Effects of Width Scaling, Length Scaling, and Layout Variation on

Electromigrationin in Dual Damascene Copper Interconnects

M. H. Lin1,2, K. P. Chang1, K. C. Su1 and Tahui Wang2 senior member, IEEE
1
United Microelectronics Corp., No. 3, Li-Hsin Rd. II, Science-Based Industrial Park, Hsinchu City, Taiwan
2
Department of Electronics Engineering, National Chiao-Tung University, Hsinchu City, Taiwan
Tel: 886-3-5789158 Ext. 33923; Fax: 886-3-578-2965; E-mail: [email protected]

1. Introduction and then slightly decrease to a constant value in Fig. 4. Since

Once the Cu diffusion along the top surface is sufficiently thickness of the electroplated Cu file before CMP is 1Pm, the
slowed, grain structures or interface boundary should affect the grain size is limited by deposited Cu thickness rather than by
reliability. Effects of width scaling were evaluated after Cu/cap metal width. MTF as a function of linewidth is plotted in Fig. 5.
process optimum. Short-line immortality was examined under MTF slightly increases with linewidth w but reaches a maximum
extremely long time testing. at w ~ 1Pm and then sharply decreases to a minimum at w =3.5Pm.
2. Experiment As w > 3.5Pm, the MTF slightly increases to a constant value. The
The EM tests were carried out at the package level. Testing observed behavior of width dependency is explained below. For
was conducted typically at 300oC with a current density of polycrystalline line structure, the drift velocity can be written as
1.2~3.0 MA/cm2. The EM failure criterion was defined as 10% [2]:
increase in resistance or when extrusion monitor current exceeds
1PA. Figs. 1(a)-(b) are schematic diagrams of three-level inter- Vd=[(GGB/d)(1-d/w)D0GBe(-QGB/kT)ZGB*+GS(2/w+1/h)D0Se(-QS/kT)ZS*]eUj/kT
connect structures discussed in this paper. Samples were fabri- (1) (2)
cated using fully metal deposition up to 9 layers to simulate real Where the subscript GB refers to the grain boundary; GGB denotes
product environment. The downstream case is defined according the width of grain boundary; d is the grain size; and h is the
to the electrons flowing from the upper wide metal line over the thickness. For part (2) large w range, (2/w+1/h)~1/h, surface dif-
via into the narrow metal line below. fusion velocities are almost the same for large w regions. However,
3. Results and Discussion for part (1) w=1Pm, the grain size d ~ 1Pm, the grain boundary
3.1 Various widths of EM test structures were designed to diffusion term is neglected. So MTF reaches a maximum at w ~
evaluate the EM performance. The median time to failure (MTF) 1Pm. Since d is almost the same for the w>1Pm range, as w in-
was measured at different line widths, and the results are plotted creases, the grain boundary diffusion impact will increase. MTF
in Fig. 2(a). MTF decreases sharply as the line width increases in sharply decreases to a minimum at w =3.5Pm. Activation energy
the range below 0.42Pm. Voids, which grow at the bottom of a via, (Ea) versus width in Fig.6 further supports the hypothesis. Lower
cause the line to fail [1]. Due to large current crowding, and the Ea value for w=1Pm, imply that the grain boundary term contri-
lack of liner redundancy, design rules that allow maximum width bution is neglected. When w > 5Pm, dielectric slots are automati-
with a single via show the worst MTF. For 0.42Pm wide lines, cally generated to prevent CMP dishing; EM induced atom diffu-
