Chapter 2.

Linear and Nonlinear Diffusion

• The subject of linear diffusion is first discussed with the two governing laws
“derived”/analyzed;

• We will highlight one of the key assumptions made during the above derivation,
which serves as the basis for the need of discussing nonlinear diffusion;

• The nonlinear effects becomes obvious and non-negligible when the gradient of the
chemical potential becomes large, as is often the case in nano-scale structures;

MSE 201 Ya-Hong Xie

 2.1 Introduction

• The importance of diffusion to our lives can be traced back in history with
examples such as C diffusing into Fe to make steel, n-type and p-type dopants
diffusing in Si to make integrated circuits, and Li diffusing in Si and other
matrix in energy technologies;

• Experimental observations show that when the concentration gradient or the

chemical potential gradient driving diffusion is small, e.g. diffusing distance
being >100 nm, linear expressions such as Fick’s first and second laws can be
used to satisfactorily predict the diffusion outcomes. However, when the
diffusing distance is reduced to e.g. 10 nm or 1 nm, these laws do not work as
well and higher order factors need to be taken into consideration.

MSE 201 Ya-Hong Xie

 2.2 Linear Diffusion

𝜕𝐶
Fick’s first law: 𝐽 = −𝐷 Units:
𝜕𝑥
• J: # atoms/cm2s
Linear diffusion: Flux being
• C: # atoms/cm3
proportional to driving force
• x: cm;

• D: cm2/s

• This form of flux is highly similar to that of thermal conduction, the conduction of
electricity, viscosity in fluid flow, and others;

• The diffusion mechanism we are interested in this course is substitutional diffusion.

The basic assumptions for Fick’s first and second laws are given on the next slide.

MSE 201 Ya-Hong Xie

 2.2 Linear Diffusion

• It is a thermally activated uni-molecular process in contrast to e.g. most

chemical reactions in which more than one atoms participating;

• It is a defect mediated process requiring vacancies thus Kirkendall effect;

• There exists an activation energy barrier with an atom’s chance of scaling the
barrier following Boltzmann’s equilibrium distribution function;

• The diffusion process takes place under close to equilibrium conditions, i.e.
the system is not in an extremely excited state;

• The direction of individual jump of atoms is in the way of random walk;

• Diffusion of atoms on a micro- or macroscopic scale requires a chemical

potential gradient, or a “driving force”.

MSE 201 Ya-Hong Xie

 2.2.1 A Close Look at Atom Flux

#𝑎𝑡𝑜𝑚 1 #𝑎𝑡𝑜𝑚 𝑐𝑚 #𝑎𝑡𝑜𝑚 𝑐𝑚2

× = × = ×
𝑐𝑚2 𝑠 𝑐𝑚3 𝑠 𝑐𝑚4 𝑠

The three ways of conceptualize atom diffusion flux and the physics implication:

• Left-hand: A plane of atoms jumping at a frequency equals to 1/s;

• Middle: A volume of atoms diffusing at a given velocity;

• Right-hand: the product of diffusivity (cm2/s) and concentration gradient, i.e.

Fick’s first law.

MSE 201 Ya-Hong Xie

2.2.2 A Close Look at Fick’s First Law

a) Without a driving force or chemical potential gradient;

b) With a driving force or chemical potential gradient;

MSE 201 Ya-Hong Xie

 The Physics Description of Single Atom Jumps

The frequency of successful jumps to the neighboring lattice site via saddle points:

−Δ𝐺𝑚
𝜈 = 𝜈0 exp
𝑘𝑇

where n0 is the attempt frequency

MSE 201 Ya-Hong Xie

 The Physics Description of Single Atom Jumps

Under the influence of a chemical potential gradient or a driving force:

𝜆 𝜆
Δ𝜇 = Δ𝑥 tan 𝜃 = tan 𝜃 𝑤ℎ𝑒𝑟𝑒 Δ𝑥 ≡
2 2
Δ𝜇
𝐹=− = tan 𝜃
Δ𝑥

MSE 201 Ya-Hong Xie

 The Physics Description of Single Atom Jumps

The frequency of net jumps in the forward direction:

−Δ𝐺𝑚 + Δ𝜇 𝜆𝐹
𝜈 + = 𝜈0 exp = 𝜈 exp +
𝑘𝑇 2𝑘𝑇

−Δ𝐺𝑚 − Δ𝜇 𝜆𝐹
𝜈 − = 𝜈0 exp = 𝜈 exp −
𝑘𝑇 2𝑘𝑇

𝜆𝐹 𝜆𝐹 𝜆𝐹
∴ 𝜈𝑛 = 𝜈 + − 𝜈 − = 𝜈 exp + − exp − = 2𝜈 sinh
2𝑘𝑇 2𝑘𝑇 2𝑘𝑇

𝑒 𝑥 − 𝑒 −𝑥
𝑟𝑒𝑐𝑎𝑙𝑙 sinh 𝑥 =
2

𝜆𝐹 𝜆𝐹
When the driving force is small, i.e.: << 1 then: 𝜈𝑛 = 𝜈
2𝑘𝑇 𝑘𝑇

MSE 201 Ya-Hong Xie

 Fick’s First Law is Valid Only for Small Driving Forces

Under this core assumption to linear diffusion, i.e. small driving force:

𝜆𝐹
𝜈𝑛 = 𝜈 note that n is independent of F;
𝑘𝑇

We can write out the velocity of diffusion:

𝜈𝜆2
𝒗 = 𝜆𝜈𝑛 = 𝐹
𝑘𝑇

𝜈𝜆2 𝜈𝜆2
∴ 𝐽 = 𝐶𝒗 = 𝐶 𝐹 = 𝐶𝑀𝐹 𝑤ℎ𝑒𝑟𝑒 𝑀= (2.8)
𝑘𝑇 𝑘𝑇

M is known as atomic mobility in the unit of cm2/eV-s;

Note that C in Eq 2.8 is that of vacancy NOT that of the atoms!!

