Microelectronics and Reliability, Vol. 14, pp. 43l to 433. Pergamon Press, 1975. Printed in Great Britain
A MODIFIED RELIABILITY EXPRESSION FOR THE
ELECTROMIGRATION TIME-TO-FAILURE
A. BOBBIO and O. SARACCO
Istituto Elettrotecnico Nazionale "Galileo Ferraris" Torino, Italy
AImtraet--Electromigration life data reported by different authors show that a single exponent for the
dependence of the mean time-to-failure (MTF) on the current density cannot be defined. A modified
expression is proposed in which, through the self heating, the current density is directly inserted in the
exponential term.
Furthermore a simple relationship between the mean rate of the resistance change and the time-tofailure of each specimen is presented.
Electromigration accelerated life tests have been widely
carried out in recent years in order to provide a simple
relationship between the mean time-to-failure (MTF)
and the applied stress. However a wide spread has been
noticed, as reviewed recently in [1], among the
experimental results of different authors.
An Arrhenius like dependence of the MTF on the
temperature has been always observed with activation
energies varying from 0.5 to 0"8 eV, while for the
current density dependence a single power law (of the
type MTF oc j-~) has been already proposed. Experimental values of the exponent n have been found
generally in the range of 1 to 3, but larger values, up to
10, have been also reported. In any case the assumption
of a constant exponent n implies that the weight of the
current density on the degradation rate should be
unaffected by the current density level itself.
Recently detailed physical models, which consider
electromigration damage as due to the growth of
localized voids acting as sinks for the diffusing
vacancies, have been presented. These models, although
differing in the shape of the growing cavity, spherical
[2], cylindrical [3] and crack shaped cavities [4] have
been considered, provide a very similar behaviour for
the dependence of MTF on the stress conditions,
particularly showing the role of the Joule heating on
the degradation rate.
In agreement with the kinetics models a modified
reliability formula is suggested in the following.
In this formula the linearity between MTF and
current density, as provided by the diffusion equation
of the Huntington and Grone's theory [5] is maintained
but a current density dependence is directly inserted in
the temperature term through the self heating.
Let us consider the simple equations that relate the
conductor resistance, the temperature and the dissipated power:
R = RA(1 + c t ( T - Ta)),
(1)
T = TA + bW.
(2)
Equations (1) and (2) can be combined to yield an
expression of the temperature increase above the
ambient, due to the self heating, as a function of j2:
AT(j2)
M&R 14/5--42
bRaj2
1/$2 _ 6Rao:j 2 ,
(3)
431
where b is the thermal resistance, R a the electrical
resistance at the ambient temperature TA, ~ the temperature coefficient of resistance and S the film cross
sectional area. All the parameters appearing in (3) are
easily measurable by preliminary testing. Thus the
proposed reliability formula may be written as:
MTF - AJ exp
K(T.4
TU2) ,
(4)
where A is a constant characteristic of the specimen,
~bthe activation energy and K the Boltzmann constant.
The suggested relationship takes into account,
through the function AT(j2), tile properties of heat
dissipation of the device, and its geometrical dimensions; moreover it puts explicitly into evidence
that the deviations from linearity can be ascribed to
the overheating of the film with respect to the environment in which the device operates and are mainly
related to the thermal resistance b of the film which
can be evaluated directly.
In Fig. 1 the conductor life is plotted on a log-scale
vs. the current density with the environmental temperature as parameter. Calculations have been performed assuming for the various parameters the
following values derived from previous experiments [2].
~b = 0"6 eV
Ra = 3"25 f~
K = 1'380 x 10-16 erg/°C
A = 0'05
b = 197-6°C/W
S =(15 x 1-6) = 24#m 2
= 0.00449oc- 1
The constant A, which has been obtained fitting
experimental results, lies in the range of values
reported by Black [6].
The discrepancies already observed in the literature
about the dependence of MTF on the current density
may be justified examining (4) and Fig. 1; in fact it
appears that a single exponent n cannot be univocally
determined since the degradation rate is a function
both of the applied stress, and of the thermal properties
of the conductor.