three different via arrangements were designed, i.e. a rectangular sion will be scattered by dielectric slot. MTF will slightly increase
via, a dual via-row, and a dual via-column. MTF increased more for w=6Pm. EM behavior is absolutely different from w < 5Pm.
than three times compared with a single via structure. The varia-
tion of MTF for different dual via designs is shown in Fig. 2(b). It 3.2 Electromigration critical length effect was evaluated using
is obvious that for wide lines with a dual via-row, which are along various lengths of test structures. The cumulative lifetime data for
the line length, the MTF is the longest due to the requirement for a lines 400Pm to 50Pm in length were evaluated. The TTF distribu-
large volume of Cu depletion. The MTF of the rectangular via tion at 50Pm is wider than that of the others. The dispersion of
decreases around 20% compared to that of a dual via-row. For a TTF was supposed to be the influence of back-stress force. The
0.14Pm narrow line, the MTF of the dual via-row is around 1.6 values of V in the log-normal distribution versus jL2 product are
times that of a single via. In comparison with a 0.42Pm wide line, plotted. Lower jL2 product shows large V values because of
the MTF of a dual via-row increases to 4.6 times that of a 0.42Pm back-stress-induced TTF dispersion. In order to avoid the Blech
wide line with single via. It is interesting that for the structure effect, the EM test structure length should be longer than 200Pm.
with dual via design, the MTF of 0.42Pm is very close to that of A relative resistance versus time plot, shown in Fig. 7, further
0.14Pm. The cases for a 0.42Pm wide line, in which a single via indicates that TTF dispersion is enhanced by back-stress force at
limits the EM performance, are eliminated. The cumulative life- lengths below 50Pm. It should be noticed that the time of resis-
time data for the 1Pm to 3.5Pm wide lines show the values of V, tance onset (e.g. R increase of 1%) decreases as L decreases. A
even in a wider line, are all about 0.3. The failure mode remains substantial back-stress force is developed quickly in a short length,
unchanged. An SEM image for a 1.68Pm failure sample is shown and then it suppresses resistance increase. The relative resistance
in Fig. 3. The void shape indicates that the void probably started change is saturated around 103% for L=25Pm. We observed two
at the first via, and then grew up to reach the second via. No voids modes of resistance evolution. The percentage of abrupt
formed directly under the via bottom, such as a very thin slit void, open-circuit failure was found to decrease as length decrease. As
were found. It seems reasonable to assume that no via-limited length decrease to 25Pm, most of all samples show that the resis-
issue degrades EM lifetime. The TCR value was found to increase tance gradually increases on time evolution. Extremely long test-
monotonically with linewidth but to reach a maximum at w ~ 1Pm ing time up to 5100 hrs show that the immortal behavior occurs at
0-7803-9498-4/06/$20.00 ©2006 IEEE IEEE 06CH37728 44th Annual International Reliability
671 Physics Symposium, San Jose, 2006
 L=10Pm under testing conditions; the resistance change is very small
around 2 to 4%, as shown in Fig. 8. To determine the threshold–length
2
product (jL) value precisely, Fig. 9 plots the reciprocal of the MTF 1.8
1.6
versus current density. A linear relationship between (jL) and 1/MTF 1.4

