can be seen experimentally:




This shows the change of D
~
from the
Darken regime (on the right) to the
Nernst-Planck regime (on the left) as the
length scale of interdiffusion is reduced.
The system is amorphous Ni-Zr.





2.3.1 Chemical potential

Under the action of a force
A
F , atoms of A drift with a mean velocity v, given by:

A A
F B v =
where
A
B is the mobility of A.

The flux of A atoms is then given by:

A A A A A
F B C v C J = =

where
A
C is the concentration of A atoms (
3
m

).


The force is a gradient of some
potential, and could have many
different causes. One possible cause is
a composition gradient, in which case
the relevant potential is the chemical
potential,
A
:



The force on an A atom due to the composition gradient is then

x N
F
d
d 1
A
a
A

= .
So that

x N
B
C J
d
d
A
a
A
A A

= .
Comparing with Ficks First Law, we arrive at:

A
A
a
A
A
A
a
A
A
A
a
A
A A
ln d
d
ln d
d
d
d
X N
B
C N
B
C N
B
C D

= = = .


2.3.2 Diffusion in an ideal solution

To illustrate the effects of the thermodynamics of the solution in which the diffusion is
occurring, we take the regular solution (Part IA Course C), for which the free energy G is
given by:

mix mix B B A A
S T H G X G X G A A + + =
( )
B B A A B A B B A A
ln ln X X X X RT X X G X G X + + O + + = .
But since

B B A A
X X G + =

we can equate the coefficients of terms in
A
X and
B
X to arrive at
( )
A
2
A A A
lnX RT X G + 1 O + = (with the equivalent for
B
).
A special case of the regular solution is the ideal solution in which 0 = O , giving

A A A
lnX RT G + = .

In an ideal solution, diffusion is driven by the entropy of mixing alone, and the diffusivity is
given by:
kT B RT
N
B
X N
B
D
A
a
A
A
A
a
A
*
A
ln d
d
= = =

.
Here
*
A
D is the tracer diffusion coefficient, i.e. that in the absence of chemical effects, such
as for a radioactive tracer.


2.3.3 Diffusion in a non-ideal solution

|
.
|

\
| O
=
RT
X X
D D
B A
*
A A
2
1
u =
|
.
|

\
| A
=
*
A
mix
*
A
2
1 D
RT
H
D
where u is the thermodynamic factor ( 1 = u , for an ideal solution).

Similarly,
u =
*
B B
D D .
A
D and
B
D are the chemical
diffusion coefficients. The
Darken relation for
interdiffusivity becomes

( )u + =
~
*
B A
*
A B
D X D X D .

For systems with a negative heat
of mixing, D
~
is enhanced
relative to an ideal solution, the
highest measured value of u
being ~33, as shown here


In contrast, in a phase-separating system, the
heat of mixing is positive, and

|
.
|

\
| A
= u
RT
H
mix
2
1
can be negative. In this regime, the solid
solution is unstable against infinitesimal
composition fluctuations. This is the
process of spinodal decomposition, and it
shows uphill diffusion, i.e. negative D
~
:




Development of spinodal decomposition.


2.4.1 Self-diffusion

Self-diffusion in silicon relies on vacancies. We write the mole fraction of vacancies as:
] [V
Si V
Si
X .
As is familiar for metals, the population of vacancies increases with temperature:

|
.
|

\
| A
|
.
|

\
| A

kT
H
k
S
X
vac vac
V
exp exp
Si
.
But in a semiconductor, the vacancies lead to new states in the band gap:



There are three charged states with significant populations, and so self-diffusion in silicon has
contributions from four diffusing species (each with a distinct activation energy, etc.):

+
+ + + =
Si
2
Si Si Si
V V V V
Si
D D D D D .
The mole fraction of charged vacancies is affected by the state of doping. Let us consider for
example n-doped silicon:

+
+ h V V
Si Si
.
The equilibrium constant for this reaction is

] V [
] V [
Si
Si
p
K

= .
In a doped semiconductor:

2
i
n np =
where
i
n is the population of carrier electrons in the intrinsic semiconductor.

It follows that:
n

] V [
Si

and that

i undoped Si
doped Si
] V [
] V [
n
n
=

.

Overall, in doped silicon we have

|
.
|

\
|
+
|
|
.
|

\
|
+
|
|
.
|

\
|
+ =
+
n
n
D
n
n
D
n
n
D D D
i
V
2
i
V
i
V V
Si
Si
2
Si Si Si
.
In n-type Si, the self-diffusion is mediated mainly by

Si
V ; in p-type, by
+
Si
V .


2.4.2 Dopant diffusion

Arsenic as a solute in silicon acts as a dopant only if it is substitutional (not interstitial). On
substitutional sites, it is ionized to
+
Si
As (donating a negative carrier

e ), and interacts mainly


with

Si
V . We therefore expect that:
n D

] V [
Si As

so that

|
|
.
|

\
|
=
i
i
As As
n
n
D D
where
i
As
D is the diffusivity in the intrinsic state, i.e. at infinite dilution of dopant.

The measured dependence of
As
D on ] As [
Si
, provides evidence for the link with

Si
V . Since
As
D depends on ] As [
Si
, Ficks second law cannot be straightforwardly applied.

For dopants in general, there are many types of behaviour:

As n-type

Si
V
B p-type
+
Si
V
P n-type
Si
V
Sb n-type
Si
V


2.4.3 Impurity diffusion

Na, K, Cu interstitial, very fast, even at room temperature.

Au in Si very strange behaviour!

Gold applied to one surface of a silicon wafer
only.

But gold diffusion profiles are measured at
both surfaces!

Au in Si can be both:


Interstitial very fast diffusion ( eV/atom 39 . 0
diff
= Q ), but very low solubility.

Substitutional slow diffusion ( eV/atom 0 . 2
diff
= Q ), but some solubility.

But gold can dissolve substitutionally in
silicon, only with a supply of vacancies:

In principle, vacancies can be generated at
grain boundaries and dislocations, but in
semiconductor-grade silicon wafers, the
only obvious source is the surface.


This is the Frank-Turnbull mechanism. Similar effects are found for Fe, Ni, Cr. Fast
interstitial transport of these transition metals is very bad, because they form deep electronic
states and destroy the carrier mobility in silicon. Even picking up a silicon chip with stainless
steel tweezers can be disastrous.










2.4.4 Diffusion at contacts

On silicon integrated circuits (ICs), we must have electrical contacts and connections between
these (interconnects). We consider a contact of Al (the usual metallization) on Si.

The silicon is oxidized. A contact window is
etched through the oxide. The silicon in the
window still has a thin coating of native
oxide. Aluminium is deposited on top. This
is annealed at 400-500C for the Al to reduce
the native oxide and make contact with the
silicon. There is no reaction between Al and
Si (there are no aluminium silicides), but Si is
soluble in Al. The Si dissolves through weak
points in the native oxide and metal spikes
into the device (through the doped layers,
with disastrous consequences).



The old solution to this problem is to add ~1wt%Si to the Al; this is roughly the solubility
limit at 550C so therefore the Al resists penetration of Si up to 550C. But for T<550C,
the silicon comes out of solution in the metallization and forms an epitaxial p-type layer at the
contact.

