Reservoir effect and the role of low current density regions on ...

Reservoir effect and the role of low current density regions on ...

ong>Reservoirong> ong>effectong> ong>andong> ong>theong> ong>roleong> ong>ofong> ong>lowong> ong>currentong> ong>densityong> ong>regionsong>

on electromigration lifetimes in copper interconnects

Z.H. Gan, a) W. Shao, S.G. Mhaisalkar, ong>andong> Z. Chen

School ong>ofong> Materials Science ong>andong> Engineering, Nanyang Technological University, Singapore 639798

Hongyu Li

Institute ong>ofong> Microelectronics, Singapore 117685

K.N. Tu

Department ong>ofong> Materials Science ong>andong> Engineering, University ong>ofong> California—Los Angeles,

Los Angeles, California 90095-1595

A.M. Gusak

Cherkasy National University, Cherkasy 18017, Ukraine

(Received 19 February 2006; accepted 18 April 2006)

Electromigration (EM) in copper dual-damascene interconnects with extensions

(also described as overhang ong>regionsong> or reservoirs) in ong>theong> upper metal (M2) were

investigated. It was found that as ong>theong> extension length increases from 0 to 60 nm, ong>theong>

median-time-to-failure increased from 50 to 140 h, representing a ∼200% improvement

in lifetimes. However, furong>theong>r increment ong>ofong> ong>theong> extension length from 60 to 120 nm did

not result in any significant improvement in EM lifetimes. Based on calculations ong>ofong>

ong>currentong> densities in ong>theong> reservoir ong>regionsong> ong>andong> recently reported nucleation, void

movement, ong>andong> agglomeration-based EM phenomena, it is proposed that ong>theong>re isa

critical extension length beyond which increasing extension lengths will not lead to

longer EM lifetimes.


It is well known that ong>theong> interfacial diffusion at Cu/

Si3N4 interfaces adversely affects ong>theong> electromigration

(EM) lifetime in Cu interconnects, in contrast to grainboundary

diffusion in Al alloy conductors. 1,2 Considerable

attention has been paid to modify this interface to

improve EM lifetime. Some researchers have used different

cap materials, 3–5 ong>andong> oong>theong>rs have proposed different

process treatments prior to ong>theong> deposition ong>ofong> ong>theong> dielectric

cap. 6 Anoong>theong>r good choice for delaying ong>theong> EM

failure is to introduce an extension (also described as

overhang region or reservoir) in ong>theong> interconnects. 7 It is

thought that ong>theong> extension serves as a reservoir for void

growth, thus prolonging ong>theong> median time to failure. Recently,

ong>theong> reservoir ong>effectong> in Al–Cu interconnect with W

vias was reported, 8,9 where voids are generally initiated

at ong>theong> W/Al interface by large Al atomic flux divergence.

Lower levels ong>ofong> stress ong>andong> vacancy concentration in ong>theong>

longer reservoir were proposed to contribute to ong>theong> better

electromigration reliability ong>ofong> interconnects. However,

a) Address all correspondence to this author.


DOI: 10.1557/JMR.2006.0270

ong>theong> reservoir mechanism in dual-damascene Cu interconnects

is clearly different from that ong>ofong> W/Al via structures

because voids nucleate at ong>theong> Cu/Si 3N 4 interface far from

ong>theong> cathode ong>andong> move along ong>theong> interface. 10 Because ong>ofong>

ong>theong> technological importance ong>ofong> dual-damascene Cu interconnects,

it is necessary to understong>andong> how extension

lengths impact electromigration lifetimes. In this work,

ong>theong> reservoir ong>effectong> ong>andong> ong>theong> ong>roleong> ong>ofong> ong>lowong> ong>currentong> ong>densityong>

ong>regionsong> on electromigration lifetimes in Cu dualdamascene

interconnects were investigated.


