Gupta D. (Ed.). Diffusion Processes in Advanced Technological Materials

Подождите немного. Документ загружается.

various types have been placed as APDBs between two reacting films.

During the early years of the microelectronics era (the 1960s and early

1970s), gold-beam leaded interconnection technology led to very reliable

devices and circuits.

[3]

In this technology, the interconnections to the out-

side world were made by using Au, Pd, or sometimes a bilayer of Ti and

Pd between the Au and Si contacts. Contacts to Si were made using a sili-

cide of Pd or Pt, although some work was reported on the use of electro-

less- plated Ni.

[4]

Au could not be used directly on Si because of (1) its very

high diffusivity in Si, (2) its inability to bond with SiO

, which is always

present as a native oxide layer on the surface of silicon, and (3) the exis-

tence of a low temperature (370°C) eutectic in the Au-Si system.

[5]

Adirect

extension of the gold-beam leaded technology, when Al was introduced as

the contact to Si, was not possible because of the metallurgical interactions

between Al and Au and between Al and Pd. Such interactions, occurring at

relatively low temperatures, made the metallization mechanically and elec-

trically unstable. Asuitable barrier was needed. Ti/Pt bilayer and Ti/TiN/Pt

trilayer APDBs were found to satisfy this need and to be reliable under

actual fabrication and use conditions.

[6]

Note that a study of the thermody-

namics of the metal system (individual metals and their reaction products),

appropriate phase diagrams, and interdiffusion was essential for making

the right choice of the APDB between Al and Au. The need for an effective

Ti-Pt diffusion barrier between Al and Au, with a thickness of at least 200

nm of each individual metal, was recognized.

[6]

To investigate their relative

effectiveness as diffusion barriers, several two-metal, three-metal, four-

metal, and five-metal layers were fabricated and tested for interdiffusion,

failures, and preventing a reaction between Al and Au.

Interaction between two materials is a result of fast interdiffusion. In

thin films, a large number of fast diffusion mechanisms are operating,

leading to an obvious suggestion to use high-melting-point metals/alloys.

For diffusion through a lattice (so-called lattice diffusion), the self-diffusion

coefficient D

self

is indirectly related to the melting point (T

) of the mate-

rial by an empirical relation between the activation energy (Q

self

) of self-

diffusion and the melting point:

self

 34 T

, (2)

where T

is given in degrees Kelvin and Q

self

is given in calories per mole.

In most cases, D and Q are related through the Arrhenius relationship:

D  D

exp(QRT), (3)

where R is the gas constant. Thus the higher the melting point, the higher

the activation energy and the lower the self-diffusion coefficient. However,

242 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Ch_05.qxd 11/29/04 6:19 PM Page 242

DIFFUSION BARRIERS IN SEMICONDUCTOR DEVICES, MURARKA 243

in thin films, which are generally deposited by vapor deposition techniques,

the diffusion is dominated by the grain boundaries. For self-diffusion

through grain boundaries, we can approximate the activation energy to be

about half that for the lattice diffusion given by Eq. (2). Thus, most thin-

film interactions are the result of grain boundary diffusion and other fast

diffusion paths, such as dislocations. If these fast diffusion paths can some-

how be blocked (or stuffed

[2]

), the interaction rate can be significantly

reduced. These relationships provided the concept of using refractory met-

als with stuffed grain boundaries for application as APDB thin films and

led to the use of W, Ta, Nb, Mo, Hf, Zr, V, Cr, and Ti metals. The electri-

cally conducting nitrides, carbides, and borides of these metals have very

high melting points. In early applications, these materials were suggested

as APDBs, and some were used, not only because they have high melting

points, but because the presence of a slight excess of the nitrogen, carbon,

and boron, respectively, in these materials stuffs the fast diffusion paths.

With decreasing device size in microelectronics (which caused

decreasing junction widths and depths, leading to the concern about the

use of Al contacts on Si) of the late 1970s and early 1980s, a diffusion bar-

rier was needed between Al and Si. The most commonly suggested barri-

ers (which have one common factor common: the high melting point)

have been the following:

1. Silicides

2. Refractory nitrides, carbides, or borides

3. Ti-W alloy

4. Refractory metal-noble (or near noble) metal alloy

5. A refractory metal alone

The most commonly used barriers between Al and Si have been the Ti-W

alloy, TiSi

TiN, TiSi

WTiTiN, CoSi

TiN, CoSi

TiTiN, W, and

WTiN. In most advanced Si integrated circuits, a CoSi

TiTiNW thin-

film structure is used in the Al or Cu interconnect technology.Unlike the

Al interconnects on SiO

, which do not require any APDB material due to

their natural adhesion property without interdiffusion, APDB material is

needed between Cu and SiO

. This has led to a renewed search. Presently,

a nitride of Ta (TaNx) is being used between SiO

and Cu, and between W

(in the contact windows) and Cu.

