Académique Documents
Professionnel Documents
Culture Documents
From
these
results we that
see
can
even
though
the REFEKKNCES
leakage current increases and the transconductance [1] A. G. Revesz and E;. H.Zaininger, RCA Ret’,,VOI 29, p. 22,19ciE,.
decreases, the transistors are still “alive” and operative, [ 2 ] K. H. Zainingerand A. G.Holmes-Siedle, RC.1 R e v . , vol. ,253,
p. 208, 1967.
Thus, in the case of neutron bombardment, A1203 hiOS [3] K. H. Zalninger,“Irradiation of N I S capacitorswith ;7igh
transistors
also
seem
quite
be
to“hard.” energy
electrons,” I E E E Trans. iVZlCka7 Science, vc~l.NS-13,
pp. 237-247, December 1966.
[4] E.H.Snow, A. S. Grove, and I). J . Fitzgerald, “Effects of icmiz-
VI. CONCLUSION ing radiation on oxidized silicon surfaces and planar devices,”
Proc. I E E E , vol. 5 5 , pp. 1168-1185, July 1967.
I t hasbeendemonstratedthatfabrication of MOS [5] J. P. IIitcheIl,“Radiation-inducedspace-chargebuildupin
510s structures,” IEEE Tvans. Electvon Devices, vol. ED-14,
devicesemployingagateinsulatorobtainedbyplasma pp. 764-774, November 1967.
anodization of evaporated aluminum is feasible. These [6] A. G.Revesz, K. H. Zaininger,and R. J . Evans, to bepublished
in J . Electrochem. SOC.
devices exhibit excellent characteristics and are remark- [ j ] J , Lindyayer and \v. p. Sable, J r , , “Radiation resistant AIOS
resistant
radiation.
ably
to devlces, I E E E Trans. Electron Devices, vol. ED-15, pp. 637-
640, September 1968.
[SI A. S.Lf’axman and K. H. Zaininger, AppZ. Phys. Letters, vol.
ACKNOWLEDGMENT 12, p. 109, 1968.
[9] J . Aliles and P. Smith, J . Electrochem. SOC.,vol. 110, p. 1250,
F’ and R* have assisted in the [IO] *\96i; \vaxman,Electronics,p, 88, .\Iarc! 18, 1968.
fabricationofthedevicesand in measurements; F. [11] A. G. Stanleyand H. A. R. IVegener, Effects of electron ir-
Kolondracarriedoutthebombardmentand P. Rappa- radiation of metal-nitride-semiconductor insulated gate field-
effect transistors,” Proc. I E E E (Letters), vol. 54, pp. 784-785,
port gave constant encouragement, Alay 1966.
is still
unknown
of the processes
involved in electromigration;however,todate,much
[12]-[19]. T h e Rensselaer group
studied the current induced motion of surface scratches
in bothmoltenand solid metals[1]-[3].Recently,
on bulk metals, and they have concluded that a metal
hlanuscript received December 31, 1968. This work was supported ionwhich has beenthermallyactivatedand is a t its
in part by the Rome .Air Development Center, Air Force Systems saddle point (lifted o u t of its potential well and is es-
Command, Griffiss AFB, Rome, N. Y.,under Contract F30602-67-
C-0166, Project 5519. This paper was presented at the IEEE Re- sentially free of the metal lattice), is acted on by two
liability Physics Symposium, LVashington, D. C., December 1968. opposing forces in anelectricallyconductingsingle
The author is with hIotorola Inc., Semiconductor Products Di-
vision, Phoenix, Ariz. 85008. band metal. (See Fig. 2.)
BLACK: ELECTRO34IGRATION 339
LMTF kT
ductingelectrons colliding with
the
activated
metalions will exertaforce on themetal ion in where
direction
the of electron flow. hITF =medianfailure
time
to in hours,
A = a constant which contains a factor involving
Because of shielding electrons, the force on the ion
the cross-sectional area of the film,
due to the electric field is quitesmall;therefore,the
J=current density in amperes per square centi-
predominating force is t h a t of the “electron wind.” As a
meter,
result,activatedmetal ionswhich areupstream in
4 =an activation energy in electron volts,
terms of electron flow from a vacancy have a higher
k = Boltzman’s constant, and
probability of occupying the vacancy site than do other
T = film temperature in degrees Kelvin.
near neighbor ions which surround the vacancy. hIetal
ions,therefore,traveltowardthepositiveend of the Experinlents carried out with readily cooled metal films
conductor while vacanciesmovetowardthenegative enabledtheexperimentalcurrentdensitytorange
end. The vacancies condense to form voids while the greater than 5 : 1 and appear to confirm the J 2 relation-
ions condense a t certain discontinuities to form crystals, ship.
