Growth kinetics of planar binary diffusion

couples: ’’Thin-film case’’ versus ’’bulk cases’’

Cite as: Journal of Applied Physics 53, 3252 (1982); https://doi.org/10.1063/1.331028
Published Online: 04 June 1998

U. Gösele, and K. N. Tu

ARTICLES YOU MAY BE INTERESTED IN

‘‘Critical thickness’’ of amorphous phase formation in binary diffusion couples

Journal of Applied Physics 66, 2619 (1989); https://doi.org/10.1063/1.344229

First phase nucleation in silicon–transition-metal planar interfaces

Applied Physics Letters 28, 624 (1976); https://doi.org/10.1063/1.88590

General Relationship for the Thermal Oxidation of Silicon

Journal of Applied Physics 36, 3770 (1965); https://doi.org/10.1063/1.1713945

Journal of Applied Physics 53, 3252 (1982); https://doi.org/10.1063/1.331028 53, 3252

© 1982 American Institute of Physics.

Growth kinetics of planar binary diffusion couples: "Thin-film case" versus
"Bulk cases"
u. Gbsele and K. N. Tu
a)

IBM Thomas J. Watson Research Center, Yorktown Heights, New York /0598

(Received 5 October 1981; accepted for publication 4 December 1981)

It is proposed that inte~facial reaction barriers in binary A / B diffusion couples lead to the
absence of phases predIcted by the equilibrium phase diagram, provided that the diffusion zones
are sufficiently thin (thin-film case). With increasing thickness of the diffusion zones the
influence of interfacial reaction barriers decreases and the simultaneous existence of diffusion-
controlled growth of all equilibrium phases is expected (bulk case). Selective growth of the first
and second phases and the effect of impurities are discussed with the influence of interfacial
reaction barriers and with references to the known cases of silicide formation.
PACS numbers: 81.30.-t, 66.30.Dn

I. INTRODUCTION and specifically pointed out very recently by Geguzin et at. 14

It is well established experimentally that in planar bina- In Sec, II the reader is reminded of known cases of the
ry thin-film diffusion couples not all of the compound phases growth kinetics of a single compound layer in the presence of
predicted by the equilibrium phase diagram have been ob- interfacial reaction barriers and the growth kinetics of multi-
served to be present simultaneously. J,2 In the case of silicides phase layer structures without interfacial reaction barriers.
it is even the common case that only one compound phase In Sec, III we will study the influence of interfacial reaction
occurs between the components silicon and metal. For ex- barriers on the kinetics of a specific model system consisting
ample, nickel films deposited on silicon (or vice versa) form of two compound phases between saturated phases which
an intermediate layer of Ni 2Si at temperatures above supply the compound components A and B. In this context,
- 200 ·C with no indication of the presence of the otherequi- and by the discussion in Sec. IV, the difference between the
librium phases Ni 3Si, Ni 3 Si 2 , Ni j Si2 , NiSi, or NiSi 2 as long as bulk case and thin-film case will be made clear-at least for
both unreacted nickel and silicon are still available. 3 .4 The the model system described. In Sec. IV we apply our model
thicknesses of compound layers in thin-film diffusion cou- system to the sequence of phase formation and the growth
ples, mainly investigated by ion backscattering techniques, kinetics of silicon-metal thin-film diffusion couples. Finally,
are typically on the order of a few hundred to several thou- a summary is given in Sec. V.
sand Angstroms. For much thicker diffusion couples, say
above 10 ",m, usually most or all of the compound phases II. BRIEF REVIEW OF PREVIOUS THEORIES
predicted by the phase diagram have been found. We will A. Influence of interfacial reaction barriers on the
term the latter case the bulk case throughout this paper in growth kinetics of a single compound layer
contrast to the thin-film case where all but one of the predict-
The growth kinetics of a single compound layer is deter-
ed equilibrium phases are clearly missing.
mined by a combination of two types of processes, namely by
According to available theories of multi phase diffusion-
(i) the diffusion of matter across the compound layer where
controlled layer growth 5- 9 all equilibrium phases should be
the diffusion flux slows down with increasing layer thick-
present. Various reasons have been given as to why not all of
ness, and by (ii) the rearrangement of the atoms at the inter-
them may be observed in thin-film diffusion couples. It is
faces required for the growth of the compound layer which
argued that either all equilibrium phases are in fact present
may involve a reaction barrier. If the diffusion process is rate
but some of them cannot be observed because oflimited ex-
limiting and controls the growth, the corresponding kinetics
perimental resolution,5 or that some of the phases could not
is termed diffusion controlled and the layer thickness in-
nucleate I (J but would certainly grow if they could overcome a
creases proportional to the square root of time t. If the inter-
nucleation barrier. It has also been claimed that the forma-
facial reaction barriers control the kinetics, it is termed inter-
tion of the missing compound phases is thermodynamically
face controlled (or boundary or reaction controlled) and the
impossible at the generally lower diffusion temperatures in
layer thickness increases linearly with time. Although var-
thin-film compared to bulk diffusion couples. II In the pre-
ious treatments for the influence of interface reaction bar-
sent paper we investigate the influence of interfacial reaction
riers on the growth kinetics of a compound phase are avail-
barriers on the growth kinetics of multi phase layer struc-
able in the literature,14-17 none of them has taken into
tures. We find that certain compound phases may indeed be
account all the specific features which are required to facili-
kinetically unstable due to interfacial reaction barriers in the
thin-film case as suggested by a number of authors 3 .4,J2,J3 tate the general treatment of the model of a double com-
pound layer diffusion system in Sec. III. Therefore, we re-
derive the growth kinetics of a single compound layer in
alO nleave of absence from the Max-Planck-Institut fur Metallfors-
some detail here as a prerequisite of Sec. III.
chung, Stuttgart, Federal Republic of Germany, We consider the following simple situation illustrated

3252 J, AppL Phys. 53(4), April 1982 0021/8979/82/043252-09$02,40 © 1 982 American Institute of Physics 3252
 time-independent equilibrium value .d C pq = C'Pi. - C p~ by
ceq
I
I
I
:
combining Eqs. (4) and (5) with Eq, (3), The result is
a~: I
CA 1-----;:..:.:.; ~Without} interface reaction
(6)
With

~
barriers
eq I where K";is an effective interfacial reaction barrier for the
/C~a : ApB layer composed of the two interfacial reaction barriers
I

I, '-.......

