cnse.albany.edu

Background:

Statistical Electromigration Analysis: Constant Current vs Constant Voltage Jonathan Gentile, Andrea Hubbard, Dr. Lloyd

College of Nanoscale Science and Engineering, SUNY, Albany NY, 12203

Objective:Compare and contrast statistical electromigration (EM) data retrieved from Constant Current and Constant Voltage accelerated electromigration tests in order to better predict reliability lifetimes of Cu interconnects.

Motivation

Results

Conclusion

References[1] Breu, F., Guggenbichler, S. & Wollmann, J., 2008. Grain size dependance of narrow EM failures. Vasa. Available at: http://medcontent.metapress.com/index/A65RM03P4874243N.pdf [Accessed December 5, 2013].

[2] Hau-Riege, C.S., 2004. An introduction to Cu electromigration. Microelectronics Reliability, 44(2), pp.195–205. Available at: http://linkinghub.elsevier.com/retrieve/pii/S0026271403004475 [Accessed December 5, 2013].

[3] Hauschildt, M. et al., 2006. Analysis of electromigration statistics for Cu interconnects. Applied Physics Letters, 88(21), p.211907. Available at: http://link.aip.org/link/APPLAB/v88/i21/p211907/s1&Agg=doi [Accessed December 5,

2013].

[4] Hauschildt, M., 2005. Statistical analysis of electromigration lifetimes and void evolution in copper interconnects. Available at: http://adsabs.harvard.edu/abs/2005PhDT........75H [Accessed December 5, 2013].

[5] Ho, P.S., Uehling, T. & Ramanathan, L.N., 2006. Electromigration Lifetime Statistics for Pb-Free Solder Joints with Cu and Ni UBM in Plastic Flip-Chip Packages. 56th Electronic Components and Technology Conference 2006,

pp.650–656. Available at: http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=1645719.

[6] Laboratories, S.N., 1997. E l e c t r o m i g r a t i o n : a review. , 37(7), pp.1053–1072.

[7] Li, B. et al., 2006. Threshold electromigration failure time and its statistics for Cu interconnects. Journal of Applied Physics, 100(11), p.114516. Available at: http://link.aip.org/link/JAPIAU/v100/i11/p114516/s1&Agg=doi [Accessed

December 5, 2013].

[8] Lloyd, J. & Clement, J., 1995. Electromigration in copper conductors. Thin solid films, 262, pp.135–141. Available at: http://www.sciencedirect.com/science/article/pii/0040609094058067 [Accessed December 5, 2013].

[9] Schafft, H. & Lechner, J., 1988. Statistics for electromigration testing. … 1988. 26th Annual …. Available at: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=23449 [Accessed December 5, 2013].

What is electromigration?:

• Definition: The transport of material caused

by the gradual movement of the ions in a

conductor due to the momentum transfer

between conducting electrons and diffusing

metal atoms. (Figure 1)

• Affects devices with high current densities

(Nanoelectronics).

• Typical current density in a lamp cord ~ 𝟏𝑨

𝒄𝒎𝟐

• Typical current density in copper

interconnects ~ 𝟏𝟎𝟓𝑨

𝒄𝒎𝟐

• Voids are formed typically near the via,

killing your devices. (Figure 2)

Accelerated Electromigration Testing:

• We need to be able to get 10 year lifetimes in

a matter of hours.

• In order to do that, we test specific test

structures and elevate the temperature and

voltage applied to the structure. (Figure 3)

• We take the data from test structures

containing a single failure element and we

need to relate that to the failure of a chip

with perhaps 100,000 to 1,000,000 failure

elements.

• We plot lognormal failure distributions and

extrapolate fitting parameters in order to get

useful information.

• Electromigration is the main reason of

why high performance nanoelectronic

devices fail over time.

• The reliability of nanoscale electronics

is very important to industry and

corporations to bring down RMA and

return expenses. The customer and

Manufacturer are happier as a result.

• Electromigration has been difficult to

treat in the past. This is mainly due to

the difficulties in testing large sample

populations.