Ea(eV.)
is assumed, as indicated by the linear extrapolation to zero. The criti- 1.2
1
cal current jc was calculated to be 2.18 MA/cm2 for 250 oC. The cur- 0.8
0.6
rent exponent n value in the Black equation was found to be 3.6 for 0.5 1 1.5 2 2.5 3 3.5 4

50Pm. The activation energy was found to be 1.0 for 50Pm, which Linewidth w(Pm)

indicates no early failures affecting the result. The temperature de-

pendence of threshold–length product for 50Pm and 25Pm is shown in FIG 6. Plot of activation energy (Ea) as a func-
Fig. 10. The threshold–length product (jL) value at 300oC was deter- tion of width. Lower Ea value for w=1Pm,
imply that the grain boundary term contribution
mined to be 4300 A/cm based on 25Pm data. For L=25Pm, testing is neglected.
time up to 3100hr show small resistance change under testing condi-

Relative R Change (%)

tions. The jL product at 300 oC is 4000 A/cm at 25Pm further support 103
25um
this threshold–length extraction method. The maximum stress differ- 102
200&100um
ence in interconnects for void nucleation was determined to be 400 101
50um
MPa at 250 oC. 100

4. Conclusion 99

The effects of width scaling, length scaling, and configuration of 1 10 100 1000

Stress Time (a.u.)

via layout were evaluated. Two scenarios for width scaling were
shown. A theory was proposed to explain the observed behavior. FIG 7. Relative R change vs. time plots for
Blech effect in a dual damascene process and its temperature depend- L=25, 50, 100 and 200Pm. Different resis-
ence were demonstrated under long time testing. tance onset times are shown for various
lengths. Stress conditions are the same.
References
[1] M. H. Lin et al., in proc., IRPS., pp.229-233, 2004.
[2] C.-K. Hu, et al., in proc., AMC., 1999.

Relative R Change (%)

110
Down-stream
L=10Pm, j=1.6 MA/cm2
e- 105
Up-stream
M3

e- M2 100
M1
L 95
1 10 100 1000 10000
(a)
Stress Time (hrs)
single via

FIG 8. Relative R change vs. time shows slight

Dual via -
column
FIG 3. SEM image for w=1.68Pm wide line resistance increase up to 5100 hrs of testing for
structure. Void was found in via bottom. L=10Pm, j=1.6 MA/cm2, T=300C, and sample
Dual via -
row Large trench void cross other via is depleted. size over 24. jL|1600 A/cm.
Rectangular
3.75 0.03
via
(b) 3.65 50um-300C
1/MTF (a.u.)
TCR [10-3 oC-1]

3.55 0.02 50um-250C

FIG 1. (a) Via terminated line structure. The stripe 3.45
50um-250C
length L is 5, 10, 25, 50, 100, 200, or 3.35 0.01
400Pm.Downstream is evaluated. (b) Various EM 3.25 Max
structures for width dependence, left w=0.14, mid- 3.15
Min
0
dle w=0.42, right w>0.42Pm L=400Pm.
75th %
25th % 0 1 2 3 4
3.05
0.14 0.3 0.36 0.42 1 1.68 2.1 3.5 6 Median
Current Density (MA/cm2)
˄˃˃˃ Linewidth w (Pm)

2via FIG 9. Reciprocal MTF as a function of current den-

2via
FIG 4. Plot of TCR as a function of line width. It sity for L=50Pm. The critical current jc was calculated
MTF(a.u.)

˄˃˃ 1via reaches a maximum at w ~ 1Pm and then slightly to be 1.3 MA/cm2 for 300C, 2.05 MA/cm2 for 275C,
decreases to a constant value. 2.18 MA/cm2 for 250C. Error bars represent 90%
1via
confidence intervals.
1000
Max stress difference (Mpa)

˄˃ L=25Pm  'V(L=25Pm)
MTF(a.u.)

0.1 1
 L=50Pm 'V(L=25Pm)
Linewidth w(Pm) †

12000 1000
100
jLc (A/cm)

(a) 10000
8000 700
1000 1000
6000
4000 400
MTF(a.u.)

MTF(a.u.)

10 2000
0 100
0.1 1 10
200 250 300 350
Linewidth w(Pm)
100
single via rectagular via 2 via-row
100
2 via-column rectagular via 2 via-row
Temperature(oC)
w=0.14Pm w=0.42Pm
(b)
FIG 5. Plot of MTF as a function of linewidth. 90% FIG 10. Temperature dependence of critical length
FIG 2. (a): Plot of MTF as a function of linewidth. confidence levels are lower and upper bounds. For effect and maximum stress difference in the inter-
90% confidence levels are lower and upper bound. w>0.42Pm, sufficient vias are designed to relieve the connect for L=25, 50Pm. It is observed that jL value
(b):2 via cases for 0.14 and 0.42Pm, stress condi- via-limited issue. Stress condition is the same. increase slowly at low temperature.
tion is the same.

672