[7]

The barrier thickness is in the range of

5 to 30 nm, and the resistivity is high (200 to 300 mΩ cm). For future

applications, either the thickness has to be reduced by an order of

magnitude or the resistivity of the barrier has to be similar to that of Cu.

These requirements have led to a flurry of activities in self-forming

barriers/adhesion promoters

[8]

and in a newer group of materials called

self-assembled molecules (SAMs)

[9]

in which a monolayer of molecules,

Ch_05.qxd 11/29/04 6:19 PM Page 243

generally about 1 nm thick, is formed on a given surface and could act as

an APDB layer between the ILD and Cu. In the future, a lower dielectric

constant (k) interlayer dielectric (ILD) film,

[10]

which may be a polymer,

fluorinated/carbonated SiO

, or a porous material, may replace pure SiO

as an ILD. Such replacements may also place new constraints on the prop-

erties of APDB materials. These new concepts and their applications are

reviewed in Sec. 5.6.

5.3 Brief Review of Diffusion and the

Influencing Material Factors

5.3.1 Diffusion in the Lattice and Grain Boundaries

Equation (1) simply states that the rate of transfer of a diffusing

species across a unit area perpendicular to the x direction is proportional

to the concentration gradient in that direction. If we consider diffusion

across an element of volume and use the law of conservation of matter, we

reach a conclusion that the rate at which the concentration changes in this

volume must equal the local decrease of the diffusion flux, leading to the

one-dimensional Fick’s second law of diffusion:

∂c(x, t)∂t  ∂{D [∂c(x, t)∂x]}∂x. (3)

For concentration-independent D, Eq. (3) simplifies to:

∂c(x, t)∂t  D [∂

c(x, t)∂x

]. (4)

To determine D, we must carefully determine the concentration of the

diffusing species as a function of the depth in the sample for a given dif-

fusion treatment (that is, time, temperature, and ambient). Equation (3)

or (4) is then solved using the experimentally placed boundary condi-

tions. Avariety of solutions fitting various boundary conditions are avail-

able.

[11]

Two of the most commonly used solutions of Eq. (4) are Gaussian

and error-function distributions given, respectively, by the following

equations:

c(x, t)  M√(pDt)  exp(x

4Dt) (5)

and

c(x, t)  c

 erfc[x2√(Dt)]. (6)

244 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Ch_05.qxd 11/29/04 6:19 PM Page 244

DIFFUSION BARRIERS IN SEMICONDUCTOR DEVICES, MURARKA 245

Here M is the total amount of the diffusing substance that is constant, c

is the time-independent surface concentration of the diffusing species,

and erfc is the complimentary error function related to the error function

by:

erfc (z)  1  erf (z). (7)

Note that the solutions given by Eq. (5) or (6) are for concentration-

independent D and are useful in most cases of interest. For concentration-

dependent D, several solutions (and approximations) of Eq. (3) have been

examined and presented.

[11]

To obtain a reasonable estimate of the penetration by a diffusing

species, we can calculate the diffusion depth, x

, for a diffusion time t and

at a temperature for which a value of D has been obtained. Theoretically,

should be infinite, since x

will be determined by the condition c  0

at t  0. For most practical purposes, x

is defined as:

 4Dt. (8)

Use of this definition to determine a typical diffusion depth simply indi-

cates that cc

has a predetermined value: for Gaussian solution (Eq. 5),

cc

 1/e  0.37; for the error-function solution (Eq. 6,) cc

 0.1573.

In this context, we also define the transport velocity, V(cm/sec),

across a material of thickness d, which is given by:

V  Dd. (9)

Thus, to minimize the mass transport, Vshould be reduced by minimizing

D across the APDB and/or by increasing the thickness, d, of the APDB.

The latter is difficult to achieve, leaving the choice of selecting a very thin

film of APDB material with the lowest possible D of the impurity.