Test structures consisting ong>ofong> M1 ong>andong> M2 via-fed structures

(Fig. 1) were fabricated using 0.18 m Cu/oxide

damascene technology. M2 trench ong>andong> via were formed

by a via-first dual damascene process. Formation ong>ofong> M2

Cu metallization in undoped silicate glass (USG)

trenches involved deposition ong>ofong> a stack ong>ofong> 25-nm Ta barrier,

150-nm physical vapor deposited (PVD) Cu seed,

ong>andong> 600-nm electrochemical plated (ECP) Cu layers. A

cap ong>ofong> 50 nm Si3N4 was deposited after Cu chemical

mechanical polishing (CMP) process. Layers ong>ofong> 800-nm

USG, 50-nm Si3N4, ong>andong> 500-nm USG were ong>theong>n deposited

as intermetal dielectric. A 50-nm-thick Si3N4 served

as ong>theong> trench 2 etch-stop layer. M2 was 350 nm thick ong>andong>

J. Mater. Res., Vol. 21, No. 9, Sep 2006 © 2006 Materials Research Society 2241

FIG. 1. Schematic cross section ong>ofong> M2-via test structure.

via diameter was 260 nm. The M1 lines connected to

pads were short ong>andong> wide such that voids would be

formed in ong>theong> narrow ong>andong> long M2 test line between ong>theong>

vias. The M2 test lines were 500 m long ong>andong> 280 nm


EM tests were performed in a Qualitau (QualiTau,

Inc., Santa Clara, CA) package level EM test system. To

study ong>theong> reservoir length ong>effectong> on EM, M2 structures

with extensions ong>ofong> 0, 60, ong>andong> 120 nm were tested at

300 °C with a ong>currentong> ong>densityong> ong>ofong> 1.2 MA/cm 2 . In each

case, 12 samples were tested. Focused ion beam (FIB)

ong>andong> transmission electron microscopy (TEM) were used

for post-EM failure analysis.


The cumulative distribution ong>ofong> failure time for three

different extensions ong>ofong> M2 test structures are shown in

Fig. 2. The distributions could be described well with

log-normal distributions with similar slopes, thus indicating

similar failure modes in ong>theong>se structures. As ong>theong>

M2 extension was increased from 0 to 120 nm, ong>theong> EM

lifetime was improved correspondingly. For samples

without extension, Mean-time-to-failure (MTTF) was

about 50 h, whereas it was about 140 h (i.e., close to a

200% increment) in ong>theong> case ong>ofong> 60-nm extension. However,

as ong>theong> extension length was furong>theong>r increased from

FIG. 2. Cumulative Distribution Function (CDF) plot ong>ofong> median-timeto-failure

for 280-nm-wide M2 copper lines with three different M2



Z.H. Gan et al.: ong>Reservoirong> ong>effectong> ong>andong> ong>theong> ong>roleong> ong>ofong> ong>lowong> ong>currentong> ong>densityong> ong>regionsong> on electromigration lifetimes in copper interconnects

J. Mater. Res., Vol. 21, No. 9, Sep 2006

60 to 120 nm, EM lifetimes did not show significant

furong>theong>r improvement.

A post-EM cross-sectional FIB image ong>ofong> a 120-nmextension

with 10% resistance increase is shown in Fig.

3(a). It is clearly seen that at ong>theong> initial stage ong>ofong> EM, void

nucleation takes place preferentially at ong>theong> Cu/Si 3N 4 interface

in ong>theong> M2 line away from ong>theong> cathode via. It is

interesting to note that in departure from ong>theong> ong>currentong> understong>andong>ing

ong>ofong> ong>theong> EM mechanism based on tensile

stress-based void nucleation 11 ong>andong> ong>currentong> ong>densityong> gradient

induced void diffusion, 7 ong>theong> corner ong>ofong> ong>theong> M2 line

above ong>theong> via is free from voiding. Because ong>ofong> ong>theong> stressinduced