There is a need for another solution especially now that copper is becoming preferred for
metallization. (Unlike the case for Al, for Cu no direct contact with Si is permissible.) The
solution is to use a diffusion barrier layer.




2.5.1 Frenkel and Schottky defects

These have vacancies and interstitials, but factors to be considered are:

constraints of sub-lattice: cations diffuse only on the cation sub-lattice, etc.
need to maintain charge neutrality: in stoichiometric compounds, this is achieved with two
main types of defect:

Frenkel defect cation moves to an interstitial site and creates a vacancy

Schottky defect vacancies on anion and cation sites are balanced to ensure neutrality.


2.5.2 Diffusion and doping in oxides

Oxides forming on metals are non-stoichiometric, for example:

ZnO normally has an excess of Zn, accommodated as interstitial ions, and arising from
( ) g O e 2 Zn ZnO
2
2
i
+ +
+
.



In this structure, the excess

e are more
mobile than the cations. ZnO is an n-type
semiconductor. In a higher pressure of
oxygen, ZnO is more stoichiometric, has
lower ionic conductivity, and has lower
electronic conductivity. For oxide film
growth, both ionic and electronic
conductivity are required.

There are other possibilities, for example:
NiO metal-deficit p-type

2
TiO anion-deficit n-type.

Diffusion in oxides is relevant for oxide film growth:

In most cases, including ZnO, the dominant
ionic carriers are cations. For oxide growth
to be possible, we need:
ionic conductivity
electronic conductivity (by

e or
+
h ).
The rate is limited by the slower ionic
conductivity, which can be controlled by
doping.


Oxides can be doped by adding ions of different valence. For example
+ 3
Cr can be
substituted into ZnO. To maintain charge neutrality, one must substitute
+ 3
Cr 2 for
+ 2
Zn 3 ;
this reduces the total number of cations, there are fewer interstitials, and the conductivity is
decreased. An opposite effect is found by doping with ions of lesser valence, e.g.
+
Li .


2.5.3 Diffusion in compound semiconductors

We have seen from silicon that defects such as vacancies can be charged and that
consequently their populations are affected by doping. We have seen from oxides that
stoichiometry is very important, deviation from stoichiometry generally giving faster
diffusion. In a compound semiconductor such as GaAs all these effects come together and the
overall behaviour is exceedingly complex. Relevant species include:

Ga
V
As
V
i
Ga
i
As
As
Ga
Ga
As
vacancies interstitials anti-site defects

each of these can have various effective charges. Their concentrations are affected by
stoichiometry and by doping.



2.6.1 Effective charge

In an electrical conductor, the electron current density is given by
E j o = (this gives the electron flux, antiparallel to conventional current)
where o is the conductivity (
1 1 1
m m S

O ), and E is the applied field (
1
m V

). For an
isolated charge (z = 1, 2, etc.) in a field E, the force on it would be:
zeE F =
where e is the magnitude of the electronic charge ( C 10 602 . 1
19
+ = ).

However, the force on metal ions in the conductor has two components:

(i) a direct force
d
F arises because of their charge, but screening effects modify this by a
factor a (<<1)
(ii) a wind force
w
F due to momentum transfer from the electrons (the counterpart of
electrical resistance)

azeE F =
d
eE z F
w w
=
where
w
z must be negative and is inversely proportional to the resistivity (

b
z

=
w
).

The total electromigration force on an ion of the conductor,
EM
F , is given by
eE z F F F
*
w d EM
= + =
where
*
z is the effective charge, given by

b
az z =
*

(where a and b are both positive).
In general,
*
z can be positive or negative. For low metals (e.g. Al and Cu, as used in ICs),
*
z is negative, i.e. the electron wind dominates and the ions migrate in the same direction as
the electrons. (For Al atoms in aluminium, 10
*
~ z .)
The flux of A atoms in the conductor is given by:

j
e z
kT
D
C F
kT
D
C F B C J
o
* *
A
A EM
*
A
A EM A A A
= = =

so that overall the electron current density j causes a slight biasing of the diffusional jumps of
the atoms.


2.6.2 Electromigration damage

We normally expect
EM
J to be negligible. With high currents the effect can be used to purify
metal rods (migration of impurities to one end or the other).


But electromigration became really
significant as the major cause of failure in
integrated circuits (ICs). Miniaturization
leads to interconnects of smaller cross-
section: typically 0.18 m wide and of
similar thickness, and always getting
smaller.



Also, ICs have greater and greater total
length of interconnect lines on them now
several km. To make all the required
interconnections, it is necessary to use up to 5
layers of metallization, connected by vias.

Currents are small ( mA 1 < ), but current
densities are very high (
2 10
m A 10

> ).
(Roughly 1000 the current density in a standard 13A flex.)

A small uniform atomic flux along an interconnect would cause few problems.
Electromigration damage is caused by non-uniformities in the atomic flux, giving:
regions of atomic depletion VOIDS (open circuit)
regions of atomic accumulation HILLOCKS and WHISKERS (short circuit):

Why do we have depletion and accumulation? Focusing on:
j
e z
kT
D
C J
o
* *
A
A A
=
the key parameter is
*
A
D . One reason why this might change is a change of material, for
example W vias between Al lines:

W has a very high
m
T , and consequently a
very low D compared to Al. The W via then
acts as a barrier to atomic diffusion


But we see electromigration damage even in lines of one metal. In such a case, non-
uniformities in
*
A
D can arise from variations in temperature (which can arise from heating in
the metallization itself or in the underlying semiconductor), or variations in grain size.

For Al-based metallization, a typical operating temperature
m
4 . 0 T T ~ , so that diffusion is
predominantly along grain boundaries and
*
A
D therefore depends on grain size.



When (as for Al-based metallization) atomic
diffusion is along grain boundaries, we can
expect many local divergences in atomic flux,
for example as here

But in practice we do not find a profusion of damage sites. This is because of:


2.6.3 Critical length for electromigration damage

A gradient in vacancy concentration would give a net migration of metal atoms, and is part of
the mechanism of creep. But even in the absence of such a gradient, with a uniform vacancy
concentration, atoms respond to a gradient in hydrostatic pressure
H
o . This is the
phenomenon of mechanodiffusion, the force on each atom being:

x C
F
d
d 1
H
0
M
o
= (where O
0
1
C
, the atomic volume).
A non-uniform electromigration flux of atoms gives regions of depletion and accumulation.
The associated local stresses lead to stress gradients that oppose the electromigration flux:


Between two divergences of opposite sign, the stress gradient is

L x
H H
d
d o o A
= .
This grows with time, and correspondingly
M
F increases, until
M
F is equal and opposite to
EM
F . The stress distribution then saturates as there is no further transport in the segment of
length L. Setting
EM
F and
M
F equal and opposite:

L C
eE z
H
0
*
1 o A
=
but o j E = (where o is the conductivity of the metal), so that:

H *
0
o
o
A =
e z C
jL .
The right-hand side of this equation
consists mainly of materials constants, of
which
H
o A has a critical value for
damage initiation (derived from the
absolute stress levels required for
damage). Thus we have damage only if
( )
crit
jL jL > , where:
( ) ( )
crit H *
0
crit
o
o
A =
e z C
jL .
Critical conditions of this kind are
revealed in drift-stripe experiments:



For a given current density j, there is no damage if ( ) j jL L
crit
< .
For a given segment length L, there is no damage if ( ) L jL j
crit
< .