It is also evident that extrapolation of accelerated
test data using a typical value n = 2 yields too
432
A. BOBBIOand O. SARA('(:()
108
I(?
•
I0
10 3
o~
LD
~6
*
• E×perlmental
points
"m"a'~l~, C y l i n d r i c a l
2cu
io
d
Crackand~~
c
2
.c
l0
IO T
c
o
~: Jc
id v
10
I0"
IO;
Life,
F-
IOt>
,'/*
hr
Fig, 2. Mean rate of change in resistance at 5 % vs. specimen
life-time. Experimental points are obtained from three
different series of measurements:
• TA = 70°C; j = 1"2 x 106 A/cm 2
•
Z A - 4 0 ° C ; j = 0"9 x 106 A/cm 2
• TA - 60°C; j - 0"8 x 106 A/cm 2
IO
103
104
Current
105
density,
406
A/cm
FOr
2
Fig. 1. Mean time-to-failure vs. current density with the
environmental temperature TA as parameter.
optimistic values of the conductor life at normal stress
levels.
It should still be noted that the suggested formula
neglects the formation of an hot spot in the weakest
section during the last stages of the conductor life;
this effect could be included adding in the temperature
expression terms of higher order in j ; these terms,
however, are difficult to evaluate experimentally.
For reliability purposes it is of interest to define a
useful life, that is a time during which the device can
correctly operate. In thin film conductors the useful life
may be conveniently referred to the conductor resistance change, which is an easily measurable parameter
related to the device wear-out.
For this reason we have applied the electromigration
damage models to find the relation between the rate of
the change in resistance in the first stages of the film
degradation and the time-to-failure.
The following expression has been derived:
TF =
C
(5)
where T F is the time-to-failure of each specimen,
[ ( 1 / R o X d R / d t ) ] y % the mean rate of the resistance
change calculated at a typical level y%, and C is a
constant.
The general character of (5) lies in the fact that the
constant C is quite independent on the applied stress
and on specimen characteristics, for all the considered
theoretical models.
The results are plotted in Fig. 2 for a mean rate of
the resistance change calculated for instance at 5 °'
q).
The straight lines obtained from the cylindrical,
spherical and crack models correspond to a value of
the constant C equal to 0.1 [3]. Available experimental
points in the range of accelerated tests lie very closely
on the theoretical curve.
This formula can provide in any case a rough
estimate of the time-to-failure of a thin film conductor
through resistance change measurements, that is by
non destructive experiments. But vice versa if a useful
life level, related to the change in resistance is
established, it can be evaluated from accelerated life
measurements.
CONCLUSIONS
A modified reliability formula is presented, in which
the role of the self-heating is emphasized, inserting
directly the current density dependence in the temperature term. Extrapolation of accelerated life data to
normal stress levels through this formula leads to less
optimistic values of the conductor life.
The introduction of the thermal function AT(j 2)
may be very useful in the statistical analysis of the life
test results. In fact, as it has been experimentally
proved [7], the spread in the time-to-failure distributions are in a large extent to be ascribed to the spread
in the distribution of the specimen thermal resistances,
i.e. in the distribution of the true operating temperatures.
Finally a simple theoretical relationship which
connects the time-to-failure with the mean rate of
resistance change, and which agrees with experimental
results, is reported. By means of this relationship a
useful life level may be determined from accelerated
experiments.
A Modified Reliability Expression for the Electromigration Time-to-Failure
REFERENCES
1. L. Braun, Microelectron. & Reliab. 13, 215 (1974).
2. A. Bobbio, A. Ferro and O. Saracco, IEEE Trans.
Reliab. R-23, 194 (1974).
3. A. Bobbio, A. Ferro and O. Saracco, Proc. "II Congr~s
National de Fiabilit6", Perros-Guirec, France (1974).
433
4. R. A. Sigsbee, J. appl. Phys. 44, 2533 (1973).
5. H. B. Huntingtone and A. R. Grone, J. Phys. Chem.
Solids 20, 76 (1961).
6. J. R. Black, IEEE Trans. Electron Devices ED 16, 3381
(1969).
7, A. Bobbio and O. Saracco, Thin Solid Films 17, S-13
(1973).