C~a
"<

'-
C~y :
'\1
I
I
Kpa and Kp,.. according to
l/Kpff = l/KPa + l/Kp,... (7)
'-,- The change of the layer thickness xp = x p,.. - xaP with time
ceq is given by
Aa B A~B fJy AyB
ceq dxfJldt= [l/(C~'b -Cpa )+ l/(Cp ,.. -C~'b)]J;;.
x~ .,
x
y~

x
We may further simplify Eq. (7) by assuming a narrow range
of homogeneity for the compound ApB, so that in the com-
(8)

xa~ /h position factor in the square brackets in Eq. (8) CfJa and Cp,..
FIG. I. Schematic of concentration profile of A atoms across an may be approximated by the same value /31[no( 1 + /3 I],
AaB / ApB / AyB diffusion couple with or without interface reaction where no is the volume per A or B atom, for simplicity as-
barriers. The phases AaB, ApB, and AyB are assumed to be saturated sumed to be constant throughout the sample. The quantity /3
(a>/hyl· is the number of A atoms per B atom in the AfJB compound.
ExpressingJ;; by Eq. (6) we thus arrive at
dxfJldt = Gp.dCpqKpff1(1 + XfJKpffIDp), (9)
in Fig. 1: A compound layer ApB of thickness xp grows
where GfJ is a constant determined by the composition of the
between two saturated phasesAaB andA,..B where the sub-
three phases AaB, ApB, and A,..B via
scripts a >/3> r characterize the composition of the com-
pounds. If a and r are chosen as 00 and 0, respectively, AaB Gp =n()(1 2( 1
+/3) - - + - - . 1) (10)
and A,..B represent the pure phases A and B. In Fig. 1 the a-/3 /3-r
concentration profile of the component A in the absence of Again, a and r represent the number of A atoms per B atom
interfacial reaction barriers is indicated by a dashed line. in the A"B or A,..B compound, respectively. Equation (9) is
Then the concentration of A at the interfaces is given by its basically the same as already given by Deal and Grove '5 and
corresponding equilibrium values C!fa and C p~. In the pres- Farrel et al. '6 for one diffusing species and similar to that
ence of interfacial reaction barriers, the concentration gradi- given by Geguzin et al. '4 for a slightly different situa~ion.
ent across theApB layer is smaller and the actual concentra- The growth kinetics described by Eq. (9) are different below
tions at the interfaces are Cpa < C p;' and Cp,.. > Cp~. We and above a changeover thickness xb' defined by
assume that theApB layer may be characterized by a!1 essen-
tially constant chemical interdiffusion coefficient Dp. Fol- Xp* = D-PI KfJ'
eff (11)
lowing Kidson 5 we calculate the change of the position of the From Eq. (9) it follows that
two interfaces with time t by dXpldt-GfJ.dCpqKpff, for xp<x~, (12a)
(1) dXpldt-GfJ.dCpqDplx p , for xp>x~, (12b)
for the a/3 interface, where C ~ refers to the concentration of or in integrated form
A in the ApB layer, and by
xp 0::. t for xfJ <x~, (13a)
(Cp,.. - C~'b)dxp,../dt = - Dp(dC~/dx)p,.., (2)
and
for the /3r interface. Assuming a steady state of the diffusion 2
x p o::.t ' / forx p >xl (13b)
fluxes and compound formation only at the interfaces we
Equations (13a) and (13b) again show the well-known result
obtain for the diffusion fluxJ;; of A atoms in the ApB phase
that interface-controlled growth kinetics will always change
J;; = - D(3(dC~/dx)pa = - Dp(dC~/dx)(3,.. over to diffusion-controlled growth kinetics if the layer has
(3) grown to a sufficiently large thickness xp>x~.

This fluxJ;; may also be expressed in terms of the reaction B. Diffusion-controlled multi phase layer growth
constants Kpa and Kp,.. at the two interfaces according to
In considering the growth of a set of compound layers
J;; = KPa(C p;' - Cpa)' (4) with thicknesses Xu, x fJ ' x,.., etc., we find that so far all the
or theoretical treatments 5 - 9 assume explicitly or implicitly that
the material transport across the compound layers by diffu-
(5)
sion is rate limiting, which is certainly justified if all the lay-
Since both C(3a and CfJ ,.. depend on the layerthicknessxfJ and ers considered are sufficiently thick, namely
therefore on time it is convenient to express the actual con- Xa >x:, x(3 >x~, x,.. >x~, etc. These theoretical treatments
centration difference CfJ ,.. - CfJy in Eq. (3) in terms of the have two essential points in common: (A) All phases predict-