2.6.4 Improving the reliability of Al metallization

This is achieved by alloying the Al with 1 to 4 wt% Cu. The copper segregates to grain
boundaries, can form Cu Al
2
precipitates, and blocks grain-boundary diffusion. The lifetime
under given testing conditions is 100 better than pure Al.


2.6.5 Use of copper for metallization on integrated circuits (ICs)

Al-Cu metallization is rapidly being replaced by Cu in high-performance devices.

Intrinsic gate delay and interconnect
delay as a function of feature size.
The interconnect delay is dominant in
modern devices. It is given by the
product of the line resistance R and the
capacitance of the inter-line dielectric
C.





Miniaturization leads to significant
capacitative coupling of metallization lines.
This graph of capacitance as a function of
feature size shows that of the various possible
components, it is the line-to-line capacitance
that is dominant in modern devices.


CPU cycle time (calculated, arbitrary
units) as a function of the number of
metallization wiring levels for different
combinations of conductors and
insulators. The cycle time is minimized
by choosing a conductor with low
resistivity and an insulator with low
dielectric constant k.

Comparison of resistivities suggests that copper would be a good choice:
Cu cm - O 1.58 =
Al cm - 61 . 2 O =
Al-Cu cm - 8 . 2 O ~ (all values quoted are for bulk, not thin film).

Comparison of dielectric constants suggests that fluoropolymer (FP) should be better than
2
SiO .

Copper is also preferable to Al-Cu because it should have an intrinsically higher resistance to
electromigration. Comparing melting points ( C 660
m
= T for Al; C 1085
m
= T for Cu),
copper should have a lower atomic diffusivity at the operating temperature. Comparing, for
example, bulk lattice diffusion at 50C: ( )
10
Cu Al
10 2 ~ D D ! But copper metallization is not

10
10 better In practice, interconnect lifetimes improve by
4 2
10 to 10 (compared to Al-
Cu).

With copper metallization it is possible to use
copper vias, but these are still associated with
electromigration damage:


The activation energy for electromigration failure of Cu is 75 to 100
1
mol kJ

(compared to
1
mol kJ 204

= Q for bulk lattice diffusion in copper).

For copper lines, the main transport path is the top surface (prepared by polishing). There is
little or no transport in the bulk (so that microstructure, i.e. grain size, is not important).
There is also little or no transport on the bottom or side surfaces of the copper, which is
deposited in contact with a Ta or TaN diffusion barrier.


The resistance of a copper line rises as a void
is formed, and brings the lines useful life to
an end. The lifetime of copper metallization
can be improved (by ~100) by applying a
thin (10-20 nm) top coating to stifle
diffusion, e.g. Co 3%W 6%P.


An additional factor favouring Cu is that it is
stronger than Al-Cu, so that ( )
crit H
o A is
higher. The ( )
crit
jL value for Cu is roughly
three times that for Al-Cu.


Comparison of Al and Cu for metallization
conductivity: Al is good, but Cu is better
diffusion and reaction: Al is inert, but Cu must be isolated from
2
SiO and from Si
bulk transport: much slower in Cu than in Al (because of higher
m
T )
surface transport: unknown for Al because of its tenacious oxide, but this is a significant
reliability issue for Cu
strength: Cu is better
corrosion resistance: Al is better.




2.7 Atomic Transport in Liquids

2.7.1 Temperature dependence of mobility

Atomic mixing in liquids is usually dominated by convective currents rather than true
diffusion. Diffusivities in liquids are therefore difficult to measure. Measurements are often
made in capillaries to stifle convection. There is also interest in microgravity experiments.

For liquids, it is easier to characterize the atomic transport by the viscosity q. The ideal self-
diffusivity (i.e. in the absence of convection) is related to the viscosity of the liquid q or to the
fluidity F ( q 1 = ), by the Stokes-Einstein relation:

a
kTF
a
kT
D
3 3
= =
q

where a is the molecular diameter (the ionic diameter for metallic liquids).

The viscosity is of direct interest for glass formation: the glass transition is at s Pa 10
12
= q .


The temperature dependence of liquid fluidity is
characteristically non-Arrhenius, always showing
a higher activation energy at lower temperature.
The behaviour is well fitted by the Vogel-Fulcher-
Tammann equation

|
|
.
|

\
|

0
exp
T T
B
F
which is valid for a wide variety of liquids:
simple molecular liquids (benzene, toluene)
metals, alloys
polymers
silicates (though the plot is then not so curved).

In understanding liquids, the volume is very important.
The thermal expansion of a liquid is much greater
than that of the corresponding crystal. The liquid
expansion is due not only to anharmonic vibration
(as in the crystal) but also to structural change.

The glass transition occurs when the viscosity
(increasing on cooling) becomes so high that it
prevents the structural change keeping pace with
the rate of cooling. With slower cooling, the
system stays in (metastable) equilibrium for
longer, and gives a denser glass.


On heating a liquid, its volume increases and it becomes more fluid and gas-like. But if we
heat a liquid to high temperature at constant volume (i.e. by applying the necessary pressure),
then the fluidity is roughly independent of temperature. Thus it is volume that has the key
effect on mobility. The analysis of atomic transport/diffusion in liquids is greatly
complicated by the variety of local environments. Characterizing the environments in terms
of their local volume v is the basis of


2.7.2 Free-volume model

The free volume
f
V is the excess over the ideal glassy state
( )
0 f
T T V =o
and is assumed to be freely distributed over the N atomic sites in the system

f
V v N
i
i
i
=


where N N
i
i
=


is a geometrical constant
, necessary to allow for
overlap

i
v are the site volumes.

The local diffusivity for one site ( ) v D is a function of the site volume. Atomic rearrangement
is possible only if the local volume is big enough. In the model, we take
For
*
v v < , ( ) 0 = v D
For
*
v v > , ( ) gau v D =
where g is a geometrical constant, taken to be 6 1
a is the molecular diameter
u is the gas kinetic velocity.

It is assumed that the number of ways of distributing the local volumes is maximized, giving a
distribution function:
( )
|
|
.
|

\
|
|
|
.
|

\
|
=
f f
exp
v
v
v
v p


where
N
V
v
f
f
= .
( )
|
|
.
|

\
|
|
|
.
|

\
|
=
f f
exp
v
v
v
v p




where
N
V
v
f
f
= .

The overall diffusivity must take account of the variety of local sites:
( ) ( ) ( )
|
|
.
|

\
|
= = =
} }

f
*
* 0
exp d d
v
v
gau v v p gau v v p v D D
v


where
*
v molecular volume.


The importance of the local volume is made
clear when we look at how the volume of the
diffusing species itself affects its mobility.

The graph shows a survey of data on tracer
diffusivities (at 573 K) of various species in
amorphous Ni-Zr alloys (of 50:50
composition).


Since ( )
0 f f
T T V v , the temperature dependence of the atomic mobility ( F D )
matches the form of the Vogel-Fulcher-Tammann equation:

|
|
.
|

\
|


0
exp
T T
B
F D .
Predictions:
any liquid will form a glass if crystallization can be suppressed
some free volume
f
V is always frozen-in at the glass transition.