MSE 201 Ya-Hong Xie

 More on the Chemical Potential Pertaining to Diffusion

Consider an ideal solution, which by definition has DHmix=0:

Δ𝐺𝑚𝑖𝑥 = Δ𝐻𝑚𝑖𝑥 − 𝑇Δ𝑆𝑚𝑖𝑥 = −𝑇 𝑋𝐴 ln𝑋𝐴 + 𝑋𝐵 ln𝑋𝐵

For a dilute alloy assuming XB<<1, i.e. XA~1:

𝑑Δ𝐺𝑚𝑖𝑥
𝜇𝐵 = = 𝑘𝑇 ln𝑋𝐵 ; 𝜇𝐴 = 𝑘𝑇 ln𝑋𝐴 ≈ 𝑘𝑇 ln( 1) ≈ 0; 𝜇 ≈ XB 𝜇𝐵
𝒅𝐶𝐵

In other words, the chemical potential (pertaining to the free-energy of mixing) of an

ideal solution/alloy is dominated by the dilute component, or the solute, NOT the solvent.

MSE 201 Ya-Hong Xie

 2.2.3 Different Ways of Representation of Concentration

The relationship between the fractional concentration (Xi) and concentration (Ci):

𝐶Ω = 1

𝑋𝐴 𝑋𝐵
𝐶𝐴 = ; 𝐶𝐵 =
Ω Ω

Note that Xi is unitless but C is in the unit of density: # of atoms/volume;

MSE 201 Ya-Hong Xie

 Fick’s First Law is Valid Only for Small Driving Forces

The relation derived on the previous slide allows us to establish:

Δ𝜇𝐵 Δ𝜇𝐵 Δ𝑋𝐵 𝑘𝑇 Δ𝑋𝐵

𝐹=− =− =−
Δ𝑥 Δ𝑋𝐵 Δ𝑥 𝑋𝐵 Δ𝑥

𝜈𝜆2 𝑋𝐵 𝜈𝜆2 𝑘𝑇 Δ𝑋𝐵 𝜕 𝑋𝐵 /Ω 𝜕𝐶𝐵

𝐽𝐵 = 𝐶𝐵 𝐹= − = −𝜈𝜆2 = −𝐷𝐵
𝑘𝑇 Ω 𝑘𝑇 𝑋𝐵 Δ𝑥 𝜕𝑥 𝜕𝑥

As such, we have successfully “derived” Fick’s first law based on the

assumptions given on the second slide of Chapter 4.

As will be shown later in this chapter, the key assumption for linear diffusion, i.e.
small driving force is often invalid in nano-scale materials and thus the need for
discussion nonlinear diffusion.

MSE 201 Ya-Hong Xie

Fick’s Law being a Direct Consequence of the Small Driving Force

𝐽
Furthermore: = 𝐷 = 𝜈𝜆2
− 𝜕𝐶/𝜕𝑥

𝐷
𝑀=
𝑘𝑇

This is sometimes known as Einstein relation, frequently used in electron transport in

semiconductors relating the mobility (designated as m in this case) and the diffusivity D.

MSE 201 Ya-Hong Xie

 2.2.4 Fick’s Second Law of Diffusion

It should be realized that Fick’s first law assumes thus is only applicable to steady
states where the flux does not change with time and location (within a range);

Most situations encountered when studying diffusion involve flows of atoms that
changes with both time as well as location to which Fick’s first law is not applicable;

The equation used for describing transient behaviors of diffusion is Fick’s second
law. It can be derived based on the consideration of, thus is a mathematical formulism
of mass conservation:

the mass flowing out of an enclosed volume must equal to the decrease of mass
inside the volume.

MSE 201 Ya-Hong Xie

 Fick’s Second Law of Diffusion

Fick’s second law of diffusion is a mathematical statement of mass conservation.

It can be “derived” as follows.

Consider an imaginary cube volume as shown in the figure, the net flow through
each face is the dot product of the flow vector J with the surface norm:

MSE 201 Ya-Hong Xie

 Fick’s Second Law of Diffusion

The rate of change of the number of atoms inside the cube can be calculated as follows:

Δ𝑁1 = −𝐽1 𝐴1 Δ𝑡

6 6

෍ 𝐽𝑖 𝐴𝑖 Δ𝑡 = ෍ −Δ𝑁𝑖 = −Δ𝑁 The sign of DN is positive if the

𝑖=1 𝑖=1
number inside the cube increases.