3253 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3253
ed by the equilibrium phase diagram should be present in an As in Sec. II the volume no per A or B atom is assumed to be
A IB diffusion couple; and (B) All the layer thicknesses constant in all phases. We introduce an effective interfacial
X a , x p , x y' etc. grow proportional to t 1/2, which means that reaction barrier K~ff for the A yB layer composed of two inter-
the ratios of the layer thicknesses, e.g., Xalxp or Xplxy re- facial reaction barriers Kyp and Kyb in an analogous way as
main constant in time at constant temperature. For details given for the ApB layer in Eq. (7). Applying the same con-
the reader is referred to the original literature. 5-8 cepts as in Sec. II for the shift of the interfaces we may calcu-
late the change of x/3 and Xy with time by
III. MODEL CALCULATIONS FOR TWO INTERMEDIATE
COMPOUND LAYERS dX/3ldt = Gp./J - G/3ylr, (15)

A. Basic assumptions and equations ( 16)

We investigate the influence of interfacial reaction bar- with the (positive) diffusion fluxes of A atoms in theApB and
riers in a multi phase system by considering as a model sys- theAyB layer given by
tem two intermediate compound layers of composition ApB iJ =L1C 8QK8ff/(1 +X/3K8ff ID(3)' (17)
and AyB between saturated phases AaB and AliB, where
a> {3 > y> 8 (see Fig. 2), similarly as has been done by Sha-
Ir = L1C~QK~ff/(1 + XyK~ffIDy). (18)
tynski et al. 8 for the diffusion-controlled case. In order to The various quantities with the subscript y have the analo-
avoid the complications introduced by a possible nucleation gous meaning for the AyB compound layer as the corre-
barrier for anyone of the layers,we assume that in our model sponding quantities with the subscript {3 for the A/3B com-
system the two compound layers ApB and A yB are already pound layer. The quantities G/3' G/3y' GY' and Gy/3 account as
present with given thicknesses xp and x y' respectively. Our in Eq. (9) for the change in composition at the interfaces.
procedure will be to check whether one of the layers will They maybe easily calculated as G/3 already given in Eq. (10):
shrink away completely under various kinetic conditions. If
this can actually take place, this phase will not form in a Gp
1{3 + {3 -1)y ,
= no( 1 + {3) 2(a _ (19)
diffusion experiment for the given kinetic conditions, which
are characterized by a set of kinetic parameters (thickness of
other compound layer, diffusivities, interfacial reaction bar- Gy = no(l + y)
2( 1y + y _1)
{3 _ 8 ' (20)
riers). We shall call it a kinetic or growth instability. In such a
way we can separate a kinetic instability induced by interfa- Gpy = Gyp = n o(1 + {3)(1 + y)/( {3 - y). (21)
cial reaction barriers from the case of phase suppression by a We mention explicitly that in our model system the fluxesiJ
nucleation barrier (which can be present in addition); a kine- and Ir are considered to be independent of each other and
tically unstable phase shrinks away (or does not form) even if may be described, e.g., at the interface between ApB and
it can nucleate. AyB by
We characterize our model system by chemical inter- (22)
diffusion coefficients Dp and_Dy in the ApB and AyB layer,
respectively, where, e.g., Dp may approximately be ex- and
pressed in the most simple case in terms of component diffu- (23)
sion coefficients D; and D: by The quantity K/3y characterizes the reaction barrier against
Dp =D;/({3+ 1) + {3Dg/({3 + 1). (14) the growth oftheA/3B layer at the expense oftheAyB layer,
and Ky/3 characterizes the reaction barrier against the growth
of the AyB layer at the expense of the A/3B layer. In princi-
ple, special situations at the interface may be envisaged
which may not be described consistently by the boundary
conditions (22) and (23). If, e.g., compound formation at the
interface were not possible at all, but diffusion across the
interface were easily accomplished, an additional parameter
I
I would be required to describe the direct transfer of atoms
I across the interface. In the following, we will use results
I
I based on the boundary conditions (22) and (23) although we
I
I are aware of their limitations in specific cases.
I In Eqs. (15) and (16), the change of xp and Xy with time
I
I in each case is composed of a positive and a negative term. 18
I I This structure for dXpldt and dXyldt is well known from
I Af3B I
I I the theory of diffusion-controlled growth of multilayer
I I
I-- xf3 ......II....- - x y - I structures. 5- 9 The basic difference between our formulation
of growth and that of diffusion-controlled growth consists in
x
the occurrence of interfacial reaction barriers which are de-
FIG. 2. Schematic of concentration profile of A atoms in the
AaBIApBIArBIA6B model diffusion couple dealt with in Sec. III scribed in terms of K8ff and K<';. For a fast reaction at the
(a>fl>r>8). interface, KIt and K~ff_OO and Eqs. (15)-(18) can be simpli-

3254 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3254
fied to the corresponding expressions for diffusion-con- C. Transport across both layers dlffuslon-controlled
trolled multilayer growth kinetics. We assume that the starting thicknesses xp and Xy of
the two compound layers in our model system are such that
B. Criteria for growth/shrinkage behavior
the transport across each compound layer is diffusion con-
The condition for growth oftheApB layer, dxp/ trolled. This means xp >x: and Xy >x~, where x: is the chan-
dt> 0, may be expressed in terms of the (positive) ratio r of geover thickness for the ApB layer given by Eq. (11) and x~
the diffusion fluxesn andtr, that for the AyB layer defined by
r= n/tr, (24)
(30)
as
Under these circumstances the flux ratio r is simplified to
(25)
The analogous condition for the growth of the AyB layer is (31)