The glass transition is kinetic, not
thermodynamic:

q as
0
T T




3.1 Effective Pressure on Interfaces

So far, we have considered kinetics only in terms of the mobility of atoms or molecules. But
there are many cases in which it is more useful to consider the motion of interfaces. We
consider the simplest possible case of a one-component system with two phases (e.g. ice and
water) with a free-energy difference between them. (For such a system, A AG .)
Phases 1 and 2 have free energies per unit
volume
1 V,
G and
2 V,
G . If
1 V, 2 V,
G G > ,
then there is a driving force for an
interface between the two phases to
migrate so that the higher-energy phase is
consumed. The free-energy difference
V 1 V, 2 V,
G G G A = per unit volume. If the interface moves a distance ox, the energy change
per unit area of interface is x G
V
A . The interface has an effective pressure P acting on it, and
the work done by this pressure in moving unit area of interface a distance dx is Pdx. Clearly
the energy change and work done must be the same, so that:

V
G P A = .
Possible origins of
V
G A include:
a phase difference (e.g. crystal/liquid, or different crystalline polymorphs)
different dislocation densities in the same phase (relevant for recrystallization)
different strain energies in the same phase (relevant for abnormal grain growth).

A special case is that in which there is also a pressure if the phase is identical on both sides of
the boundary, but the boundary is curved. The interface has two principal radii of curvature
1
r and
2
r . The increase in pressure on the concave side is given by:

|
|
.
|

\
|
+ = A
2 1
1 1
r r
P
where is the interfacial energy (
2
m J

). This is equivalent to an increase in the free
energy per unit volume on the concave side. We can check that this approach is consistent
with analyses met in earlier courses by considering the case of a critical nucleus:
___________________________________________________________________________
A spherical embryo, radius r, of a new phase
forms within an original phase. The work of
formation W varies with r and passes through
a maximum. The maximum W represents the
critical nucleus that is in unstable
equilibrium. The curvature of the spherical
surface is:


r r r
2 1 1
= + .
This generates a pressure that, in equilibrium, must balance the free-energy difference per unit
volume. Thus:

V
2
G
r
A = ,
So that the critical nucleus radius
*
r must be given by:

V
*
2
G
r
A
=


which is the standard result.
___________________________________________________________________________

The migration of an interface must involve a
net flux of atoms across it. From transition
state theory:

the forward rate is

|
.
|

\
|
=

RT
Q
n A J exp
1 1 2
v

where
2
A is the fraction of atom sites on side 2 able to receive atoms from side 1

1
n is the number of atoms per unit area on side 1 adjacent to side 2
v is the vibration (i.e. attempt) frequency,

and the backward rate is

( )
|
.
|

\
| A +
=

RT
G Q
n A J exp
2 2 1
v .
In equilibrium ( 0 = AG ), there is no net flux, so that

2 2 1 1 1 2
v v n A n A = .

The overall flux (forward minus backward):

(

|
.
|

\
| A

|
.
|

\
|
=
1
RT
G
RT
Q
n A J J exp 1 exp
1 2
v

RT
G
RT
Q
n A
A
|
.
|

\
|
=
1
exp
1 2
v for small departures from equilibrium.
The boundary velocity v is given by the flux the atomic volume (
a mol
N V = O ):
( )
a
mol
N
V
J J v

=
where
mol
V is the molar volume

a
N is Avogadros number.

Overall, for small departures from equilibrium, the interface velocity is:

V
G M v A =
where M is the interfacial mobility, given by

|
.
|

\
|
=
RT
Q
RT N
n A V
M exp
a
1 1 2
2
mol
v
.
M can be reduced, for example, by solute drag:
the maximum effect is when the solute velocity ( a D
L
) is roughly the interface velocity.

The linear dependence of v on
V
G A is representative of an irreversible, dissipative process
(analogous to E j o = ).



3.2.1 Origins of grain structure

Cast alloys are often subsequently deformed (e.g. by rolling) and then annealed. In such
cases, the alloy is likely to recrystallize (nucleate a new grain structure in the solid state).
This is likely to be followed by grain growth, which is often the major influence on the final
grain structure. In thin films, unless there is epitaxy on a single-crystal substrate, a fine-
grained equiaxed structure in contact with the substrate often grows as a columnar grain
structure. Again, there is often subsequent annealing, so that the final microstructure is
largely a result of grain growth.


3.2.2 Normal grain growth

Final microstructures, whether in bulk materials or thin films, are often determined by grain
growth, driven by reduction in the total grain-boundary area (and therefore energy). Grain
growth is significant in metals for
m
5 . 0 T T > .
In normal grain growth:

the average grain size increases with time
the grain structure evolves to remain self-
similar: the grain size distribution
(typically approximately log-normal) is
time-invariant.



Grain growth involves boundary migration
under the action of boundary curvature:










Each boundary moves with a velocity given by:

|
|
.
|

\
|
+ =
2 1
1 1
r r
M v
Geometrical similarity dictates that:
D r r etc. , ,
2 1

where D is the average grain diameter. Therefore

D t
D 1
d
d

and so,
Kt D D + =
2
0
2

Kt ~ if
0
D is very small.

Therefore we expect that for ideal grain growth:

2 1
t D .
This behaviour is indeed found for soap froths and for very pure metals at high temperature.
Otherwise,
n
t D , where 5 . 0 < n , typically 0.3. This deviation from the ideal behaviour is
largely due to solute drag, caused by segregation of solute to grain boundaries.


Many solutes segregate to boundaries, but there is less segregation at higher temperature and
entropy is favoured:

|
.
|

\
| A
=
RT
G
X X
b
0 b
exp
where
0
X and
b
X are the solute mole fractions in the bulk and on the grain boundary.

Grain growth may also be impeded by
second-phase particles. This is Zener drag.
A spherical particle of radius r exerts a drag
force tr on a migrating grain boundary,
where is the grain-boundary energy.





3.2.3 Abnormal grain growth

(also known as secondary recrystallization)





2
H
2
O


3.2.4 Grain growth in thin films

An additional mechanism for pinning grain
boundaries is thermal grooving at the
surface. The groove pins the boundary, but
the groove can form only if the boundary
velocity falls below a critical value.

The pinning therefore occurs in the later stages of grain growth; the phenomenon is called
stagnation and it stops further grain growth.
In narrow lines of metallization on an IC,
grain growth would ultimately lead to a
bamboo grain structure:
This would be very good for electromigration
resistance.



But stagnation often prevents this being
achieved. The resulting variability in
transport (some bamboo segments, some
polycrystalline segments) is very bad for
reliability.








3.3 Kinetics of the Solid-Liquid Interface

3.3.1 Contributions to the total supercooling

We will focus on the interface itself, but we need to recognise that the rate of solidification
depends on a number of factors.

We consider the growth of a dendrite into a
supercooled alloy melt. Before the dendrite
is nucleated and starts to grow, the liquid is at
a uniform temperature
bulk
T that is below the
liquidus temperature of the alloy
liq
T . We
focus on the conditions at the dendrite tip.