Δ𝑁
Δ𝐶 = ±
𝑉

6
𝑑𝑁 𝑑𝐶
=𝑉 = ෍ 𝐽𝑖 𝐴𝑖 (2.15)
𝑑𝑡 𝑑𝑡
𝑖=1

MSE 201 Ya-Hong Xie

 Fick’s Second Law of Diffusion

The expression on the previous slide can be expressed in a more compact fashion:
6

෍ 𝐽𝑖 𝐴𝑖 = ∇ ⋅ 𝐽 𝑉
𝑖=1

෍ 𝐽𝑖 𝐴𝑖 = ර 𝐽Ԧ ⋅ 𝑛 𝑑𝐴 = ∇ ⋅ 𝐽 𝑉 (2.17) Gauss theorem

𝑖 𝐴

𝑑𝐶
∴𝑉 = ර 𝐽Ԧ ⋅ 𝑛 𝑑𝐴 = ∇ ⋅ 𝐽 𝑉 (2.18)
𝑑𝑡
𝐴

𝑑𝐶
∴ = −∇ ⋅ 𝐽 (2.19)
𝑑𝑡

Gauss theorem is a mathematical formulation of continuity as a direct result of

conservation of a quantity, which could be mass, momentum, energy, etc.
MSE 201 Ya-Hong Xie
 Fick’s Second Law of Diffusion

The expression takes on a simpler form in a one-dimensional problem:

𝜕𝐶
𝐽 = −𝐷
𝜕𝑥

𝜕𝐶 𝜕 𝜕𝐶
= 𝐷
𝜕𝑡 𝜕𝑥 𝜕𝑥

𝜕𝐶 𝜕2𝐶
∴ =𝐷
𝜕𝑡 𝜕𝑥 2

The expression in the red box is the form of Fick’s second law commonly taught
in classrooms.

Note that because we used the expression of Fick’s first law, there is the implicit
assumption of small driving force.

MSE 201 Ya-Hong Xie

 2.2.5 Flux Divergence
Fick’s second law can be “visualized” from yet another perspective: flux divergence.

The discussion below references the figure on this slide and considers only the flow
along x-axis:

𝐽Ԧ𝑥1 ⋅ 𝑥Ԧ1 = 𝐽𝑥1 𝑥1 cos 180∘ = −𝐽𝑥1 𝑑𝑦𝑑𝑧

𝜕𝐽𝑥 𝜕𝐽𝑥
𝐽Ԧ𝑥2 ⋅ 𝑥Ԧ2 = 𝐽𝑥1 + 𝑑𝑥 𝑥2 𝑐𝑜𝑠0∘ = 𝐽𝑥1 𝑑𝑦𝑑𝑧 + 𝑑𝑥𝑑𝑦𝑑𝑧
𝜕𝑥 𝜕𝑥

𝜕𝐽𝑥
𝐽𝑥2 − 𝐽𝑥1 𝑑𝑦𝑑𝑧 = 𝑑𝑥𝑑𝑦𝑑𝑧
𝜕𝑥

This last equation represents the net

flow out of the box along x-axis.

MSE 201 Ya-Hong Xie

 Flux Divergence

In deriving the last equation on the previous slide, we used the fact that since flux
is a physical quantity, it is continuous and smooth which means:

𝑑𝐽𝑥
𝐽𝑥2 = 𝐽𝑥1 + 𝑑𝑥
𝑑𝑥

Carrying out similar analyses along y- and z-axes, we have:

𝑑𝐽𝑦
𝐽𝑦2 − 𝐽𝑦1 𝑑𝑥𝑑𝑧 = 𝑑𝑥𝑑𝑦𝑑𝑧
𝑑𝑥

𝑑𝐽𝑧
𝐽𝑧2 − 𝐽𝑧1 𝑑𝑥𝑑𝑦 = 𝑑𝑥𝑑𝑦𝑑𝑧
𝑑𝑥

We have also used the symbol: xi=nxidydz; yi=nyidxdz; and : zi=nzidxdy

MSE 201 Ya-Hong Xie

 Flux Divergence

Summing up flows along x-, y- and z-axes:

6
𝜕𝐽𝑥 𝜕𝐽𝑦 𝜕𝐽𝑧
෍ 𝐽𝑖 𝐴𝑖 = + + 𝑑𝑉 𝑤ℎ𝑒𝑟𝑒 𝑑𝑉 = 𝑑𝑥𝑑𝑦𝑑𝑧
𝜕𝑥 𝜕𝑥 𝜕𝑥
𝑖=1

Expressing the summation in integral form, we have the well-known Gauss theorem:

𝜕𝐽𝑥 𝜕𝐽𝑦 𝜕𝐽𝑧

න 𝐽 ⋅ 𝑛𝑑𝐴 = ∇ ⋅ 𝐽 𝑉 = + + 𝑉
𝜕𝑥 𝜕𝑥 𝜕𝑥
𝐴

𝜕𝐶
− ∇⋅𝐽 =
𝜕𝑡

We have arrived once again Fick’s second law but from the angle of flux divergence

MSE 201 Ya-Hong Xie

2.2.6 Tracer Diffusion for the Measurement of D in Self-diffusion
Apparently, diffusivity (D) can be said to be the most important quantity to know
in order for us to be able to correctly predict any diffusion process.

Tracer experiment is a commonly employed experimental technique determining

the diffusivity (D) of self-diffusion. Diffusion of B in A is a much simpler
situation without the need for the use of tracers most of the time.