r<Gy/Gyp = r2 , We suppose that interdiffusion experiments start with a

(26)
thickness ratio x y /xp so that either r> r2 (case a), or .
with r l < r < r2 (case b), or r < r] (case c). The results are schemati-
r 1 =(1 +y)(a-/3)/(1 +/3)(a-y), (27) cally indicated in Fig. 4. For r> r2 the ApB layer grows,
and whereas the AyB layer shrinks. Since ro:.xy/xp, this behav-
r2 = (1 + y)( /3 - 15)/(1 + /3)(y - D), (28) ior leads to a decreasing value of r until r is so small (r < r2 )
where that the A y B layer starts to grow simultaneously with the .
r2 >r 1 , (29)
ApB layer. Finally, a time-independent flux ratio r"q IS
reached where both layers grow proportional to V t. This
holds, since (a - /3 ) < (a - y) and (/3 - 15 ) > (y - 15 ). Con-
flux ratio r"q is reached independent of the actual starting
trary to the ratio r, which may depend on xp and Xy and
ratios rand Xy/x/3 and is the same for cases (a-c). The diffu-
therefore may change with time, the ratios r l and rz depend
sion system is self-regulating always leading to the proper
on the compositions of the compounds only. Since r2 > r l , in
ratio r"q. This kinetic behavior reflects the predictions of all
principle a combination of parameters x p , x Y ' D p , D y' Kpff,
theories based on pure diffusion-controlled transport across
and K"yff can exist for which both compound layers grow si-
multaneously. If the flux ratio r is not in the range between r l
the compound layers. On the basis of such theories s- 9 x:
and
x* are equal to zero and no phas'e present in the equilibrium
and r2 , one of the layers will shrink, whereas the other one
phase diagram can shrink away because the diffusion flux
grows. The different growth and shrinkage regimes are sche-
across theApB layer, e.g., is proportional to l/x/3 and math-
matically indicated in Fig. 3. In the following subsections the
ematically xp~o would lead to an infinitely high diffusion
kinetic behavior of our model system for various specific
flux. This high diffusion flux, in turn leads to the growth of
parameter combinations is dealt with in more detail.

Qt
)(
......
)0..
)( r2
~ req
'-
0
II
+=
0
'-
f,
-
o
o
'-
)(
:::J
:;:

-
)(
:::J

FIG. 4. Schematic of growth/shrinkage behavior of the model diffu-

FIG. 3. Schematic of various growth and shrinkage regimes for the dif- sion system of Fig. 2 in dependence of time, starting diffusion with
fusion couple depicted in Fig. 2 in dependence of the ratio r of the dif- various thickness ratios xr/xp at time to' The transport across both
fusion fluxes of A atoms in the ApB layer (/p) and in the ArB layer layers, ApB and AyB is assumed to be diffusion controlled, xp>xb and
(/rl· xr>x~,

3255 J. Appl. Phys., Vol. 53, NO.4, April 1982 U. Gosele and K. N. Tu 3255
-~
41).,.

-
~ ;A}'B shrinks
.......
4 I)Ql.
~
)(
).,.
~
r2 ~ r2
~
... ...
0
:;::
...
c
rl
-...
0
c

-
)(

-
)( :::J
.:!

FIG. 5. As in Fig. 4, except that it is now assumed that the transport FIG. 6. As in Figs. 4 and 5, except that it is now assumed that the
across both layers, ApB and AyB, is interface controlled, xp <x~ and transport across the AyB layer is diffusion controlled (Xy>x;':) whereas
Xy <x~. In this case the flux ratio r is independent of the layer thick- the transport across the A(JB layer is interface-controlled (x(J <x~). At
nesses and determined by the effective interface reaction barriers de- time t, in case (c) the growing AyB layer has reached its critical thick-
scribed by x: and K"yff. ness x~."t above which the A(JB layer can grow.

the ApB layer and prevents further shrinkage. Physically, may be expressed as
diffusion fluxes cannot become infinite since they are limited
by the rearrangement processes at the interfaces. Conse- (33)
quently, it can be expected that under appropriate condi-
tions phases may shrink away completely. The two less interesting cases are those in which for the ini-
tial value of r either r> r2 (Fig. 6, case a) or r 1 < r < r2 (Fig. 6,
D. Transport across both layers interface-controlled case b ) holds, since in both cases the two layers can coexist all
the time. In case (a) the ArB layer shrinks until it reaches a
We now assume, in contrast to the case in the preceding
constant thickness and theApB layer grows all the time. 19 In
subsection, thatxp andx r are so small (xp<xt, xr<x~) that case (b) the ArB layer grows until it reaches the same con-
the transport across each layer is controlled by interface re- stant thickness as in case (a) and theApB layer grows all the
action barriers. Then the flux ratio r is given by time.
(32) The most revealing case for our understanding of the
The ratio r does not depend on time or on xp or xr as differences between the thin-film case and the bulk case is
long as xp <xt and xr <x~ remain fulfilled. If, by chance, K'}t case (c) in Fig. 6 in which the initial value of r is smaller than
and K~ff are such that r lies between r 1 and r2 both layers grow r I' TheArB layer grows and theApB layer shrinks as long as
simultaneously (Fig. 5, case b). In the more likely case that r xr is so small that r < r I still holds. The flux ratio rand
is somewhere outside this range one of the phases will shrink therefore also the shrinkage behavior of the ApB layer is
away completely (Fig. 5., cases a and c), which obviously also independent of its thickness xp (as long as xp <t). Therefore,
means, that this phase, e.g., ApB in case (c) would not start to we conclude that for kinetic reasons in a usual interdiffusion
experiment the ApB layer (even if its nucleation poses no
grow at all in a usual diffusion experiment as long as xr < x~
problem) cannot coexist with the ArB layer as long as xr is
holds. Eventually the thickness xr of the growing compound
below a critical thickness x~it defined by
layer will exceed x~ and its growth will become diffusion
controlled. Such a situation will be considered in the follow-
(34)
ing subsection.
Equation (34) withxr <x~rit expresses the same condition as
E. Different transport-control mechanisms across the r < rl' For xr >x~rit the ApB can grow simultaneously with
two layers
the ArB layer. Since the ApB layer keeps growing, in the
We now deal with the case that the transport across one long run the condition xp <xt will no longer be fulfilled.
layer, say the ArB, is diffusion controlled (x r >x~) whereas Finally, the transport across both layers will be diffusion-
the transport across the other layer, ApB , is interface con- controlled and both layers will grow proportional to V t with
trolled (xu <xt). Under these circumstances the flux ratio r a constant thickness ratio xrlxp.