To use Eqs. (5) and (6) rigorously, diffusion measurements are gener-

ally made in the lattice of ideal solids that contain only point defects in

thermal equilibrium that provide energetically favored migration paths. In

real solids and especially in thin films, their microstructures become very

important and a large number of fast-diffusion paths, such as grain bound-

aries, surfaces, and dislocations, exist and contribute to enhancing the

low-temperature diffusion coefficient by several orders of magnitude. At

diffusion temperatures of less than half the melting point (in degrees

Kelvin), surface, grain boundary, and dislocation diffusion coefficients are

higher than the lattice diffusion coefficient by about 8, 7, and 5 orders of

magnitude, respectively.

[12]

We can derive relationships between J

and J

that represent grain boundary and lattice fluxes, respectively, when diffusion

Ch_05.qxd 11/29/04 6:19 PM Page 245

from the grain boundary into the adjoining grains is permitted:

J

 2dD

D

. (10)

and D

are diffusion coefficients in the grain boundary and in the lat-

tice, respectively, and d is the grain boundary width. This relationship is

obtained by considering a situation in which the concentration on the front

and back surfaces of the slab of the polycrystalline material is maintained

constant, and a steady state is assumed because all concentration gradients

parallel to the plane of the slab are eliminated. Note that in these consid-

erations, grain boundaries offer unrestricted diffusion. Any segregation of

the diffusing species and others present in the films will seriously affect

both fluxes. Gupta et al.

[13]

have explained the solute (impurity) effect in

changing the diffusion of solvent atoms (for example, Al in an Al-Cu

alloy) phenomenologically. They considered interactions of the solute

(impurity) atoms with defects in the lattice, in the grain boundaries and

the equilibrium solute adsorption at the grain boundaries. The solute effect

on the grain boundary diffusivity has been shown to depend not only on

the associated solute-binding free-energy difference (∆G

) between the

grain boundary and the lattice sites, but also on the changes in the lattice

diffusion, as given by:

[(D

gbp

D

gba

)  (D

D

]

12

 1  c

exp(∆G

RT), (11)

where subscripts p and a refer to pure and alloyed metal, respectively, and

is the solute concentration in the lattice. For c

 1, Eq. (11) predicts

diffusion enhancement at high temperatures and retardation at low

temperatures.

5.3.2 Interdiffusion Between Two or More

Materials in Contact

There are no universal models of diffusion/interdiffusion in thin films

of various materials in contact with each other. Balluffi and Blakeley

[14]

have examined the complexities in the details of diffusion processes in

thin films and point out the following:

“Diffusion in thin films may be expected to have special characteris-

tics for a number of reasons, which include the following: (1) thin films

are invariably diffused at relatively low temperatures because of their

poor thermal stability; (2) all volume elements are in close proximity to

either a free surface or interphase boundary of some kind; (3) films gen-

erally contain high densities of low-temperature short circuits for diffusion

246 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Ch_05.qxd 11/29/04 6:19 PM Page 246

DIFFUSION BARRIERS IN SEMICONDUCTOR DEVICES, MURARKA 247

such as grain boundaries and dislocations; (4) large biaxial stresses in the

plane of the film are often present; (5) relatively high concentration of

uncontrolled impurities may be present as a result of special fabrication or

diffusion conditions; (6) disordered or metastable structures may be present;

(7) diffusion often occurs over short distances under the influence of large

concentration or electrostatic potential gradients; and (8) steep-chemical

gradients and low temperature may affect the possible maintenance of

local equilibrium at phase boundaries during multiphase diffusion.”

It is then expected that at the typical post-metal-anneal temperatures

of 300 to 450°C, grain boundary diffusion will dominate in films of the

high-melting-point metals of interest, that is, Ti, Ta, Pt, W, metallic sili-

cides, and the refractory nitrides, carbides, and borides. On the other hand,

in metals such as Al and Mg, lattice diffusion will dominate at the same

temperatures. Diffusion behavior in copper is somewhere in between. In

developing an APDB film, our goal is therefore to eliminate or minimize

the interaction or reaction between materials of interest and the APDB

film. The reactivity can generally be estimated by evaluating the interdif-

fusion coefficients in thin films. Interdiffusion coefficients are strongly

related to the microstructure of these films; the mutual solid solubilities;

the intermetallic or compound formation, as suggested by the phase dia-

grams; the free energies or heats of fomation of these intermetallics and

compounds; and the temperature. For the devices/circuits in use, the ILDs

sandwiched between two interconnect layers are exposed to very high

electric fields (of the order of 1 MV/cm), which causes diffusion of metal

ions into the ILD. Thus, an APDB material is required between the metal

and the ILD.