The bases of this approach are as follows:

1. Isotopic atoms with only the mass difference of one neutron is expected to
possess the same diffusivity;

2. There are techniques allowing for precision measurement of the

concentration profile of the isotope.

3. Use the sample geometry shown on the next slide to simplify a 3-D problem
into a 1-D one.

MSE 201 Ya-Hong Xie

 2.2.6 Tracer Diffusion

Here we use the sample geometry as shown in panel (a) of the figure:

Here in panel (b) we expand

conceptually the real experiment
into a symmetrical one to simplify
the mathematics involved in
predicting
: the outcome of the
experiment using Fick’s second
law.

MSE 201 Ya-Hong Xie

 2.2.6 Tracer Diffusion

The solution form of Fick’s second law under the boundary/initial condition of the
imaginary symmetric experiment is:

𝑄 −𝑥 2 please verify that this is a solution of

𝐶(𝑥, 𝑡) = 𝑒𝑥𝑝
𝜋𝐷𝑡 4𝐷𝑡
Fick’s second law
𝑄
𝑎𝑡 𝑥 = 0, 𝐶 → 𝑎𝑠 𝑡→0 Q – total amount of isotope;
𝐴𝑑
𝑓𝑜𝑟 𝑥 > 0, 𝐶 → 0 𝑎𝑠 𝑡→0 A – area of the end face;

d – isotope thickness
∞ ∞
Mass conservation further requires: න 𝐶𝑑𝑥 = 𝑄 or න 𝐶𝑑𝑥 = 2𝑄
0 −∞

MSE 201 Ya-Hong Xie

 2.2.6 Tracer Diffusion

There are two important concentration values that are helpful to be written out
explicitly from the solution to Fick’s second law:

𝑄
𝐶(0, 𝑡) = 𝐶0 = 𝑎𝑡 𝑥=0
𝜋𝐷𝑡

𝐶0
𝐶 𝜆𝐷 , 𝑡 = 𝐶𝜆 = 𝑎𝑡 𝑥 = 𝜆𝐷 = 4𝐷𝑡
𝑒

These values are shown in panel (c) on the slide before the previous one.

Note that the scale x2~Dt is a characteristic distance to any diffusion process
for getting an order of magnitude idea about the extent of diffusion.

MSE 201 Ya-Hong Xie

 2.2.6 Tracer Diffusion

Assuming the concentration profile, i.e. C(x,t) of the isotope can be precisely
measured, one could then determining D(T) and the associated energy barrier (E)
and the pre-exponential factor (D0) by, for example, performing the diffusion
experiment at different T with a fixed t, and using the last equation on the
previous slide.

Specifically, we could experimentally determine lD(T), and thus D(T);

Next, plot D(T) in an Arrhaneus plot based on the Boltzmann factor of D as

follows: −E
𝐷 = 𝐷0 exp
𝑘𝑇

From the slope of the Arrhaneus plot, both D0 and E can be determined.

MSE 201 Ya-Hong Xie

2.2.7 Diffusivity: vacancy mediated process scrutinized

Given the importance we have seen of diffusivity D pertaining to especially

to the theme of this course, namely kinetics, it is natural to proceed
with more discussion about it.

We will present conceptually deeper insight into factors influencing D

(section 2.2.7) and next the techniques for experimentally determining
D (section 2.2.8)

MSE 201 Ya-Hong Xie

 2.2.7 Diffusivity: conceptual examination

We begin by calculating the probability that an atom is neighboring a vacancy (a

necessary condition for a substitutional diffusion event to occur) by considering the
vacancy formation energy DGf and the fractional concentration of the lattice sites
being vacancies: 𝑛𝑣 Δ𝐺𝑓
= exp −
𝑛 𝑘𝑇

In identifying the above expression with the probability, we have made one
(unsaid) assumption:
The probability of an atom is neighboring a vacancy is dictated by the
availability of vacancies, NOT that of the atoms, which could occur if we are
dealing with the situation a few atoms being in a vacuum (abundance of
vacancies).

MSE 201 Ya-Hong Xie

 2.2.7 Diffusivity: conceptual examination

A second requirement for a substitutional event to take place is that an atom must
be the immediate neighbor to a vacancy, with a probability being dependent on the
specific crystalline symmetry. This can be expressed as a quantity known as nc , the
nearest neighbor coordinates. For FCC crystals, nc equals to 12.

Factoring in nc with the consideration of vacancy concentration shown on the

previous slide, the probability of an atom-vacancy exchange can be written as:

𝑛𝑣 Δ𝐺𝑓
𝑛𝑐 = 𝑛𝑐 exp −
𝑛 𝑘𝑇

MSE 201 Ya-Hong Xie

 Diffusivity: the correlation factor

Continuing our conceptual development of the physics model of diffusion, we need

to consider the fact that during the process of a swap in place between an atom and a
vacancy, there is a instantaneous change in lattice strain. The probability of a reverse
process could be rather high before the lattice strain can be relaxed, which take typically
a time duration on the order of pico-second.