Xvhiskers, and nondescript hillocks.Surface
scratch I t was also shown that the activation energyfor elec-
marks on bulk metal conductors behave as vacancies tromigrationwasstronglydependentuponthestruc-
and are observed to move along the surface of the con- ture of the film [ X ] . Small crystallite (1.2 p) aluminum
ductor in the direction of the negative terminal. The films enabling iondiffusion to take place in the grain
OPE
BLACK: ELECTROAIIGRXTIOS 341
900
600
ul
a
a
-g 700
U
. . . . . . . . . . . . . ...
................ - , ,
..... -. - . . . . . - .. . .. ....
. . . . . . ,....
500 . . . . , . . . . . . . ~.. . . . . . . ~. . . . . . .
. I + . - . :. :. : :
.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..... . ,. .. ..... .. .. . .l;;;;.
......................................
. . . . . . . - ..... ~___
...............
. . . . . . . f . ./,
. . . . . . . . , . . . . . . . . . . . . ,- . .z-.. . . . . . . . . --:-.
_ _.fiLLT
..
........ . . . . . . . I . . ........
400 . . . . . . . . . . . . , . . . . . . . .
.. .. .. .. .. .. .. .. .. .. .. .. . .( .- .. ... . . .. .. . .
........
. . .. .l ._
' ..........
........
_ . _
200
IO0
0
I00 200 300 400 500 600 '100
EXPERIMENTAL MTF ( HOURSl
Fig. 9. Experimentally determined versus calculated mean time to failure for large grained aluminum films of variable line width.
temperature near 275OC. At this temperature and above, Intheabove discussion,equations were developed
lattice diffusion is predominant over surface and grain which express the lifetime of long aluminum conductors
boundary diffusion processes; thus, film structural which contain no gradients of temperature,current
effects are not important in that temperature range. At density, or ion diffusion coefficient. If gradients exist as
temperatures lower than27j°C,however,orders of defined in Fig. 15 i t wouldbeexpected t h a t hillocks
magnitudeimprovementinlifetimecanbeobtained andvoids would preferentiallygrow a t t h e positions
BLACK:ELECTROlIIGRATION 343
f j .I ,I . . , . .. .. .. .. .. .. .. .. .. .. .. ..
. , , t
160 , ! : i
................
t
c
I
40 60 eo 100
E X P E R I M E N TM
ATLF HOURS)
Fig. 10. Experimentally determined versus calculated mean time to failure for large grained aluminum film conductors of variable thickness.
indicated in t h a t figure. This is because aluminum ions t a n t excess vacanciescondense at that point to form
in the plateau moving to the right arrive at the negative voids. In addition, voids and whiskers form more slowly
gradient faster than they are removedfrom t h a t region. in gradient-free regions because vacancies generated by
The aluminum ions therefore accumulate in this region ions flowing downstream are readily filled by ions mov-
and formcrystals,whiskers,ornondescriptextruded ing from upstream portionsof the conductor.
appearing formations just upstream (in terms of elec- Films containing positive gradients of temperature,
tron flow) from a negative gradient. In a similar man- currentdensity,or iondiffusioncoefficientwouldbe
ner,aluminum ions justdownstream fromapositive expected to fail earlier than would be predicted by the
gradient are removed from that region faster than they equationsgenerated in theprevioussection of this
are replaced by ions from further upstream. The resul- paper. This is duetothe presence of regionswhere
344 IEEETR.ANS.ICTIONS O S ELECTRONDEVICES, APRIL 1969
c
I
s
3
L
L
L
C
346 IEEE TRANSACTIONS DEVICES,
O N ELECTRON APRIL 1969
f 1
TEMPERATURE,
CURRENT
f
HILLOCK
FORMATION
DENSITY
OR
DIFFUSION
COEFFICIENT NEGATIVE POSITIVE
GRADIENT GRADIENT
(-)
CONDUCTOR LENGTH 4
E L E C T R O N FLOW 4
(-)
CONCLUSIONS
Electromigrationhasbeenidentified asapotentia 1
\\-ear-outfailuremode for semiconductordevices pos
sessingconductors of inadequate cross-sectionalarea
Althoughelectromigration in metalshas been k n o w1
for several decades it is still not well understood. Design
equationshavebeenpresented which enable the con-
struction of long line aluminum conductors nith essen-
tially infinite lifetime. This is based upon the structure
of the aluminum, the current density, conductor tem-
perature, and conductor cross-sectional area. It has also
been showm that a positive gradient (in terms of elec-
tron floxv) in the diffusion coefficient of the metal ions
producesaregionwherevacanciescondensetoform
Fig. 19. Whisker and hillock growth a t negative A I contacts to Si
diffused resistors (ohmic hole in glass is 0.5 by 0.3 mil). voids and greatly reduce film life. I t is predicted that
similargradients in temperatureandcurrentdensity
will also reduce film life.