1.00 1000.0010.00 100.00

1.00

5.00

10.00

50.00

99.00

ReliaSoft's Weibull++ 6.0 - www.Weibull.com

Probabil i ty - Lognormal

Time, (t)

Un

relia

bility, F

(t)

4/9/2014 12:05IBMJim Lloyd

LognormalEAKOF4

L2 MLE - SRM MED

F=11 / S=4CB[FM]@60.00%2-Sided-B [T2]

1.00 1000.0010.00 100.00

1.00

5.00

10.00

50.00

99.00

ReliaSoft's Weibull++ 6.0 - www.Weibull.com

Probabil i ty - Lognormal

Time, (t)

Un

relia

bility, F

(t)

4/9/2014 12:10IBMJim Lloyd

Lognormal27SEE0

L2 MLE - SRM MED

F=13 / S=1CB[FM]@60.00%2-Sided-B [T2]

1.00 1000.0010.00 100.00

1.00

5.00

10.00

50.00

99.00

ReliaSoft's Weibull++ 6.0 - www.Weibull.com

Probabil i ty - Lognormal

Time, (t)

Un

relia

bility, F

(t)

3/5/2014 12:45IBMJim Lloyd

Lognormal09TMG5

L2 RRX - SRM MED

F=11 / S=4CB[FM]@60.00%2-Sided-B [T2]

1.00 1000.0010.00 100.00

1.00

5.00

10.00

50.00

99.00

ReliaSoft's Weibull++ 6.0 - www.Weibull.com

Probabil i ty - Lognormal

Time, (t)

Un

relia

bility, F

(t)

3/5/2014 12:48IBMJim Lloyd

Lognormal58SJB5

L2 RRX - SRM MED

F=12 / S=4CB[FM]@60.00%2-Sided-B [T2]

10.00 1000.00100.00

1.00

5.00

10.00

50.00

99.00

ReliaSoft's Weibull++ 6.0 - www.Weibull.com

Probabil i ty - Lognormal

Time, (t)

Un

relia

bility, F

(t)

3/5/2014 12:50IBMJim Lloyd

Lognormal40TMH2

L2 RRX - SRM MED

F=10 / S=3CB[FM]@60.00%2-Sided-B [T2]

Constant Current Data: Lognormal Distributions

Constant Voltage Data: Lognormal Distributions

Lots of roughly 16 samples

were run under accelerated

testing conditions on 5

different wafers using

Constant Current and

Constant Voltage methods.

CC CV

Temperature 300°C 300°C

Voltage ~ 200mV

Current 40µA ~

Accelerated Test Conditions:

Current Situation and Methods

• Traditionally Electromigration (EM) testing is performed

using constant current.

• However, variations in geometry (line width, thickness,

cross section) will produce corresponding variations in

current density if the same current is passed through each

structure. This contributes to the variation in lifetime,

• For a given length of conductor, the electric field along the

conductor will be identical at a given applied voltage

regardless of the cross sectional area.

• Although cross sectional area can vary significantly across

a wafer (thickness and conductor width) the length

proportionately does not.

– Therefore, at constant voltage the current density will be

constant regardless of these inherent variations.

Figure 1: Electromigration concept diagram

Figure 2: Electromigration void nucleation and SEM images

Figure 3: Test structure and wirebonding diagrams

• Constant Current vs Constant Voltage shows some clear differences.

• The standard deviation, which represents the spread of failure times, is showing a decrease in value for all

of our Constant Voltage tests up to a factor of 2.

• Most T-50’s that we’ve recorded, which is the median time to failure, have been within an order of magnitude

difference. We can note that constant voltage on average seems to take less time to fail. This could be due to

under-stressing the constant current tests.

JGEM03 JGEM04 JGEM05 JGEM06 JGEM07

T50 (h) σ

Test # Wafer # CC CV CC CV

JGEM03 09TMG5 100.5847 69.17434 1.0762 0.574278

JGEM04 58SJB5 195.6251 59.98551 0.9617 0.661062

JGEM05 40TMH2 111.8994 87.58686 0.6683 0.582221

JGEM06 EAK0F4 37.13277 38.96727 0.9215 0.494421

JGEM07 27SEE0 38.34024 55.49667 1.0995 0.649492