These are different from what we might
expect for four reasons:

1. The release of latent heat raises the temperature at the tip (i.e. the tip is less supercooled
than we might have expected):

bulk tip th
T T T = A
2. Rejection of solute by the growing solid raises the local solute content in the liquid and
lowers the liquidus temperature:

tip liq, bulk liq, sol
T T T = A
3. Curvature of the dendrite tip (radius r) lowers the effective liquidus temperature still
further:

V
curv
2
S r
T
A
= A


4. Finally some kinetic supercooling
kin
T A is needed to drive the motion of the interface
itself.

So the total supercooling (
bulk liq total
T T T = A ) has four components (thermal, solutal,
curvature and kinetic):

kin curv sol th total
T T T T T A + A + A + A = A .


3.3.2 Kinetic supercooling

We focus on the kinetic supercooling
kin
T A , although this is usually very small and not rate-
controlling. We consider a non-faceted solid-liquid interface such as is usual for metals:
atoms add anywhere on the surface of the solid (not just at growth ledges)
there is continuous, normal (i.e. in the direction perpendicular to the interface) growth.

In considering interface migration (3.1), the velocity was found to be of the form
| |
(

|
.
|

\
| A
=
RT
G
factor kinetic v exp 1 .
For small supercooling, this approximates to:
| |
RT
G
factor kinetic v
A
=
and the velocity is linearly proportional to the supercooling (as T G A A ).

For large supercooling, the thermodynamic factor saturates (all atoms jump in the same
direction):
1 exp 1
(

|
.
|

\
| A

RT
G

so that the kinetic factor is seen to be the limiting velocity
0
v . For metals, two possibilities
are considered for understanding
0
v :

1.
0
v is diffusion-limited
this does not refer to limitation by long-range diffusion, but rather is the concept that the
atomic motion at the solid-liquid interface during solidification is similar to the jumps
necessary for diffusion in the liquid. This has been our picture so far, and it gives:

1 L
0
s m 10

~ ~
a
D
v
where
L
D is the atomic diffusivity in the liquid and a is the atomic diameter. But in highly
supercooled melts, measured dendrite velocities can exceed
1
s m 100

(and the true
0
v is
presumably still higher). So we need another approach.

2.
0
v is collision-limited

here the concept is that solidification does not
involve diffusive jumps; rather the atoms merely
shuffle their positions as the solid-liquid
interface moves past them. In this case:

1
0
s m 1000 sound of speed

~ ~ v .
Collision-limited growth appears to apply to
pure metals and disordered solid solutions (no
chemical ordering is needed at the interface).

Whether diffusion-limited or collision-limited, interfacial supercoolings are very small. Even
for high solidification velocities
K 10
2
kin

s AT .
So, in analysing a solidification problem it is usually an excellent approximation to take all
parts of the solid-liquid interface to be at the local equilibrium temperature.

3.3.3 Loss of local equilibrium: Solute trapping

Rapid solidification can give extended solid
solubility. But it is usually considered that
the solid and liquid compositions at the
interface lie on the extrapolated solidus and
liquidus lines. This is expressed by saying
that the composition of the solid that is
forming still obeys local equilibrium:

i
L 0
i
S
X k X =

where
0
k is the equilibrium partition coefficient. But we have seen that for collision-limited
growth, velocities can be much greater than the typical diffusive speed of solute in the liquid
~
1
s m 10

:

a
D
v
L
>>
Under such rapid solidification, there must then be a loss of local equilibrium at the solid-
liquid interface. The partitioned solute cannot diffuse fast enough to form a solute spike
moving with the interface. Also, the width of any diffusion boundary layer ( v D
L
) would be
much less than the atomic diameter a.

In the limiting case, solute is trapped by the
next layer of solid before there is time for
equilibration with the liquid. This is certainly
a non-equilibrium process because the
chemical potential of the solute increases!


Such solute trapping can, of course, occur only if the chemical potential of the solvent falls
sufficiently to give an overall decrease in the free energy. This is a thermodynamic criterion
for solute trapping. Between the solidus and liquidus lines on a phase diagram there is the
0
T
line, representing the equilibrium temperature for partitionless freezing. The thermodynamic
criterion is then simply that
0
T T < .





The kinetic criterion for solute trapping is that the interface velocity be sufficient. The solute
partition coefficient is dependent on the interface velocity:

( )
( )
L
L 0
1 D av
D av k
k
+
+
=
where k is the velocity-dependent
partition coefficient
( 1 for complete solute trapping)

0
k is the equilibrium partition
coefficient
a is the atomic diameter

L
D is the diffusivity of the solute
in the liquid.





Some data from laser-quenching
experiments on the velocity-dependent
partition coefficient for Bi in Si. Note
that the equilibrium partition coefficient
0
k

for Bi in Si is only
4
10 7

, so that
very considerable enhancements in
solubility have been achieved. The line
fitted to the data has been calculated
using a stepwise growth model.

Solute trapping is important technologically in the heavy doping of silicon by ion
implantation. The implantation is necessary for some dopants that cannot be diffused in
because they have very low equilibrium solubilities. Also, implantation allows excellent
control of the dopant concentration as a function of depth. But the implantation damages and
ultimately amorphizes the Si. There is a need to regrow crystalline Si. Normal surface
melting allows epitaxial regrowth of high quality (low defect density) Si to the surface, but
the relatively slow regrowth allows time for partitioning of the solute (dopant) back to the
surface. This is avoided by rapid solidification, achieved by laser melting and regrowth:

Typically:
Implanted profile peaks 60-100 nm
Amorphization 150-200 nm
Melt depth 300 nm


The rapid diffusional mixing in liquid silicon
gives a broadened profile of As
concentration. The As profile after regrowth
can be explained only if there is complete
solute trapping ( 1 = k ). After regrowth, all
the As is on substitutional sites, as is
necessary for it to act as a dopant. The
maximum As concentration exceeds the
equilibrium solubility limit by a factor of
four.



4. DIFFUSIONAL CONTROL OF TRANSFORMATION RATES

4.1 Precipitation




For a plate growing in thickness x,
Dt
X
X
x

ss
2
A
A
= .
For a growing sphere of radius r,
Dt
X
X
r
2 1

ss
2
|
|
.
|

\
|
A
A
~ .
Thus we can often take diffusion control to mean that a characteristic linear dimension will
scale with the square root of time. However, this does not apply:

1. in the very early stages of growth, when there is interface control (interfacial processes are
rate-limiting, rather than long-range diffusive transport);

2. in the late stages of growth when there is soft impingement of the diffusion fields (we can
no longer treat the precipitates as growing independent of each other).

We now examine how other forms of time dependence (not only t r ) can arise under
diffusional control.


4.2 Ostwald Ripening

This is the coarsening of precipitate distributions driven by the precipitate-matrix interfacial
energy.

Consider precipitates of a pure phase |
(consisting of B atoms only) dispersed in
a matrix of o phase, which we take to be
an ideal solution of B atoms in A. The
curved interface on the precipitates,
raises their internal pressure and their
free energy per unit volume by

r
G
2
V
= A
where r is the particle radius.



Comparing two neighbouring precipitate
particles, if
2 1
r r < , then ( ) ( )
2 1
r X r X > ,
giving us the mechanism for coarsening (the
big get bigger):

To analyse the kinetics of this process, we need to calculate how much

X is altered by
interface curvature.