void nucleation mechanism, voids would be expected

to nucleate at ong>theong> cathode end ong>ofong> ong>theong> via as well as

at ong>theong> via bottom during EM, where high tensile stress

develops due to ong>theong> blocking boundary formed by ong>theong> Ta

liner. When ong>theong> critical tensile stress for void nucleation

is reached, voids may form ong>andong> grow, leading to via

opening. 12 It is also evident that ong>theong> ong>currentong> crowdinginduced

vacancy flux mechanism 7 does not apply eiong>theong>r

to ong>theong> observed failure mode [Fig. 3(a)]. This ong>currentong>

crowding-based mechanism postulates that a ong>currentong> ong>densityong>

gradient force drives vacancies from high ong>currentong>

ong>densityong> to ong>lowong> ong>currentong> ong>densityong> ong>regionsong>. Accordingly,

voids would nucleate at ong>theong> top corner ong>ofong> ong>theong> cathode end

in ong>theong> M2 interconnect.

FIG. 3. (a) Cross-sectional FIB image induced by electromigration

after 10% resistance increase, ong>andong> (b) cross-sectional TEM image after

30% resistance increase.

Z.H. Gan et al.: ong>Reservoirong> ong>effectong> ong>andong> ong>theong> ong>roleong> ong>ofong> ong>lowong> ong>currentong> ong>densityong> ong>regionsong> on electromigration lifetimes in copper interconnects

At ong>theong> 30% failure criterion for a 120-nm extension

structure, ong>theong> cross-sectional TEM image [Fig. 3(b)]

shows that ong>theong> EM-induced void has moved/grown along

ong>theong> Cu/dielectric-cap interface ong>ofong> ong>theong> M2 trench against

ong>theong> electron fong>lowong> direction, mainly driven by ong>theong> electron

wind force. 10 These observations lend furong>theong>r support to

ong>theong> hypoong>theong>sis that voids form heterogeneously at ong>theong>

Cu/Si3N4 interface away from ong>theong> cathode, migrate via

interfacial diffusion paths at this interface driven by ong>theong>

electron wind force, accumulate near ong>theong> cathode, ong>andong>

eventually lead to opening ong>ofong> ong>theong> via ong>andong> open circuit


The localized ong>currentong> ong>densityong> at ong>theong> via/M2 extension

region was evaluated by finite element analysis (FEA).

Figure 4 shows ong>theong> contours ong>ofong> ong>currentong> ong>densityong> (solid

curves) around ong>theong> M2 extension/via region for three

extension lengths. The contours ong>ofong> ong>currentong> ong>densityong> gradient

(dotted curves) have been overlayed on ong>theong> ong>currentong>

ong>densityong> contours. After ong>theong> interfacial void is generated in

ong>theong> middle section ong>ofong> ong>theong> metal strip away from ong>theong> via

[Fig. 3(a)], where ong>theong> ong>currentong> is uniform ong>andong> parallel to

ong>theong> interface, ong>theong> void migration along ong>currentong> direction

is driven by ong>theong> electron wind force (i.e., Fc): Fc =−Z*ej , (1)

where Z* is ong>theong> ong>effectong>ive charge number ong>ofong> ong>theong> diffusing

Cu atom, is ong>theong> resistivity ong>ofong> Cu, ong>andong> j is ong>theong> ong>currentong>

ong>densityong>. The void velocity is related to ong>theong> ong>currentong>

ong>densityong> by ong>theong> folong>lowong>ing equation 13 :

=−3NDskTaA Z*ej , (2)

where N is atomic ong>densityong>, is ong>theong> atomic volume, k is

ong>theong> Boltzmann constant, T is ong>theong> temperature in Kelvin, a

is ong>theong> radius ong>ofong> ong>theong> void, D is ong>theong> diffusion constant, ong>andong>

A is a factor related to ong>theong> void location.

Anoong>theong>r driving force due to ong>currentong>-ong>densityong> gradient

(i.e., Fg) 7 may also influence ong>theong> void migration

Fg =−dPdr , (3)

where r is ong>theong> three-dimension vector, P ( qv|j|ARv) is ong>theong> potential energy ong>ofong> an excess vacancy driven by ong>theong>

ong>currentong> ong>densityong> j, qv is ong>theong> charge ong>ofong> ong>theong> vacancy, A is ong>theong>

scattering cross section ong>ofong> ong>theong> vacancy, ong>andong> Rv is ong>theong>

resistance ong>ofong> ong>theong> excess vacancy.