3256 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3256
IV. DISCUSSION AND APPLICATIONS IN THIN-FILM I
l~~~AyB~
(oj
SILICIDE FORMATION
A. Restrictions to applicability !
I
Before we discuss the results obtained for our simple
!
model system and apply these results to thin-film metal-sili-
con diffusion couples we comment on complications which ..
:I
~
u
?:
First
Compound Ay B _ I Second
I Compound
I ApB_
may be encountered in real binary diffusion couples as com-
pared to our model system. Some of these complications ex-
....
f-

o I
clude the direct applicability of our results to various diffu-
...J
!
I
sion systems.
o~----------------~~---------
(i) The diffusion in a compound phase has been_de- I cri ! lime
scribed by a single chemical interdiffusion coefficient D. If
the diffusion processes of A and B atoms in a compound !
phase are independent of each other, the kinetic description
becomes more complicated and involves more parameters,
(b) ~--;-yB-~-~-~ I ~hE ryj
but qualitatively our basic conclusions remain unchanged. I
(ii) In our model system only two competing compound .,on -----1---------
phases, ApB and AyB , have been dealt with. In most binary
~
.. I
I
u
systems the phase diagram predicts more than two com- :<;
f-
pound phases. 2o Qualitatively our conclusions may still be .....
applied if one keeps in mind that now all the phases are o
...J
competing with each other.
(iii) Energy contributions and changes associated with
the interfaces have not been considered. If, e.g., the ApB lime
layer nucleates between AyB and AaB the single interface FIG. 7. Schematic of growth/shrinkage behavior of first forming com.
AaB IAyB is replaced by two new interfacesAaB IApB and pound (AyB) and growth of secondly forming compound (ApB) in an
ApB IAyB which most likely represents a higher energy A / B diffusion couple, starting from the components A and B. In (a) the
AyB layer reaches its critical thickness x~it at time I, where unreacted
state. The energy balance of the various interfacial energies
A and B is still unavailable. In (b)B is transformed completely into
and the compound formation energies involved may lead to AyB at time In < I I (case of second phase formation induced by supply
a nucleation barrier for a compound phase. 10 Such a nuclea- limitation).
tion barrier, if present, has to be superimposed on the kinet-
ic-instability effect, which probably in many cases is the rea-
son for the absence of all but one compound phase in thin-
couples. Neglecting nucleation problems we conclude that
film diffusion couples.
an AlB diffusion couple starting from the components A and
(iv) The effective diffusivity in a compound layer has
B shows the following kinetic behavior within the frame-
b(.!n assumed as independent of time and layer thickness. In
work of our model system: the compound phase with the
thin films grain boundaries often represent the most impor-
lowest effective interface reaction barrier, say AyB, will
tant diffusion paths. 2l ,22 If during layer growth the micro-
form first and initially grow linearly with time and then with
structure of the layer changes, e.g., by grain growth, nonpar-
a parabolic growth law. After the thickness Xy of the first
abolic layer growth will result. 16,23 Even for this case the
compound phase has exceeded a critical thickness x~ri"
concept of an interface-induced kinetic instability of com-
which depends on the microstructure of the compound layer
pound phases in thin-film diffusion couples remains applica-
and in general also on temperature, the second compound
ble, although only in its qualitative aspects.
phase (ApB ) will form and grow [Fig. 7(a)], first almost lin-
(v) The possibility of metastable thin-film phases has
early in time and later on diffusion controlled (as AyB) pro-
not been taken into account. Such metastable phases may be
portional to v't with a time-independent ratioxylxp. Which
stabilized by mismatch stresses24 or by excessive lattice dis-
of the compound layers is thicker in this final diffusion-con-
order, e.g., due to ion implantation. 25
trolled stage does not depend on which phase forms first. For
(vi) The influence of impurities has been neglected. In
completeness we mention that in exceptional cases the inter-
Sec. IV C we will comment on that aspect in the context of
silicides. face reaction barriers may accidentally be such that both
phases AyB and ApB can grow simultaneously from the
B. General behavior of thin-film diffusion couples very beginning (see Fig. 5, case b ).
In sufficiently pure thin-film silicon-metal diffusion
In Sec. III we have dealt with the kinetic behavior of a couples, which may form a number of silicides according to
model system with two intermediate compound phases ApB the phase diagram, indeed usually only one compound phase
and AyB between two saturated phases AaB and AlJB. In can be observed as long as unreacted metal and silicon are
the following we consider these saturated phases basically as still available. 3 •4 The silicide layer thicknesses investigated
the pure components A and B (a_ 00 ,0-0) where, e.g., A by backscattering techniques are typically on the order of a
stands for silicon and B for metal in silicon-metal diffusion couple of hundred to some thousand Angstroms. In this