This is represented by a quantity known as correlation factor. The correlation factor

f relates to the probability of “reverse jumps”, i.e. after swapping place with a vacancy,
the process can be reversed before the lattice fully relaxes to equilibrium configuration.
f=0 if the probability is 100%; typically 0<f<1;

MSE 201 Ya-Hong Xie

 Diffusivity: the correlation factor

The correlation factor is dependent on the specifics of a crystal. For FCC metals,
f=0.78. Factoring in f, we have the functional expression of D:

Δ𝐺𝑚 + Δ𝐺𝑓
2
𝐷 = 𝑓𝑛𝑐 𝜆 𝜈0 exp − recall D=nl2
𝑘𝑇

Δ𝑆𝑚 + Δ𝑆𝑓 Δ𝐻𝑚 + Δ𝐻𝑓 𝑬

𝐷 = 𝑓𝑛𝑐 𝜆2 𝜈0 exp exp − = 𝐷0 exp −
𝑘 𝑘𝑇 𝑘𝑇

where thermally activated hopping rate as shown below is assumed:

−Δ𝐺𝑚
𝜈 = 𝜈0 exp
𝑘𝑇

MSE 201 Ya-Hong Xie

 2.2.7 Diffusivity: conceptual examination
Therefore:
Δ𝑆𝑚 + Δ𝑆𝑓
𝐷0 = 𝑓𝑛𝑐 𝜆2 𝜈0 exp 𝑎𝑛𝑑 𝑬 = Δ𝐻𝑚 + Δ𝐻𝑓
𝑘

These expressions provide clear physical meanings to both D0 and E.

Using the table on the next slide, which was established using tracer
experiments, self-diffusion of Al at room temperature can be calculated:
Δ𝐻
𝐷 = 𝐷0 exp − = 0.047𝑒 − 1.28×23,000 / 2×300
=
𝑘𝑇

= 0.047 × 10− 1.28×23,000 / 2.3×2×300

= 0.047 × 10−22 𝑐𝑚2 /𝑠

which translates to a diffusion distance of 1 Å per day.

MSE 201 Ya-Hong Xie

:
MSE 201 Ya-Hong Xie
 2.2.7 Diffusivity: conceptual examination

As an engineer, it is helpful and often necessary to have order of magnitude

level of idea about certain important quantities, in this case D and the
consequences in practice.

The first order matter to keep in mind is that for typical metals at temperatures
significantly below their respective melting temperatures, 𝑫~10−22 𝑐𝑚2 /𝑠
is a common value. When at Tm, D~ 10-6 cm2/s typically.

MSE 201 Ya-Hong Xie

Diffusivity: the difference between substitutional and interstitual

Diffusion coefficient of
various dopants in Si

MSE 201 Ya-Hong Xie

 Diffusivity: the difference between substitutional and interstitual
• Throughout our discussion till now, substitutional diffusion, i.e. diffusion takes
place through the exchanging places between a diffusing atom and a vacancy has
been assumed.

• It needs to be pointed out that there exists an entirely different diffusion mechanism,
namely, interstitual diffusion, i.e. diffusing atoms zip through the lattice in between
lattice sites independent of vacancies. Such diffusion mechanism is typically
dominant in the case of small atoms such as Li diffusing through Si as shown in the
figure on the previous slide with much higher D at low T and smaller slope.

• One unique case is Cu diffusing through Si with majority of the atoms take
substitutional path with only a tiny fraction being via interstitual pathways, which
ended up having an overwhelming impact on the outcome.

MSE 201 Ya-Hong Xie

 2.2.8 Diffusion: experimental measurements

The equations derived before illustrates that the entropy portion of the diffusion
barriers contribute to the pre-exponential part of D whereas the enthalpy
portion contribute to the Boltzmann factor:

𝑬
𝐷 = 𝐷0 exp −
𝑘𝑇

2
Δ𝑆𝑚 + Δ𝑆𝑓
𝐷0 = 𝑓𝑛𝑐 𝜆 𝜈0 exp 𝑎𝑛𝑑 𝑬 = Δ𝐻𝑚 + Δ𝐻𝑓
𝑘

We have interchangeably used E and DH to express the energy barrier:

Δ𝐻 = Δ𝐻𝑚 + Δ𝐻𝑓

MSE 201 Ya-Hong Xie

 2.2.8 Diffusion: experimental measurements

Here we discuss one example of the experiments for determining DH (the Simmons-
Balluffi method) including decoupling DHm from DHf:

We begin by considering the dependence of nv on T:

𝑛𝑣 Δ𝐺𝑓 Δ𝑆𝑓 −Δ𝐻𝑓

= exp − = exp × exp
𝑛 𝑘𝑇 𝑘 𝑘𝑇

When we raise the temperature, the volume of a given metal specimen increases due to
two factors:

1. Thermal expansion, i.e. the increase in atom-atom spacing, a;

2. The increase in vacancy concentration, nv;

MSE 201 Ya-Hong Xie

 Diffusion: experimental measurements
Our objective is to measure nv(T). It can be shown:

Δ𝑛𝑣 Δ𝐿 Δ𝑎
=3 −
𝑛 𝐿 𝑎

Overall volume
expansion by By XRD
changing length
of a metal wire

Simmons-Balluffi experiment
of Al wire

Δ𝑛𝑣 0.76𝑒𝑉
= exp 2.4 exp −
𝑛 𝑘𝑇
MSE 201 Ya-Hong Xie
 Diffusion: experimental measurements

Next we need to determine DHm , the enthalpy barrier to self-diffusion. This can be
done through the “thermal expansion, quench plus resistance measurement” method
depicted by the figure.