The |-phase precipitates are taken to be nearly pure B. Given that their free energy when of
radius r is increased by:

r
G
2
V
= A
the chemical potential of B atoms in the precipitates (and of B atoms in equilibrium
concentration in the surrounding o) is raised by:

r
V
mol
B
2
= A
where the molar volume
mol
V volume (
1 3
mol m

) just converts quantities per unit volume to
per mole.

For an ideal solution
X RT G ln
B B
+ = (writing
B
X as X for simplicity).
Then

2
mol B
2
d
ln d
d
d
r
V
r
X
RT
r

= = .
Integrating from = r to r r =

( )
( ) RTr
V
X
r X
mol

2
ln

=


where ( ) r X

is the equilibrium mole fraction of B atoms in o around a particle of radius


r.
Since
( )
( )
1

~
X
r X
, we can make the approximation that:

( )
( ) RTr
V
X
r X
mol

2
1

+ =


so that
( ) ( ) ( ) =

mol

2
X
RTr
V
X r X

.


Thus, within the matrix adjacent to precipitate particles, the enhancements of local mole
fraction of B are proportional to r 1 .


As Ostwald ripening proceeds, there is a
range of particle size, and a self-similar
size distribution is maintained. Thus the
interparticle distances are proportional to
the average radius r . Therefore
concentration gradients are proportional to
2
1 r .

The rate of coarsening of the average grain radius r is then of the form

2
1
d
d
r t
r
.
Integrating
t r r
3
0
3

and for a small initial radius
0
r

3 1
t r .
This is the characteristic time dependence when the dominant diffusion is through the bulk
lattice. There are other outcomes for other diffusion paths.


4.3 Eutectic Growth

Here two solid phases grow co-
operatively from the liquid. We assume
that the two phases are in the form of
lamellae of roughly equal volume
fractions.

As for Ostwald ripening, interface
curvatures control the rate of the
transformation, but in this case they are
constant throughout. For a eutectic
mixture of o and | phases growing co-
operatively into the liquid, the curvatures
of the o-liquid and |-liquid interfaces are
inversely proportional to the eutectic
spacing
eut
.



We need solute partitioning in the liquid to permit 2-phase growth, and it is necessary that

i
L
i
L
X X >
(noting that these are mole fractions of solute B,
B
X ).

This condition is reached by
supercooling as seen from the
extrapolated liquidus lines in
the phase diagram (above) and
from the underlying free-
energy curves:


Assuming straight phase boundaries on the phase diagram:
T X X X A A =
i
L
i
L
.
We expect the growth velocity to be proportional to the composition gradient of B solute in
the liquid:

eut eut
eut

T X
v
A

A
.
So we can write

eut
1 eut

T
K v
A
= (where
1
K is a constant).
As expected, we can see that a finer eutectic spacing would give faster growth, but what
governs
eut
? This relates to the energy stored in the interfaces between the o and | lamellae
(which is also related to the curvature of the o-L and |-L interfaces). Given an o-|
interfacial energy of (
2
m J

), the stored energy per unit volume of eutectic solid is
eut
2 .
Therefore the effective supercooling is reduced by an amount:

eut
2
eut
1

K
= (where
2
K is a second constant).
Overall, as in the analysis of Jackson and Hunt, we then have:

( )
2
eut
2 1
eut
1
eut
eut 2
1 eut

K K T K K T
K v
A
=
A
=
giving a distinctive functional form:

At 0
eut
= v ,
min eut
=
so that

T
K
A
=
2
min
.

At the extremum

3
eut
2 1
2
eut
1
eut
eut
2
0
d
d

K K T K v
+
A
= = .
Therefore, at this point of maximum velocity

min
2
2
2
=
A
=
T
K
.
For regular eutectic growth, the operating point is taken to be at the extremum. The equations
for
eut
v and for
eut
can be rearranged:

by eliminating
eut
, we obtain:
2
eut
T v A
by eliminating T A , we obtain: constant
2
eut eut
= v .
The latter relation is particularly well verified experimentally.

The influence of interfacial energies in setting the growth condition and determining the
microstructural scale is not only in eutectic growth, but in many other cases:
development of interfacial instabilities (breakdown from planar into cellular solidification)
development of spinodal decomposition
dendrite growth (the selection of dendrite tip radius).

The common features are that the velocity or
rate is limited by interfacial energies at short
length scale and by the rate of solute
diffusion at large , giving the preferred
operating condition at intermediate .




5.1 First- and Second Order Transformations

In a random solid solution:
all sites are equivalent
the probability of finding an A atom on any site =
A
X .

But if
B A mix
X X H O = A (where
|
.
|

\
| +
= O
2
BB AA
AB
E E
E Nz ) is negative, then unlike
neighbours are preferred and SHORT-RANGE ORDER (SRO) can develop. A short-range-order
parameter, s, can be defined:

( )
( ) ( ) random max
random
AB AB
AB AB
N N
N N
s

=
where
AB
N is the number of A-B bonds.

When
B A
: X X is a simple ratio, LONG-RANGE ORDER can develop in which the sites in the
solid solution can be classified into distinct sub-lattices. We will consider just two examples:

|-brass, C-Zn
the random solid solution is bcc
the ordered phase (CsCl structure,
0
2 L ) has two
interpenetrating simple cubic sub-lattices, one
for copper, one for zinc.


Au Cu
3

the random solid solution is ccp
the ordered phase ( Au Cu
3
structure,
2
1 L ) has four
sub-lattices, one for gold and three for copper.



We first consider the Bragg-Williams model for ordering of body-centred cubic into the CsCl
structure, as found in |-brass (equiatomic Cu-Zn). We take N sites in total, half occupied by
A atoms, half by B.

We assume is the fraction of A sites actually occupied by A atoms ( 1 , for perfect
order). Then we can count the number of nearest-neighbour bonds between
A on A site and A on B site ( ) 1
2
z
N

A on A site and B on B site
2
2
z
N

B on A site and A on B site ( )
2
1
2
z
N

B on A site and B on B site ( ) 1
2
z
N

where z is the co-ordination number (8 for Cu-Zn).

Adding the bond energies, we can find the enthalpy H ( internal energy for condensed
phases) of the partially ordered material
( )( ) ( ) ( ) | |
AB
2
BB AA
1 1
2
E E E z
N
H + + + =
2
.
The enthalpy of the fully ordered state is
AB
2
zE
N
H = , so that the disordering energy is
( )
( )
( ) O =
(

+
= A 1
2
1
BB AA
AB dis
E E
E Nz H .
The disordering entropy is obtained by assuming random mixing on each sublattice:
( ) ( ) | | + = A 1 ln 1 ln
dis
Nk S .



These variations of H A and S A with (and
therefore the variation of G A with ) take the
same form as the variation with mole fraction
in a binary solution:



The order parameter: = perfectly random
= 0, 1 ordered.

But it would be preferable to have a long-range-order parameter for which 0 represents
random and 1 represents perfect order. This is the long-range order parameter, q, given by:
1 2 = q .