3257 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3257
thickness range, the first silicide phase grows diffusion con- first one has been assumed in all our previous discussions
trolled proportional to V t in diffusion couples of silicon with and occurs if the supply of A (silicon) and of B (metal) re-
near-noble metals and interface controlled proportional to t mains unlimited. Then a second phase forms when the first
in silicon-refractory-metal systems. 3 .4 For a compilation of compound phase has exceeded its critical thickness x cri !.
data on the growth kinetics of various silicides, see, e.g., Tu This behavior has possibly been observed in the experiments
and Mayer 3 and Ottaviani. 4 of Petersson et af.28 on the formation of thin· film rhodium
If the first silicide is still growing interface controlled silicides,29 in Rh/Si thin-film diffusion couples first a RhSi
(as in the case of CrSi 2 and MoSi 2 ), then our results in Sec. layer forms which grows diffusion controlled. For long an-
III D predict, in agreement with experimental observation, nealing times an additional Rh 2 Si layer forms and grows
that a second silicide phase cannot grow or simultaneously simultaneously in the presence of still unreacted rhodium
exist at all together with the first silicide phase (except in the and silicon. 28 Further experiments are desirable to check this
unlikely case that the flux ratio r is accidentally between r l growth behavior in detail and to exclude impurity effects,
and r2 ). In the cases in which only one silicide layer is present which are most likely responsible for the simultaneous oc-
and this layer grows diffusion controlled, according to our currence of two silicide phases in the presence of still un-
model its thickness x must have been below its critical value reacted metal and silicon as sometimes reported for
x Cri ! defined in Eq. (34), because otherwise a second silicide Si/Ni 13,30 or Si/pe 1-33 thin-film diffusion couples. 34
phase would have started to grow. In Sec. IV C we will give In most thin· film investigations of silicon-metal diffu-
an estimate for x Crit of the first growing silicide phase (Ni 2 Si) sion couples the second silicide phase formed because of one
in Si/Ni diffusion couples. of the remaining two possibilities, namely that either the
Within our concept the first silicide phase forming and deposited metal film on silicon or the deposited silicon on the
growing is that with the lowest effective interface reaction metal has been consumed. 3,4,13,35-37 The reason for this type
barrier, it is basically a question of growth kinetics and not of second phase formation, which may be termed supply
one of nucleation 10,25 as, e.g., assumed in the rule of Walser limitation induced, can easily be understood within our sim-
and Bene26 predicting, with varying success27 the first sili- ple model system. Suppose the starting component B of an
cide phase from the phase diagram. Unfortunately, at pre- AlB diffusion couple is a thin film on a much thicker sub-
sent our concept does not allow a reasonable first phase pre- strate consisting of the other component A. All the starting
diction from the phase diagram, mainly because no obvious material B may be transformed into the first forming phase,
and clearcut procedure is known to extract information on say againAyB , beforex y >x~rit. After a short transition peri-
interface reaction barriers from the phase diagram, if such od a new steady-state diffusion situation will develop which
information is contained in the phase diagram at all (which is may approximately be described by j~ = O. Eqs. (IS) and (16)
doubtful). simplify to

C. Second phase formation dX{3 'A

dt = Gf3}f3, (35)
The compound phase which forms as the second one
depends on the materials supply conditions according to our dXyldt = - G)l! j~. (36)
model. In principle three possibilities exist (see Fig. 8). The
Equations (35) and (36) show that the Af3B layer will grow
and form if not yet ex.istent even though xl' <x~,rit whereas
theAyB will shrink [see also Fig. 7(b)]. In our simple model
system the case of second phase formation induced by supply
limitation can be realized only if the supply of B is stopped
after the first compound phase A yB has formed. Cutting off
the supply of A would stop further compound formation all
together. In binary diffusion systems with more than two
possible equilibrium compound phases, different second
Xm > x~it before:
A or B consumed
phases can be induced depending on whether the supply of A
or B is cut off. The various possibilities for a second phase
formation are indicated in Fig. 8.
It is likely, though not certain, that the second phase
forming at unlimited supply of silicon and metal will coin-
second compound second compound
cide with one of the two second phases induced by cutting off
second compound
AkB forms and AjB (j<m) forms and AnB (n>m) forms and the supply of silicon or metal which shows the lower inter-
grows; AmB grows grows; ArJ3 shrinks qrows; AmB shrinks face reaction barrier or, in more practical terms, the lower
formation temperature. Using this criterion for a Ni/Si dif-
FIG. 8. The three possibilities for second phase formation under differ· fusion couple with still unreacted Ni and Si and the data of
ent supply conditions of the components A and B in a planar A / B dif· Ottaviani et a1?S,36, we conclude that the second silicide
fusion couple. The compound AkB may coincide with either A,B or
phase which forms after the first phase Ni 2 Si has exceeded a
AnB. If the first forming compound A",B is already the compound rio
chest in A or B, a second compound phase AnB or AiB, respectively, critical thickness xcrit, consists of NiSi (and not of Ni~Si2'
is not existent. which forms at siightly higher temperatures and represents

3258 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3258
the alternative to NiSi as a possible second phase). For esti-
mating x Cri ! for the NizSi layer we need data on the NizSi and
on the NiSi growth kinetics. It is well established that NizSi
grows diffusion controlled as long as unreacted Ni and Si are
still present.3.38.39 The work of Coe and Rhoderick40 indi-
cates that after transformation of all Ni into NizSi, a NiSi
'0
layer forms and grows interface controlled linearly with II)

time. Taking their data for an estimate of x Cri ! via Eq. (34) we ...c
II)