The interdiffusion coefficient (D

) reflects the average overall move-

ment of both constituents, say, i and j (representing two solids in contact),

diffusing in a concentration gradient. For an ideal case of a binary sys-

tem, D

is given in terms of the individual diffusion coefficients, D

and

, and their respective concentrations, c

and c

, in the diffused alloy at

a depth x:

(x)  D

(x)  D

(x). (12)

is then used to replace D in Eq. (3) or (4). In thin-film interactions,

invariably, D

is obtained even when the binary system does not form the

ideal solid solution. Also in such cases, because of the lack of crystallinity,

the experimentally determined D

is an averaged grain boundary interdif-

fusion coeficient, at best. For the interaction between an APDB film and

another film in contact, D

must be negligibly small or the thickness of

the barrier film large so that the reaction does not reach the other

surface/interface.

Ch_05.qxd 11/29/04 6:19 PM Page 247

5.3.3 Role of Material Properties

Controlled Microstructure

Barrett et al. have stated:

[15]

“Practically all the useful properties of

materials are strongly dependent on their internal structure,” and “for a

material of given chemical composition, the internal structure is not con-

stant.” The effect of the internal structure or microstructure on the bulk

properties of materials is known. The microstructure of thin-film inter-

connections plays a very important role in determining two important

reliability factors: (1) the reactivity with the surroundings, and (2) the

electromigration and stress voiding.

As mentioned in Sec. 5.3.1, real materials are used for metallization

applications such as thin films of copper. Information on the lattice

parameter, density, and so forth for Cu can be readily found in handbooks.

All defects are extremely important in thin films. Chemical reactivity,

atomic diffusion, and electrical and mechanical properties are affected by

the presence of defects. Also, the properties of the materials are not

isotropic and depend on a specific crystallographic direction under con-

sideration. For example, in the face-centered cubic Al or Cu, (111) planes

have the highest density of atoms with low energy, and therefore behave

differently than (100) planes. For bulk polycrystalline metal with random

texture, isotropic behavior is obtained when the individual cystallites are

oriented throughout space with equal directional probability. The effective

macroscopic properties can be calculated by considering the directionally

dependent values averaged over all orientations in space.

For bulk material, especially the metals, the practical ways to con-

trol the microstructures to yield the desired and useful properties are

well known to metallurgists and materials scientists. The concepts are

also important for thin films to be used as APDB, but are difficult, if not

impossible, to be realized. Preparation of the substrate surfaces, surface

forces, and deposition methods, and the attendant parameters, yield a

microstructure variability that is not well understood. It is well known

that the substrate-film interface, the film surface (which may be covered

by a native oxide or other compounds), the impurities that segregate at

defect sites, and the grain boundaries are effective barriers to the motion

of dislocations, thus affecting both the film strength and film

microstructure. In Al and Cu thin films, texturing leads to deviation

from the isotropic polycrystalline thin-film case. Texture is defined as

the tendency of the individual crystallites of a material to acquire a pre-

ferred crystallographic orientation. The texture affects materials properties,

and thus affects performance. For example, the texture of electroless

copper film has been shown to affect the resulting oxidation behavior.

[16]

248 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Ch_05.qxd 11/29/04 6:19 PM Page 248

DIFFUSION BARRIERS IN SEMICONDUCTOR DEVICES, MURARKA 249

The correlation of room-temperature stress with texture and the subse-

quent influence on film resistivity have also been noted.

[17]

A recent

review discussed the observed texture responses of copper thin films

deposited by a variety of techniques.

[18]

In general, (111), (200), and

random components of texture dominate the response, although in

some cases, (220) and (520) texture components have been observed.

From surface energy considerations, the close-packed (111) oriented

grains should be favored. Frequent occurrence of an extensive fraction

of the (200) texture, which has been found the most stable, is not well

understood.

Texturing and microstructure are affected by thermal treatments, after

or during deposition, and are influenced considerably by the annealing

ambient. In certain cases, thermal annealing of copper films is found to

induce the formation of giant grains.

[19, 20]

The strain energy has been sus-

pected of playing a role in this regard.

Impurities significantly influence the microstructural evolution due to

(1) diffusion in and out of grain boundaries, (2) interaction with the host

metal, and (3) segregation in grain boundaries and dislocations, and at

surfaces and interfaces. Impurity fluxes in the grain boundaries have con-

tributed to grain boundary motion

[21]

and diffusion-induced recrystalliza-

tion,

[22]

and thus to grain growth. On the other hand, impurity segregation

at the grain boundaries and other short-circuiting paths such as disloca-

tions, surfaces, and interfaces is known to suppress the diffusion in or on

the material. Solute segregation at the grain boundaries and interfaces is

discussed later in this section. The lower the solid solubility is in the grain,

the higher the concentration is in the grain boundary.