However, analyzing Eqs. (1) ong>andong> (3) using ong>theong> material

parameters 14–16 qv Z*e 4e, A 1 × 10 −15 cm 2 ,

Rv 1×10 4 , ong>andong> 1.67 × 10 −6 cm leads to

ong>theong> inference that Fg is at least one order ong>ofong> magnitude

ong>lowong>er than Fc for ong>theong> location along ong>theong> M2/Si3N4 interface

as well as at ong>theong> ong>currentong> ong>densityong> contours (e.g., 0.4

MA/ cm 2 ), where ong>theong> values ong>ofong> |dj/dr| are between 2 ×

10 10 ong>andong>3×10 10 A/cm 3 . Therefore, ong>theong> ong>effectong> ong>ofong> ong>currentong>

ong>densityong> gradient (Fg) could be ignored in ong>theong> present


We propose that ong>theong>re exists a ong>lowong> ong>currentong> ong>densityong>

region (dead-zone) bounded by a critical ong>lowong> ong>currentong> den-

FIG. 4. Contours ong>ofong> ong>currentong> ong>densityong> (solid curves, unit: MA/cm 2 )

around ong>theong> M2 extension/via region for (a) 0-extension, (b) 60-nm

extension, ong>andong> (c) 120-nm extension. The contours ong>ofong> gradient ong>ofong>

ong>currentong> ong>densityong> (dotted curves, unit: 10 10 A/cm 3 ) in M2 are overlapped.

sity (j crit) around ong>theong> upper-left corner [Fig. 5(a)] that

highly retards ong>theong> movement ong>ofong> ong>theong> incoming voids ong>andong>

may prevent or resist void entry into ong>theong> M2 extension.

Figure 5(a), derived from Fig. 4, presents ong>currentong> ong>densityong>

contours ong>ofong> j 0.4 MA/cm 2 for extensions ong>ofong> 0 to 120

nm. It should be noted that ong>theong> ong>currentong> ong>densityong> contours

for different extensions have been overlayed along with

ong>theong> M2/via geometry to provide a direct comparison ong>ofong>

ong>theong> change in contours as a function ong>ofong> extension lengths.

J. Mater. Res., Vol. 21, No. 9, Sep 2006 2243

FIG. 5. (a) Overlapped contour lines around ong>theong> extension region corresponding

to 0.4 MA/cm 2 for varied M2 extensions, indicating a ong>lowong>

ong>currentong> ong>densityong> zone that may retard furong>theong>r movement ong>ofong> incoming

voids. (b) Normalized “ong>effectong>ive reservoir volume” in terms ong>ofong> extension


It may be envisaged that ong>theong> incoming voids are retarded

near a j crit boundary, song>lowong>ing down or even perhaps

blocking ong>theong>ir progress into ong>theong> ong>lowong> ong>currentong> ong>densityong> ong>regionsong>

located at ong>theong> far corners ong>ofong> ong>theong> reservoirs. Only a

part ong>ofong> ong>theong> extension volume may ong>effectong>ively serve as a

reservoir for accumulation.

It is observed in Fig. 5(a) that ong>theong> ong>lowong> ong>currentong> ong>densityong>

zone changes shape ong>andong> shifts slightly to ong>theong> left with

varying M2 extensions. It is thus instructive to compare

ong>theong> “ong>effectong>ive reservoir volume” above ong>theong> via/M2 interface,

which is defined here as ong>theong> ong>lowong> ong>currentong> ong>densityong>

volume was subtracted from ong>theong> overall geometrical volume.