....u
arrive at :c
I-
(37)
for the Ni 2Si layer, almost independent of temperature be- time
tween 300 and 430 ·C. The critical thickness given in Eq. (37) FlO. 9. Shrinkage and subsequent growth of the secondly forming
represents a rough estimate which in practice may vary from compound layer A/,B after new supply of un reacted B becomes avail-
one NiiSi diffusion couple to another since the parameters able (for details,see test). In case II the A~B layer shrinks away com-
involved, depend on the actual microstructure of the Ni 2 Si pletely and forms anew and grows only after the first forming ArB
film and on the cleanliness of the interfaces. Impurities at the layer has exceeded its critical thickness x~:".
Ni 2SilSi interface may strongly affect the effective reaction
barrier for NiSi formation. There are in fact, experimental
indications that NiSi does not consistently grow linearly Suppose of the first forming compound AyB in our
with time, but rather soon approaches a vtgrowth law. 41 model system has consumed all starting material B and the
This in turn would affect the value of the critical thickness second compound ApB has formed and grows, whereas
x Cri ! given above. AyB shrinks as discussed in Sec. IV B. If a new supply of B
The two compound phases NizSi and NiSi have been material is offered by depositing B material on the AyB
observed to grow simultaneously and diffusion controlled in phase, Eqs. (15) and (16) show that under these circum-
the presence of still unreacted Si and Ni if the unreacted stan.ces the AyB layer will start to grow again whereas the
silicon is amorphous. 39 It can, in fact, be expected that the ApB layer will start to shrink. Even if before deposition of
interface reaction barrier for transformation of amorphous fresh B material the AyB phase had completely been con-
silicon and Ni 2 Si into NiSi is smaller than the corresponding sumed and transformed into theApB phase, theAyB phase
one for the transformation involving crystalline silicon. will form anew and grow after B deposition.
Thus in the case of amorphous silicon the critical thickness The question arises whether the ApB (NiSi,PtSi) layer
x crit of the Ni 2 Si layer before the NiSi forms is likely to be will shrink away completely. This question has implicitly
smaller than x crit in the case of crystalline silicon. been answered in Sec. III already: If the growing A yB
As mentioned above it has been observed experimental- (Ni 2 Si,Pt 2Si) layer reaches its critical thickness before the
ly that two silicide phases can coexist in thin-film diffusion ApB layer has decomposed completely, the thickness xp of
couples even in the presence ofunreacted metal and silicon if theApB layer will pass through a minimum and then start to
the samples are impure. 3 ,13.30,33 Impurities may either re- grow again (case I in Fig. 9). Otherwise the ApB layer will
duce the effective interface reaction barrier for the second shrink away completely and start to form and grow again
phase or, more likely, limit the supply of metal to the first only after the thickness of the AyB layer has exceeded its
compound phase thus leading to the situation of a second critical thickness x~it (case II in Fig. 8). In the experiments of
phase formation induced by supply limitations as discussed Ottaviani et al.4.35.42 the first part of case II seems to be ful-
above. In this latter case,the first phase grows more slowly or filled since the reappearance of the NiSi or PtSi layer has not
even shrinks, a behavior which is in accordance with experi- been observed.
mental results, e.g., for oxygen contaminated SilNi 33 or
SilPe l thin-film diffusion couples. V.SUMMARY

D. Supply-induced second-phase shrinkage The influence of interface reaction barriers on the kinet-
ics of planar binary diffusion couples has been investigated.
Ottaviani and coworkers4 ,35.42 studied the kinetics of Within the phenomenological concept of interface reaction
various diffusion couples consisting of Si plus a layer of the barriers it can be understood that, in thin-film diffusion cou-
silicide phase forming secondly (after all metal has reacted) ples in contrast to bulk diffusion couples, most compound
and in addition a newly deposited layer of metal. Examples phases predicted by the equilibrium phase diagram do not
are the systems Si/PtSi/Pt and Si/NiSiINi, They found that occur, basically because of their kinetic instability. This ki-
the compound phase formingjirst in a usual Silicon/Metal netic instability is effective only for a sufficiently thin diffu-
diffusion couple, i.e., Pt 2 Si or Ni 2 Si in a Si/Pt or a SiiNi sion zone. For thick diffusion zones the theory presented is
system, respectively, forms anew and grows at the expense of in accordance with eqUilibrium theories of diffusion-con-
the preexisting PtSi or NiSi layers. The experimentally ob- trolled growth of multi phase planar diffusion systems as ap-
served shrinkage (decomposition) of the second phase may plicable to bulk diffusion couples.
also be understood in terms of our simple model system. It is predicted that after the first compound phase has