[23]

For example, the

segregation of Cu in the grain boundaries of Al, either as metallic Cu or

as a Cu-Al alloy, leads to a retardation of the Al migration caused by the

imposed electric field (that is, electromigration is retarded, leading to

improved mean time to failure). Grain boundary stuffing has been known

to produce effective thin-film diffusion barriers.

[2]

The effect of Mg or Al

added to Cu, in concentrations less than the solid solubility limit, on the

microstructure has been investigated to reveal the metallurgical variables

of the APDB effectiveness of these alloys used between Cu and SiO

[24, 25]

The effectiveness of Mg or Al is attributed to their thermodynamically

favorable abilities to reduce SiO

, leading to the formation of an interfa-

cial layer, between Cu and SiO

, that acts as an APDB layer. The

microstructures of annealed Cu, Cu-Al, and Cu-Mg structures, however,

show very different behaviors. Pure copper has moderate to heavy twin-

ning with relatively uniform microstructure. Cu-Al has a clear bimodal

distribution of small (∼25 nm) and large (∼160 nm) grains and heavy twin-

ning. Cu-Mg, like pure Cu, shows a bimodal microstructure with consid-

erably less twinning.

Ch_05.qxd 11/29/04 6:19 PM Page 249

Figure 5.1 compares the (111) and (200) x-ray diffraction peak inte-

grated intensities of three films as a function of annealing temperature. Cu

and Cu-Al films have a preferred texture [with (111) to (200) ratio in

annealed films as high as 15] that agrees with the observed twinning typ-

ical of (111) orientations. Cu-Mg shows a complete reversal of preferred

texture, with (200) stronger than (111) [a (111) to (200) ratio of 0.04 after

400°C and 0.01 after 600°C anneal]. Temperature also has a large influ-

ence on the growth of (200) texture compared to that of (111). It is appar-

ent that there is a remarkable difference in the behavior of Al and Mg

alloying elements in controlling the microstructure of Cu films, especially

when the added impurity concentrations are below the solid solubility

limit. These differences could have a large impact on the diffusion in the

metal and thus on the electromigration behavior.

Note that in the case of Al metal, the measured mean time to failure

for the electromigration was found to increase with the increasing grain

size and the degree of (111)-preferred orientation; decreasing the spread

of grain size distribution is beneficial.

[26]

Electron microscopic studies of

the narrow lines have shown that the grain structure takes on a so-called

bamboo appearance, with grain boundaries generally running perpendicu-

lar to the direction of the current flow (that is, along the interconnection

length).

[27]

Similarly, the superiority of the Al-Cu alloys

[28]

and of e-gun

evaporated metals or alloys is associated with the resulting preferred (111)

texture and uniform grain size distribution.

Mutual Solid Solubilities and the Phase Diagrams

Solid solubilities of i into j and of j into i play an important role in

determining the outcome of the reaction between i and j. If the solid sol-

ubilities are large, the interdiffusion leads to a solid solution prior to for-

mation of any compound. For example, titanium can dissolve significant

amounts of oxygen prior to the oxide formation. The advantage of such a

system is that no new phase appears in early periods of reaction. On the

other hand, when the mutual solid solubilities are low, compound forma-

tion occurs immediately. For example, oxides of W form readily because

W does not dissolve any significant amount of oxygen. In the absence of

a compound formation, second-phase (i or j) precipitation has been seen

on cooling from high temperatures where solid solubilities were higher.

For example, aluminum dissolves larger amounts of silicon at 450°C, the

typical metal anneal temperature. On cooling to room temperature, silicon

precipitates out as the second phase in aluminum film and at the silicon-

aluminum interface. Solid solubility does play a role in the second-phase

formation, but it does not play a direct role in influencing the diffusion

250 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Ch_05.qxd 11/29/04 6:19 PM Page 250

DIFFUSION BARRIERS IN SEMICONDUCTOR DEVICES, MURARKA 251

Figure 5.1 Area under x-ray diffraction peak vs.annealing temperature of Cu and

Cu alloys. All annealing times were 30 minutes except annealing done at 600°C,

which lasted 120 minutes.

Ch_05.qxd 11/29/04 6:19 PM Page 251