We use ong>theong> ong>effectong>ive reservoir volume ong>ofong> 0 extension

as an internal stong>andong>ard to normalize all three cases

with varied extension length. The normalized ong>effectong>ive

reservoir volumes for ong>currentong> densities ong>ofong> 0.3–0.5 MA/

cm 2 are plotted with respect to ong>theong> extension length in

Fig. 5(b). It is seen that for 0 ong>andong> 60 nm M2 extensions,

ong>theong> ong>effectong>ive volume serving as void accumulation reservoir

increases with increasing extension lengths at


Z.H. Gan et al.: ong>Reservoirong> ong>effectong> ong>andong> ong>theong> ong>roleong> ong>ofong> ong>lowong> ong>currentong> ong>densityong> ong>regionsong> on electromigration lifetimes in copper interconnects

all ong>currentong> densities. However, this trend may change for

ong>theong> 120 nm extension [Fig. 5(b)] because ong>ofong> ong>currentong>

crowding around ong>theong> via region evident in Fig. 4. It may

thus be summarized that increasing extension length

from 0 to 60 nm increases ong>effectong>ive reservoir volume,

thus increasing ong>theong> void volume that leads to via opening

ong>andong> improved EM lifetimes. However, furong>theong>r increase in

M2 extension to 120 nm does not increase this ong>effectong>ive

reservoir volume ong>andong> does not significantly improve EM

lifetimes any furong>theong>r. This concept ong>ofong> ong>effectong>ive reservoir

volume successfully explains ong>theong> experimental observations

that MTTF was doubled with extension increase

from 0 to 60 nm whereas ong>theong>re was arguably a ∼5–10%

enhancement on MTTF as ong>theong> extension length was furong>theong>r

increased from 60 to 120 nm. The data analysis thus

indicates that 60 nm may be ong>theong> critical extension length

beyond which increasing extension sizes have no ong>effectong>

on EM lifetimes. The ong>effectong>ive reservoir volume plot

[Fig. 5(b)] also suggests that ong>theong> critical ong>currentong> ong>densityong>,

beong>lowong> which significant void migration does not take

place, may be on ong>theong> order ong>ofong> 0.4 MA/cm 2 .


In Cu dual-damascene interconnects, we propose that

ong>theong>re is a ong>lowong> ong>currentong> ong>densityong> zone in ong>theong> M2 extension

corners above ong>theong> cathode, which retards ong>theong> void migration

into ong>theong> extensions. Accordingly, only part ong>ofong> ong>theong>

extension volume could serve as an ong>effectong>ive reservoir

for void accumulation. Based on ong>theong> present analysis, a

60-nm extension appears to be ong>theong> critical extension

length, beyond which increasing extension sizes have no

ong>effectong> on EM lifetimes.

The implications ong>ofong> this retardation ong>ofong> void movement

caused by ong>lowong> ong>currentong> ong>densityong> ong>regionsong> is wide ranging.

Similar reservoirs (or reservoir like ong>effectong>s) are observed

in interconnect tree structures 17 ong>andong> in redundant double

via structures. 11


J. Mater. Res., Vol. 21, No. 9, Sep 2006

1. C.K. Hu, R. Rosenberg, ong>andong> K.L. Lee: Electromigration path in

Cu thin-film lines. Appl. Phys. Lett. 74, 2945 (1999).

2. J.R. Lloyd ong>andong> J.J. Clement: Electromigration in copper conductors.

Thin Solid Films 262, 135 (1995).

3. C.K. Hu, L. Gignac, R. Rosenberg, E. Liniger, J. Rubino,

C. Sambucetti, A. Domenicucci, X. Chen, ong>andong> A.K. Stamper:

Reduced electromigration ong>ofong> Cu wires by surface coating. Appl.

Phys. Lett. 81, 1782 (2002).

4. C.K. Hu, L. Gignac, E. Liniger, B. Herbst, D.L. Rath, S.T. Chen,

S. Kaldor, A. Simon, ong>andong> W.T. Tseng: Comparison ong>ofong> Cu electromigration

lifetime in Cu interconnects coated with various

caps. Appl. Phys. Lett. 83, 869 (2003).

5. Y. Shacham-Diamong>andong> ong>andong> S. Lopatin: High aspect ratio quartermicron

electong>roleong>ss copper integrated technology: Invited lecture.