ACKKOWLEDGMENT
The scanning electronmicroscope pictures were made
by J. Devaney of theJetPropulsionLaboratory,
Pasadena, Calif.
REFERENCES
K. E.Schwartz, Electrolytisrhe TT70nderung influssizen und festen
Metallen. Leipzig: 1910.
\V. Seith, D-fiatsion in .Iletallen. Berlin:Springer, 1955.
V. B.Fiks, “On the mechanismof the mobility of ions i n metals,”
Sooiet Phys.-Sold State, vol. 1, p. 14, 1959.
I . ;i. Blech et al., RADCTech.Rept. T R 66-31, December
1965.
[5] I. X . Blech and H. Sello, presented a t the Annual Symposium
on Physics of Failure in Electronics,Columbus,Ohio, No-
vember 1966.
Fig. 20. Enlarged view of whiskerand hillock growth a t negative 161 \Y,hIutter,presented at the SpringlIeeting of theElectro-
-11 contacts to Si diffused resistors. chem. SOC., Dallas, Tex., abstract 62, p. 96, May1967.
[7] D. Chhabra and N. .qinsIey, IBM Components Division, East
FishkillFacility, N . Y., Tech. Rept. 22.419, J u l y 1967.
[8] J . R. Black, “Mass transport of aluminum by momentum ex-
change with conducting electrons,” Proc. 1967 Ann. S y m p . on
Reliability Physics, IEEE Cat. 7-lSC58, November 1967.
[91 P. B. Ghate, d p p l . P h y s . L e t t e r s , vol. 11, p. 14, 1967.
[lo] J . R. Black,“Metallizationfailures in integratedcircuits,”
R.4DC Tech. Rept. TR-68-213, October 1968.
[ l l ] -, “.Aluminum conductor failure by mass transport,” Proc.
3rdInternatl.Congr. on Micuoelertronics, llunich,November
1968.
[12] H. B.Huntington a;d X . R. Grone,“Current-inducedmarker
motion in gold wires, J . P h y s . C h e m . S o l i d s , vol. 20, nos., 1 and
2, pp. 76-87, ?961.
[13] .A. R.Grone,Currentinducedmarkertnotion in copper,” J .
Phys. Chem. Solids, vol. 20, nos. 1 and 2, pp. 88-98, 1961.
[14] R. \’. Penney, “Current-induced mass transport in aluminum,”
J . P h y s . C h e m . S o l i d s , vol. 25, pp. 335-!+5, 1964.
[15] J . F. D’.?mico and H. B. Huntingtov?Electromigrationand
thermomlgration in gamma-uranium, Rensselaer Polytechnic
Institute Department of Physics, Troy, S . Y., Rept. 1041-19.
[16] G. A. Sullivan,“Search for areversal i n copperelectrotnigra-
tion,” Rensselaer Polytechnic Institute Depxrtment of Physics,
Fig. 21. Void formation in positive .AI contacts to silicon diffused Troy, S . Y . , Rept. 1041-8.
resistors (ohmic hole in glass is 0.5 by 0.3 mil). [17] P. S. Ho, “Electromigration and soret effect i n cobalt,” J . P h y s .
Chem. Solids, vol. 27, pp. 1331-13:8, 1966.
[18] P.S. H o and H. B.Huntington,Electromigrationandvoid
observation in silver,” J . P h y s . C h e m . S o l i d s , vol. 27, pp. 1319-
minum film and of the hillocksindicatesextensive 1329, 1966.
mass transport of the aluminum to these regions. [19] .X, Ladding,“Currentinducedmotion of latticedefects in
indiummetal,” J . P h y s .C h e m . Solzds, vol. 26, pp. 143-151,
A SEI1 photomicrograph of the positive terminals of 1965.
two of the resistors is shown in Fig. 21. At these ter- 1201 J . R. Black, “Etch pit formation in silicon a t X I Scontacts due
to the transport of silicon in aluminum by momentum exchange
mifials electrons leave the silicon and enter the alumi- withconducting-electrons,” Puoc. OhmicContactsSymp., Elec-
num. The depletionof the aluminum forms a voidn-hich trochem. (?oc. Fall Meeting, hrontreal, Canada, October 1968.
[21] J . Bass,Theformationandmotionenergies of vacancies in
results in an open circuit, as is clearly indicated. aluminum,” P h i l . Mag., vol. 15, no. 136, p. 717, .April 1967.