3259 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3259
formed in a binary diffusion couple this phase must exceed a '4Ya. Yeo Geguzin, Yu. S. Kaganovskiy, L. M. Paritskaya, and V. I.
critical thickness before a second compound phase may form Solunskiy, Phys. Met. Metallogr. 47, 127 (1980).
"B. E. Deal and A. S. Grove, J. App!. Phys. 36, 3770 (1965).
and grow simultaneously with the first one. By restricting the 'hH. H. Farrell, G. H. Gilmer, and M. Suenaga, Thin Solid Films 25,
supply of one of the components of the diffusion couple, the 253 (1975).
formation and growth of a second phase may be induced "Ya. Yeo Geguzin and Ya. S. Kaganovskiy, Fiz. Met. Metalloved. 39,
even before the first compound exceeds its critical thickness. 553 (1975).
'"If the phases A"B and AyB are not saturated an additional negative
A critical thickness of about 2/im has been estimated for the term appears in both Eqs. (15) and (16).
first forming compound Ni 2 Si in an Ni/Si diffusion couple '"Even in case (a) the ArB layer may shrink away completely, if due
before the second compound phase NiSi will form and grow to its shrinkage the transport mechanism across this layer changes
in the presence of still unreacted nickel and silicon. over from diffusion to interface controlled.
2<'For example see M. Hansen Constitution of Binary Alloys (McGraw-
Finally we mention explicitly that we do not rule out
Hill, New York, 1958).
that in specific cases the formation of a given phase in a "R. W. Balluffi and J. M. Blakely, Thin Solid Films 25, 363 (1975).
binary diffusion couple may be prevented by a nucleation "For further references see, e.g., D. Gupta and P. S. Ho, Thin Solid
barrier:o rather than by the kinetic or growth instability de- Films 72, 399 (1980).
"H. H. Farrell, G. H. Gilmer, and M. Suenaga, J. AppJ. Phys. 45,
scribed in this paper.
4025 11974).
HE. S. Machlin and P. Chaudhari (to be published).
ACKNOWLEDGMENTS "For example see B. Y. Tsaur, J. W. Mayer, and K. N. Tu, J. App!.
Phys. 51, 5326 (1980).
The authors appreciate discussions with G. Ottaviani, 'hR. M. Walser and R. W. Bene, App!. Phys. Lett. 28, 624 (1976).
T. Y. Tan, and R. D. Thompson on the subject of this paper. "For notable exceptions to the rule of Walser and Bene see, e.g., R.
D. Thompson, B. Y. Tsaur, and K. N. Tu, App!. Phys. Lett. 38, 535
(1981).
'"So Petersson, R. Anderson, 1. Baglin, J. Dempsey, W. Hammer, F.
d'Heurle, and S. LaPlaca, J. App!. Phys. 51, 373 (1980).
'For examples, see Thin Films, Interdiffusion and Reactions, edited by '''Cu/Ti thin-film diffusion couples also appear to show this growth be-
J. M. Poate, K. N. Tu, and J. W. Mayer (Wiley-Interscience, New havior, see, e.g., Ref. 12.
York, 1978). "'D. M. Scott, P. J. Grunthaner, B. Y. Tsaur, M. A. Nicolet, and J.
'For examples, see Thin Film Phenomena-Interfaces and Interactions, W. Mayer, in Thin Film Interfaces and Interactions, edited by J. E.
edited by J. E. E. Baglin and J. M. Poate (Electrochemical Society, E. Baglin and J. M. Poate (Electrochemical Society, New York,
New York, 1978). 1980), p. 148.
'K. N. Tu and J. W. Mayer, in Thin Films, Interdiffusion and Reac- "J. M. Poate and T. C. Tisone, App!. Phys. Lett. 24, 391 (1974).
tions, edited by J. M. Poate, K. N. Tu, and J. W. Mayer (Wiley- 12R. J. Blattner, C. A. Evans, S. S. Lau, J. W. Mayer, and B. M. Ull-
Interscience, New York, 1978), Chap. 10. rich, J. Electrochem. Soc. 122, 1732 (1975).
4G. Ottaviani, J. Vac. Sci. Techno!. 16,1112 (1979). "J. B. Bindell, J. W. Corby, D. R. Won sidler, 1. H. Poate, D. K. Con-
'G. V. Kidson, J. Nucl. Mater. 3, 21 (1961). ley, and T. C. Tisone, Thin Solid Films 37,441 (1976).
"c. Wagner, Acta Metal!. 17,99 (1969). "In the case of Si/Co thin-film diffusion couples it is unclear at the
7A. J. Hickl and R. W. Heckel, Metal!. Trans. 6A, 431 (1975). moment whether the simultaneous growth of Co,Si and CoSi in the
"So R. Shatynski, J. P. Hirth, and R. A. Rapp, Acta Metall. 24, 1071 presence of still unreacted Si and Co is an impurity effect or not. S.
(1976). S. Lau, J. W. Mayer, and K. N. Tu, 1. App!. Phys. 49, 4005 (1978).
"See also H. Schmalzried Solid State Reactions (Academic, New York, "G. Ottaviani and M. Costa to, J. Crystal Growth 45, 365 (1978).
1974), p. 120. "'c. Canali, G. Majni, G. Ottaviani, and G. Ce\otti, J. App!. Phys. 50,
"'J. E. E. Baglin, F. M. d'Heurle, and C. S. Peterson, in Thin Film In- 255 11979).
terfaces and Interactions, edited by 1. E. E. Baglin and J. M. Poate "E. G. Colgan, B. Y. Tsaur, and 1. W. Mayer, App!. Phys. Lett. 37,
(Electrochemical Society, New York, 1980), p. 341. 938 (1980).
"S. U. Campisano, in Thin Film Phenomena-Interfaces and Interac- '"K. N. Tu. W. K. Chu, and J. W. Mayer, Thin Solid Films 25, 403
tions, edited by J. E. E. Baglin and J. M. Poate (Electrochemical So- (1975).
ciety, New York, 1978), p. 129. '"J. O. Olowolafe, M. A. Nicolet, and J. W. Mayer, Thin Solid Films
"A. E. Gershinkii, A. A. Khoromenko, and E. I. Cherepov, Phys. Sta- 38, 143 (1976).
tus Solidi A 31, 61 (1975). 4°D. J. Coe and E. H. Rhoderick, J. Phys. D 9, 965 (1976).
"c. Canali, F. Catellani, G. Ottaviani, and M. Prudenziati, App!. 41J. E. E. Baglin and G. Ottaviani (private communication).
Phys. Lett. 33, 187 (1978). ·'G. Ottaviani, G. Majni, and C. Canali, J. App!. Phys. 18, 285 (1979).

3260 J. Appl. Phys., Vol. 53, No.4, April 1982 U. Gosele and K. N. Tu 3260