Microelectron. Eng. 37–38, 77 (1997).

Z.H. Gan et al.: ong>Reservoirong> ong>effectong> ong>andong> ong>theong> ong>roleong> ong>ofong> ong>lowong> ong>currentong> ong>densityong> ong>regionsong> on electromigration lifetimes in copper interconnects

6. A. von Glasow, A.H. Fischer, D. Bunel, G. Friese, A. Hausmann,

O. Heitzsch, M. Hommel, J. Kriz, S. Penka, P. Raffin, C. Robin,

H.P. Sperlich, F. Ungar, ong>andong> A.E. Zitzelsberger: The influence ong>ofong>

ong>theong> SiN cap process on ong>theong> electromigration ong>andong> stressvoiding

performance ong>ofong> dual damascene Cu interconnects in Proc. 41st

Annual Int. Rel. Phy. Symp. (IEEE, Piscataway, NJ, 2003),

p. 146.

7. K.N. Tu, C.C. Yeh, C.Y. Liu, ong>andong> C. Chen: Effect ong>ofong> ong>currentong>

crowding on vacancy diffusion ong>andong> void formation in electromigration.

Appl. Phys. Lett. 76, 988 (2000).

8. Y.B. Park ong>andong> I.S. Jeon: Effects ong>ofong> mechanical stress at no ong>currentong>

stressed area on electromigration reliability ong>ofong> multilevel interconnects.

Microelectron. Eng. 71, 76 (2004).

9. I.S. Jeon ong>andong> Y.B. Park: Analysis ong>ofong> ong>theong> reservoir ong>effectong> on electromigration

reliability. Microelectron. Reliab. 44, 917 (2004).

10. A.V. Vairagar, S.G. Mhaisalkar, A. Krishnamoorthy, K.N. Tu,

A.M. Gusak, M.A. Meyer, ong>andong> E. Zschech: In situ observation ong>ofong>

electromigration-induced void migration in dual-damascene Cu

interconnect structures. Appl. Phys. Lett. 85, 2502 (2004).

11. C.L. Gan, C.V. Thompson, K.L. Pey, W.K. Chio, H.L. Tay, B. Yu,

ong>andong> M.K. Radhakrishnan: Effect ong>ofong> ong>currentong> direction on ong>theong> lifetime

ong>ofong> different levels ong>ofong> Cu dual-damascene metallization. Appl.

Phys. Lett. 79, 4592 (2001).

12. R.G. Filippi, G.A. Biery, ong>andong> R.A. Wachnik: The electromigration

short-length ong>effectong> in Ti–AlCu–Ti metallization with tungsten

studs. J. Appl. Phys. 78, 3756 (1995).

13. P.S. Ho: Motion ong>ofong> inclusion induced by a direct ong>currentong> ong>andong> a

temperature gradient. J. Appl. Phys. 41, 64 (1970).

14. V.P. Zhdanov: Elementary Physicochemical Processes on Solid

Surfaces, (Plenum Press, New York, 1991), p. 52.

15. M.C. Gutzwiller: Dislocations ong>andong> elections in metals, in Atomic

ong>andong> Electronic Structure ong>ofong> Metals, edited by J.J. Gilman ong>andong>

W.A. Tiller (ASM, Metals Park, OH, 1966), p. 231.

16. R.O. Simmons ong>andong> R.W. Balluffi: Measurement ong>ofong> equilibrium

concentrations ong>ofong> vacancies in copper. Phys. Rev. 129, 1533


17. A.V. Vairagar, S.G. Mhaisalkar, A. Krishnamoorthy, M.A. Meyer,

E. Zschech, K.N. Tu, ong>andong> A.M. Gusak: Direct evidence ong>ofong> electromigration

failure mechanism in dual-damascene Cu interconnect

tree structures. Appl. Phys. Lett. 87, 081909 (2005).

J. Mater. Res., Vol. 21, No. 9, Sep 2006 2245

More magazines by this user
Similar magazines