The Bragg-Williams model treats
order-disorder as a co-operative
phenomenon (analogous to
ferromagnetism). This gives:

The dashed lines indicate the effects of
some ordering on the sub-lattices. This
order-disorder transformation in CuZn
is :
at the critical temperature
c
T , the
enthalpy H and the entropy S are
continuous. There is no latent heat.

Discontinuities appear in the second derivatives of G, e.g., in the specific heat (
T
C
T
G
p
2
2
=
c
c
).



But order-disorder transformations can also
be FIRST-ORDER, for example in Au Cu
3
:





This behaviour arises from free-energy
curves of this kind:




5.2 Anti-Phase Domains

Ordered compounds can be

Reversibly Ordered
For these
liq c
T T < ;
c
T is real.
Examples are: CuZn, Au Cu
3
.

Permanently Ordered
For these
liq c
T T > ;
c
T is virtual.
Examples are: NiAl, Al Ni
3
.


How does ordering occur?

1. A continuous increase in SRO towards LRO
analogous to spinodal decomposition
applies for second-order transformations and at high driving force

2. Nucleation and growth of ordered domains
more common.

Generate anti-phase domains (APDs), with
anti-phase domain boundaries (APBs):

At low supercooling, we expect a large
domain size.

At high supercooling, nucleation is favoured,
and we expect a small domain size.


The formation of a domain structure is followed by COARSENING.

In CuZn
only two types of domain
only one type of APB.
The two APDs interpenetrate and so no metastable domain structure is formed: this leads to
rapid coarsening.

In Au Cu
3

there are four types of domain
these can form a metastable froth analogous to a grain structure
this domain structure shows rather slow coarsening (low APB energy).


Non-stoichiometric alloys
of technological interest.

disordered phase ordered precipitates + disordered matrix

The precipitates are of a different composition from the matrix, and their formation requires
long-range diffusion, e.g. ' precipitates ( Au Cu
3
type,
2
1 L ) in matrix (Ni-based ccp) in
superalloys.

Ordering in phases raises their yield stress, but tends to make them more brittle.
Conventional superalloys combine ordered and disordered phases, and establish a satisfactory
compromise of strength and toughness. There is some interest in using monolithic
intermetallic compounds as refractory alloys, in which case a fine, stable domain structure
(obtainable in compounds that are borderline between the permanently and reversibly ordered
categories) may be a route to improving toughness.


Consider an isothermal reaction in a solid system that is transforming from one state to
another. The volume fraction transformed is F (often directly measurable, for example using
calorimetry).

We might expect F to vary with time,
something like first-order kinetics in a
chemical reaction:
( ) F k
t
F
= 1
d
d
.
Integrating, we find
( ) kt F = exp 1





But this relaxation type of behaviour is not
observed for most solid-state reactions, which
instead show a sigmoidal form. To analyse
such a shape, we need to break down the
transformation into separate mechanisms.


The sigmoidal form arises from nucleation and growth. We start by considering a simple case
in which a glass crystallizes isothermally to a crystalline phase of the same composition
(polymorphic transformation). We assume that spherical crystals grow at a constant rate
t r U d d = . These crystals are growing from a fixed number of nuclei (N per unit volume)
that are active at the start of the anneal. If there is no interference between the crystals (no
impingement), then:

3 3
3
4
t U N F = .
The rate of transformation (
2
d
d
t
t
F
) thus shows clear acceleration in the early stages of
transformation.

The deceleration of the transformation rate in
the later stages is not because of any decrease
in U, but because of impingement that thus
needs to be accounted for. The expression
for F that ignores impingement in fact gives
not F, but the extended volume fraction
transformed
ext
F .



Consider a given growing sphere. In a small
increment, the sphere grows into material that is
already partially transformed, and cannot be
transformed a second time. Therefore the increment
in actual fraction transformed is less than the
increment in extended fraction:
( ) F F F = 1 d d
ext

Integrating
( )
ext
1 ln F F =
and noting that:

ext
F ranges from 0
F ranges from 1 0
we arrive at the Avrami equation accounting for the overlap between transforming volumes.
The volume fraction transformed, F, varies with the extended volume fraction
ext
F (that
which would be transformed in the absence of impingement) according to
( )
ext
exp 1 F F = .
The extended volume fraction normally has the form
n
Kt . In the case we were considering
(fixed number of nucleation events, constant growth rate):

3
ext
Kt F = (where the constant
3
3
4
U N K = )
Another commonly analysed case is that of spheres growing at a constant rate U (
t
r
d
d
= )
from sites nucleating at a constant rate I (
1 3
s m

). In that case

4 3
ext

t IU F
3
= .
The overall reaction kinetics are usually expressed in the form of the Johnson-Mehl-Avrami
equation
( )
n
Kt F = exp 1 .
Thus a plot of ( ) | | t F ln vs 1 ln ln has a gradient of the Avrami exponent, n.

The JMA equation is useful in many situations, and was first used to analyse the
transformation of austenite to pearlite. It is valid only if the distribution of nucleation sites is
random. When growth is diffusion-controlled (e.g. spheres with t r ), the JMA equation
can be used only for the early stages of transformation, before there is any soft impingement
of the diffusion fields.



QUESTION SHEET 1
(to be attempted after Lecture 4)


Revision Question (Part IB Tripos Paper, 1997)

1. For what reasons might a metallic sample show an effective self-diffusivity at
intermediate temperatures larger than that expected from an extrapolation of high-
temperature data?

Show that for an annealed polycrystalline metal, the effective self-diffusivity D is given
by
|
.
|

\
|
+ =
d
D D D
oo
gb latt


where
latt
D and
gb
D are the self-diffusivities in the lattice and grain boundary
respectively, d is the grain diameter, o is the grain-boundary width and o is a constant.

The table below gives self-diffusivities determined for polycrystalline silver with a
grain size of 500 m. From these data estimate (i) the activation energy for
latt
D , (ii)
the activation energy for
gb
D , (iii) the absolute value of
gb
D at 700 K, justifying your
choices of o and o.

T (K) D (
1 2
s m

) T (K) D (
1 2
s m

)
500
21
10 6 . 9

900
16
10 5 . 8


600
19
10 5 . 3

1000
15
10 3 . 9


700
18
10 1 . 5

1100
14
10 1 . 7


800
17
10 2 . 6

1200
13
10 9 . 3



For a silver sample with a grain size of 50 m, suggest a temperature range suitable for
measuring
latt
D .




2. In an equiatomic Ni-Pd alloy the tracer diffusivities of the Ni and Pd atoms each have
an Arrhenius form: ( ) RT Q D D = exp
0
. From the data given below, calculate the
interdiffusivity at 1050 K. How does this differ from the value in an ideal solution?

for Ni:
1 2 4
0
s m 10 3 . 1

= D
1
mol kJ 280

= Q
for Pd:
1 2 5
0
s m 10 1 . 2

= D
1
mol kJ 268

= Q
for the equiatomic alloy the enthalpy of mixing =
1
mol J 1200

+

3. For steady-state diffusion through a plate (solute entering one face, leaving through the
opposite face), what is the form of the concentration profile of the diffusing species
when the diffusivity is independent of concentration? Explain why the diffusivity of
arsenic in silicon is proportional to the concentration of arsenic. What would be the
form of the steady-state concentration profile for arsenic diffusing through a thin plate
of silicon?