MSE 201 Ya-Hong Xie

 Diffusion: experimental measurements

Primary components of the “thermal expansion, quench plus resistance measurement”

method:

1. Anneal at a high T;

2. Quench to freeze in vacancies;

3. Controlled ramping of T for annealing;

4. Vacancy diffusion with barrier DHm;

5. Extract DHm using equations on next slide;

The initial rise shown by the dashed line is due

mainly to increase phonon scattering, and
the decrease is due to the vacancies
reaching the surface of the wire;

MSE 201 Ya-Hong Xie

 Diffusion: experimental measurements

The “equation of motion” of vacancy diffusion during ramped anneal:

𝑡 𝑇 𝑇
2
𝑑𝑡 𝑑𝑡 Δ𝐻𝑚
𝑥 = න 4𝐷𝑑𝑡 = න 4𝐷 𝑑𝑇 = න 4𝐷0 exp − 𝑑𝑇
𝑑𝑇 𝑑𝑇 𝑘𝑇
0 0 0

Carry out the integration using Kissinger’s approximation, we get:

2
𝑑𝑡 𝑘𝑇 2 Δ𝐻𝑚
𝑥 = 4𝐷 exp −
𝑑𝑇 Δ𝐻𝑚 𝑘𝑇

Take loge on both sides, we get:

𝑥 2 Δ𝐻𝑚 1 𝑑𝑇 Δ𝐻𝑚
ln + ln 2 =−
: 4𝐷 𝑘 𝑇 𝑑𝑡 𝑘𝑇 Based on this, DHm
can be determined.

MSE 201 Ya-Hong Xie

 Diffusion: experimental measurements
Determining the entropy factor by assuming n0=1013 Hz by measuring D0:

Δ𝑆 Δ𝑆𝑚 + Δ𝑆𝑓 𝐷0
exp = exp =
𝑘 𝑘 𝑓𝜈0 𝜆2 𝑛𝑐

In view that diffusion leads to randomization thus increase in S, it can be argued

that the exponential should be larger than unity. As such, it is expected that
D0>fn0l2nc, with the values of the quantities on the right-hand side typically
known. For Aluminum:

𝐷0 > 0.78 × 1013 × 2.86 × 10−8 2

× 12 = 0.076𝑐𝑚2 /𝑠
:
In general, D0 in metals ranges between 0.1-1 cm2/s, therefore DS~k

MSE 201 Ya-Hong Xie

 Nonlinear Diffusion

• When the driving force is sufficiently large as for example could happen in nano-
scale materials, the assumption used in deriving linear diffusion equations is no
longer satisfactory. One needs to take into consideration of higher order terms.

• The nonlinear effect can be divided into “kinetic” and “thermodynamic” as will
be discussed in detail below.

• Theoretical foundation for describing nonlinear diffusion has been developed by

Cahn and Hilliard in the case of spinodal decomposition (with the precipitation
but not nucleation, i.e. no energy barrier of nano-scale phases), thin films or
superlattices.

MSE 201 Ya-Hong Xie

 2.3.1 Nonlinear Effect due to Kinetic Consideration
The unique situation due to large driving forces as shown in the figure where for a
non-negligible DC, Dx=Nl with N≤10 as shown in the figure below.

An alternative way of thinking about this situation is that the atomic flux or atomic
jumping frequency is no longer linearly dependent on the driving force with the inclusion
of higher order terms becoming necessary.

MSE 201 Ya-Hong Xie

 2.3.1 Nonlinear Effect due to Kinetic Consideration

We have:
𝜇2 = 𝑘𝑇 ln 𝐶2 Ω , 𝜇1 = 𝑘𝑇 ln 𝐶1 Ω
𝐶1
Δ𝜇 = 𝜇1 − 𝜇2 = 𝑘𝑇 ln where CiΩ=Xi
𝐶2

Thus:

Δ𝜇 𝑘𝑇 𝐶2
𝐹= = ln
Δ𝑥 𝑁𝜆 𝐶1

𝜆𝐹 ln 𝐶2 /𝐶1
= (2.37)
𝑘𝑇 𝑁

MSE 201 Ya-Hong Xie

 2.3.1 Nonlinear Effect due to Kinetic Consideration

For large driving force, we could have C2/C1 varies from 10-1000. Assuming N=10, the
approximation made below:

𝜆𝐹 𝜆𝐹
i.e. 𝑤ℎ𝑒𝑛 << 1, 𝑡ℎ𝑢𝑠 𝜈𝑛 ≈ 𝜈 is no longer valid
𝑘𝑇 𝑘𝑇

As such, higher order terms in Taylor expansion of the sinh function need to be included:

𝑥3 𝜆𝐹
sinh 𝑥 = 𝑥 + 𝑤ℎ𝑒𝑟𝑒 𝑥=
3! 2𝑘𝑇

𝜆𝐹
∴ 𝜈𝑛 = 2𝜈 sinh
2𝑘𝑇

3
𝜆𝜈 𝜈 𝜆
∴ 𝜈𝑛 ≈ 𝐹+ 𝐹3
𝑘𝑇 24 𝑘𝑇

MSE 201 Ya-Hong Xie

 2.3.1 Nonlinear Effect due to Kinetic Consideration

As such, the rate of successful jump, n is no longer proportional to the driving force, F.
If we take m=kTln(CW), we have from the beginning of chapter 2:

2 2
𝜕𝐶 1 𝜆 𝜕𝐶
𝐽 = 𝐶𝑉 = 𝐶𝜆𝜈𝑛 = −𝐷 1+ steady state
𝜕𝑥 24 𝐶 𝜕𝑥

This leads to the equation of continuity: 𝜕𝐶

− ∇⋅𝐽 =
𝜕𝑡

to become a 4th order equation: Transient behavior

2 2
𝜕𝐶 𝜕2𝐶 1 𝜆 𝜕𝐶
=𝐷 2 1+ 𝑤ℎ𝑒𝑟𝑒 𝐷 = 𝜈𝜆2
𝜕𝑡 𝜕𝑥 8 𝐶 𝜕𝑥

MSE 201 Ya-Hong Xie

 2.3.1 Nonlinear Effect due to Kinetic Consideration

• The above derivation is based on kinetic consideration, and thus termed kinetic
nonlinear equation;

• The use of the ideal solution expression of chemical potential in the above
derivation is a quick way for showing the idea, but is rigorously speaking not
valid in the case of large concentration gradients.

• Starting from the next slide, nonlinear effect will be considered from the angle of
the chemical potential, thus termed thermodynamic consideration.

MSE 201 Ya-Hong Xie

 2.3.2 Nonlinear Effect due to Thermodynamic Consideration

In the case of large concentration gradient, the chemical potential becomes not
only a function of the concentration, but of the concentration gradient. This
consideration has been treated in detail by Cahn and Hilliard, with the total free energy
being expressed as:

𝜕𝐶
𝐹෨ = න 𝑓 𝐶, 𝑑𝑉 where f is the free energy per unit volume
𝜕𝑥

It was shown by Cahn and Hilliard that for not too large a concentration gradient:
2
𝜕𝐶
𝑓 = 𝑓0 + 𝐾 where the dC/dx term has zero coefficient
𝜕𝑥

where f0 is the average free energy per unit volume being dependent on the
average concentration C0; K is the “gradient energy coefficient” with K=K(C).

MSE 201 Ya-Hong Xie

 2.3.2 Nonlinear Effect due to Thermodynamic Consideration
2
𝜕𝑓 𝜕𝐶
From: 𝜇≡ Noting: 𝑓 = 𝑓0 + 𝐾
𝜕𝐶 𝜕𝑥

2 2 2
𝜕𝑓 𝜕 𝜕𝐶 𝜕𝑓0 𝜕𝐾 𝜕𝐶 𝜕 𝜕𝐶 𝜕𝑥
𝜇= = 𝑓 +𝐾 𝐶 = − −𝐾 ⋅ =
𝜕𝐶 𝜕𝐶 0 𝜕𝑥 𝜕𝐶 𝜕𝐶 𝜕𝑥 𝜕𝑥 𝜕𝑥 𝜕𝐶

2
𝜕𝑓0 𝜕𝐾 𝜕𝐶 𝜕2𝐶
= − − 2𝐾 2
𝜕𝐶 𝜕𝐶 𝜕𝑥 𝜕𝑥

3
𝜕𝜇 𝜕 2 𝑓0 𝜕𝐶 𝜕 2 𝐾 𝜕𝐶 𝜕𝐾 𝜕𝐶 𝜕 2 𝐶 𝜕3𝐶
∴ 𝐽 = 𝐶𝑀𝐹 = 𝐶𝑀 − = −𝐶𝑀 − −4 − 2𝐾 3
𝜕𝑥 𝜕𝐶 2 𝜕𝑥 𝜕𝐶 2 𝜕𝑥 𝜕𝐶 𝜕𝑥 𝜕𝑥 2 𝜕𝑥

Note here the kinetic effect is not taken into consideration in using this equation.

MSE 201 Ya-Hong Xie

2.3.2 Nonlinear Effect due to Thermodynamic Consideration

In deriving the last equation, we used:

𝑑𝜇 𝜕𝜇 𝜕𝜇 𝜕𝐶 𝜕𝜇 𝜕𝐶𝑥 𝜕𝜇 𝜕𝐶𝑥𝑥
= + + +
𝑑𝑥 𝜕𝑥 𝜕𝐶 𝜕𝑥 𝜕𝐶𝑥 𝜕𝑥 𝜕𝐶𝑥𝑥 𝜕𝑥

𝜕𝐶 𝜕2𝐶
𝑤ℎ𝑒𝑟𝑒 𝜇 = 𝜇 𝐶, 𝐶𝑥 , 𝐶𝑥𝑥 , 𝐶𝑥 = ,𝐶 = was used
𝜕𝑥 𝑥𝑥 𝜕𝑥 2

MSE 201 Ya-Hong Xie

 2.3.2 Nonlinear Effect due to Thermodynamic Consideration

With this nonlinear thermodynamic effect taken into consideration, and following
Gauss theorem or flux divergence of section 2.2.4, the continuity equation becomes:

3
𝑑𝐶 𝜕 𝜕𝐶 1 𝜕 2 𝐾 𝜕𝐶 4 𝜕𝐾 𝜕𝐶 𝜕 2 𝐶 2𝐾 𝜕 3 𝐶
= ෩0
𝐷 − − " − "
𝑑𝑡 𝜕𝑥 𝜕𝑥 𝑓0" 𝜕𝐶 2 𝜕𝑥 𝑓0 𝜕𝐶 𝜕𝑥 𝜕𝑥 2
𝑓0 𝜕𝑥 3

𝜕𝐶 𝜕 𝜕𝐶 2𝐾 𝜕 3 𝐶
𝑶𝒓 = ෩0
𝐷 − if K is a constant independent of C;
𝜕𝑡 𝜕𝑥 𝜕𝑥 𝑓0" 𝜕𝑥 3

𝜕𝑓0 𝜕 2 𝑓0
where the newly introduced symbols are: 𝑓0′ = , 𝑓0" = 𝑒𝑡𝑐
𝜕𝐶 𝜕𝐶 2
෩0 = 𝐶𝑀𝑓0"
𝐷

and ∇•𝐽 involves total differentiation of m wrt x.

MSE 201 Ya-Hong Xie

 2.3.2 Nonlinear Effect due to Thermodynamic Consideration

If D0(tilda) is assumed to be neither a function of C nor its derivatives, the previous

equation becomes:

𝜕𝐶 𝜕2𝐶 𝐾𝑓0′′′ 𝜕𝐶 𝜕 3 𝐶 𝐾 𝜕4𝐶

∴ ෩0
=𝐷 ෩0
+ 2𝐷 ෩0
− 2𝐷
2 2 3
𝜕𝑡 𝜕𝑥 𝑓 " 𝜕𝑥 𝜕𝑥 𝑓0" 𝜕𝑥 4
0

If further the middle term of the above equation could be ignored as justified by
Cahn (details in chapter 5), we can write:

𝜕𝐶 ෩0 𝜕 4 𝐶
𝜕 2 𝐶 2𝐾 𝐷
෩0
=𝐷 − (2.47)
𝜕𝑡 𝜕𝑥 2 𝑓0" 𝜕𝑥 4

MSE 201 Ya-Hong Xie

 Spinodal vs Chemical Diffusion: stability & nonlinearity

• Spinodal diffusion:
✓ A homogeneous single-phase state is unstable;
✓ The decomposition takes place via atom diffusion;
✓ There is no nucleation of a second phase (no energy barrier);
✓ The inter-diffusion coefficient is negative;
✓ The nonlinear effect is most significant towards the end of the process.
• Chemical (regular) diffusion:
✓ The existence of more than one phase is unstable;
✓ The process of homogenizing takes place via atom diffusion;
✓ There is no nucleation barrier to homogenization;
✓ The inter-diffusion coefficient is positive;
✓ The nonlinear effect is most significant at the beginning of the process.

MSE 201 Ya-Hong Xie

2.3.3 Combining Thermodynamic and Kinetic Nonlinear Effects

When thermodynamic and kinetic nonlinear effects are considered together, we have:

1 𝜆𝐹 2 𝜆2ν
𝐽 = 𝐶𝑀𝐹 1 + 𝑤ℎ𝑒𝑟𝑒 𝑀= (2.48)
24 𝑘𝑇 𝑘𝑇

and:
𝜕𝐶 2𝐾 (𝜕3 𝐶/𝜕𝑥3)
𝐹= −𝑓0" 1− (2.49)
𝜕𝑥 𝑓0" 𝜕𝐶
( )
𝜕𝑥

By substituting Eq. (2.49) into Eq. (2.48), we have:

2
2𝐾 𝐶𝑥𝑥𝑥 1 𝜆𝑓0" 2𝐾 𝐶𝑥𝑥𝑥 2
𝐽 = 𝐶𝑀𝑓0" 𝐶𝑥(1 − ) 1+ 𝐶 1− (2.50)
𝑓0" 𝐶𝑥 24 𝑘𝑇 𝑥 𝑓0" 𝐶𝑥

MSE 201 Ya-Hong Xie

2.3.3 Combining Thermodynamic and Kinetic Nonlinear Effects

In Eq. (2.48) and Eq. (2.49), both of the second terms tends to be small. While
they themselves cannot be ignored, their product can certainly be ignored. As a result:

2
1 𝜆𝑓0" 2𝐾
෩0 𝐶𝑥 +
𝐽 = −𝐷 𝐶𝑥 3 − 𝐶 (2.51) flux
24 𝑘𝑇 𝑓0" 𝑥𝑥𝑥

2
𝜕𝐶 𝜕2 𝐶 1 𝜆𝑓0" 𝐾
= ෩
𝐷0 2 [1 + ෩0
𝐶 𝑥 2] − 2 𝐷 𝐶 (2.54) diffusion
𝜕𝑡 𝜕𝑥 8 𝑘𝑇 𝑓0" 𝑥𝑥𝑥𝑥

This term appears to be due to kinetic effects via comparison to Eq. (2.47)

෩0 = 𝐶𝑀𝑓0" was assumed to be independent of C and its

where for simplicity 𝐷
derivatives; furthermore, we assumed that 𝑓0 is a parabolic function of C, thus 𝑓0′′′ =0.

MSE 201 Ya-Hong Xie