4. An aluminium-alloy conductor on an integrated circuit has the composition Al2wt%Cu
corresponding to a copper concentration of
3 26
m atoms 10 1 . 5

). The dissolved copper
atoms have an effective charge in solution of 20. An electrical current density of
2 10
m A 10 1

is passed through the conductor at 400 K, and a copper concentration
gradient develops. By balancing the atomic fluxes of copper due to electromigration
and due to the concentration gradient, estimate the limiting steady-state value of the
concentration gradient (
1
m wt%

) which would be set up. Comment on this value
would significant composition differences develop across a single grain (~1 m in
diameter)?

[for the aluminium alloy under the given conditions, the resistivity = m 10 3
8
O

]



5. Compare the mechanisms of electromigration damage in Al-based and Cu-based
metallizations.



6. A glass-forming liquid shows a temperature-dependent viscosity, as follows.

T (K) q (Pa s)
700
8
10 2 . 1
800
3
10 6 . 5
900 200

Show that 600 K would be a reasonable estimate for the ideal glass transition
temperature for this material.








QUESTION SHEET 2
(to be attempted after Lecture 9)

1. A polycrystalline metal of average grain diameter D has within it a dispersion of
second-phase particles of radius r, where r D >> . Show that the number of particles
touching unit area of grain boundary is

2
2
3
r
f

where f is the volume fraction of second phase. Each particle can exert a Zener drag
force of tr on the grain boundary. This drag limits the grain size obtained by normal
grain growth. By assuming that the average grain boundary curvature in a grain
structure is D 2 , show that the limiting grain diameter is

f
r
D
3
4
max
=
.

Hence determine the maximum grain size obtainable in a dispersion-hardened copper
sample containing a 1% volume fraction of particles of radius 1.0 m.

2. In the diffusion-limited model, it is assumed that advance of a solidification front takes
place by continuous random addition of atoms to the solid by diffusive jumps in the
liquid. With this assumption, and making suitable approximations, show that the kinetic
factor in the expression for the interface velocity (i.e. the limiting velocity
0
v ) is
a D
L
~ (where
L
D diffusivity in the liquid; a atomic diameter). Hence derive a
simple expression for the dependence of interface velocity on interfacial supercooling
kin
T A , and calculate
kin
T A for aluminium with a front advancing at
1
s m 100

.
Comment on the value obtained and on how it would be different under the (more
reasonable) assumption of collision-limited growth. What is the relevance of this
supercooling for determining the rate-controlling step in solidification?

[ nm 3 . 0 ~ a
1
f
mol kJ 5 . 4

= AH K 933
m
= T
1 2 9
L
s m 10 5

= D ]

3. Show that the effective diffusivity in a material in which grain-boundary diffusion is
dominant is inversely proportional to grain size. A polycrystalline material has second-
phase particles at grain-boundary junctions. These particles undergo Ostwald ripening:

( ) ( ) t r t r
n n
0

The particles pin the grain boundaries, making it likely that the microstructure (in terms
of grain size distribution and particle size distribution) remains self-similar as
coarsening proceeds. What is the expected value of n in the above expression?

4. Explain what is meant by eutectic growth at the extremum condition. During
directional freezing of Cu Al - Al
2
eutectic, the interlamellar spacing is 3 m for a
growth velocity of
1
s m 3

. What spacing is expected if the velocity is
1
s m 1

?

5. An amorphous thin film can transform by 2-D growth of circular crystallites. Under
isothermal conditions, these crystallites nucleate at a rate I (
1 2
s m

) and grow at a rate
U (
1
s m

). Derive an expression for the volume fraction transformed as a function of
time, taking account of the impingement of the crystallites.
EXAMPLES CLASS


1. A diffusion couple is prepared by plating pure copper onto a block of o-brass with some
inert marker wires on its surface. On annealing, the zinc diffuses out of the brass faster
than the copper diffuses in, so that the inert markers move into the original brass block.
We can analyse the interdiffusion at the position of the markers. The marker velocity at
the annealing temperature is
1 11
s m 10 6 . 2

. Around the markers, the matrix
composition is 22 . 0
Zn
= X and
1
Zn
m 89

= c c x X . From the complete composition
profile established during the anneal, the interdiffusivity at the markers is found to be
1 2 13
s m 10 5 . 4
~

= D . What are the values of
Zn
D and
Cu
D in the solid solution of
composition 22 . 0
Zn
= X at the annealing temperature? [Relevant Darken relations in
the handout, p. 4.]



2. Explain why grain boundaries move towards their centre of curvature during grain
growth, but away from their centre of curvature around a grain growing during
recrystallization.

A recrystallized, dislocation-free grain is growing in a deformed aluminium sample
with a dislocation density of
2 16
m 10

. Calculate the effective pressure acting on the
boundary of the recrystallized grain. What is the critical radius of a spherical nucleus
from which the recrystallized grain could grow?

[shear modulus, Pa 10 6 . 2
10
= G Burgers vector, nm 29 . 0 = b
grain-boundary energy,
2
m J

0.5 = ]



3. Show that for a crystal growing in a supercooled liquid by the ledge mechanism, the
lateral velocity of ledges,
ledge
v , is related to the normal velocity of the solid-liquid
interface, v, by:

u sin
ledge
v
v =

where u is the angle by which the plane of the solid-liquid interface deviates from the
ideal crystallographic orientation of the ledges.

In laser-quenching experiments on single-crystal silicon ion-implanted with bismuth, it
is found that, for a constant regrowth velocity of
1
s m 5

, the effective partition
coefficient k is dependent on crystal orientation, as shown in Figure 1. By plotting k
against ( )
ledge 10
log v for example, for u = 5, 10, 20 and 40 and comparing with the
figure on p. 14 of the handout, offer a qualitative explanation of the form of Figure 1.

4. A growing spherical precipitate increases its radius r with time t according to

Dt
X
X
r

ss
2
A
A
~

where

X A is the difference in the solute mole fraction in the precipitate and in the
matrix at the precipitate-matrix interface, and
ss
X A is the supersaturation. Show that in
the case where the matrix and composition are both nearly pure components (i.e. where
1

~ AX ), the growth rate v of a particle is given by



ss
d
d
X
r
D
t
r
v A ~ = .

During Ostwald ripening, the supersaturation can be measured relative to the average
solute level in the matrix ( ) r X

, taken to be that in equilibrium with particles of the


average radius r , as shown in Figure 2. Given the form of variation of ( ) r X

with r
(handout, p. 15), show that the particles grow with a velocity given by

( )
|
.
|

\
|

=
r r rRT
X V D
v
1 1 2
m

.

Which particles grow and which shrink? How does this relate to Ostwald ripening? For
which particle radius is the growth rate maximum? Assume that this maximum
represents the rate of coarsening, i.e., that

max
d
d
d
d
|
.
|

\
|
~
t
r
t
r
.

Hence show that

( ) ( ) t r t r 0
3 3


(The same result was derived somewhat more simply, but less rigorously, in the course
handout, p. 15.)







Full worked solutions to this Examples Class, and to the two Question
Sheets on C7, will be given out just after the Examples Class

Vous aimerez peut-être aussi