Electromigration Testing on Via Line Structures with a SWEAT Method in

comparision to standard package level tests Anke Zitzelsberger, Andreas Pietsch, Jochen v. Hagen Infineon Technologies AG, Otto-Hahn-Ring 6, D-81739 Munich, Germany

Phone: + 49 89 234 46733, Fax: + 49 89 234 45822, E-mail: anke.zitzelsberger@infineon.com

PLb PLC PLd

Abstract:

250 2 250 3 225 1

Two electromigration test methods are compared with the aim to find out if highly accelerated tests can be used to guan* the relibilty of metallization. Experimental results show a good correlation between wafer level (WL) SWEAT-and conventional standard package level (PL) tests on via terminated structures. Similar tSON values are obtained extrapolating the data from both tests to norm conditions. Failure analysis show the same failure mechanism for both tests. We found a difference in current density exponent, which can partly be explained with an error in the temperature determination of the failed metal line. As a consequence the SWEAT method can not only be used for process monitoring but also to quanw the reliability of metallization. However the difference in current density exponent needs further investigation.

Introduction:

During technology qualifications, via terminated structures are tested at moderate stress conditions on Package Level to determine the electromigration life time of the interconnect system. Up to now, highly accelerated Wafer Level tests (e.g SWEAT = Standard Wafer Level Electromigration Accele- rated Test), are used for process monitoring and mainly done on pure line structures (e.g. NIST)[1,2]. Since WLR tests can give a very rapid and cost effective charcterization of the electromigration resistance it is self evident to check wether highly accelerated tests can be used not only for monitoring but also for a more quantitative assessment of the reliability. Therefore Wafer Level Reliability [WLR] tests using an improved SWEAT algorithm [3] were performed on a via- line-structure at various test conditions. The data are compared to Package Level (PL) isocurrent tests.

Experimentals

SWEAT and PL tests were performed on a via line structure containing 3 tungsten-plugs connecting metal 1 and metal 2 (Figure 1).

Figure 1: Teststructure design

2000 IRW FINAL REPORT

The cross section and metal stack is the same for both metal layers. Line temperature is proposed to be higher in metal 2 due to a worse heat dissipation. In all cases samples for both tests were taken from the same wafer (see Figure 2), to exclude the influences of wafer to wafer variations. Stress conditions are summarized in Tablel .

Figure 2: Wafermap: black stripes tested with SWEAT, dotted stripes tested with PL

For the SWEAT test the stress is controlled by keeping the failure time (tm) constant. The line temperature (Tyne) due to Joule heating is determined via the thermal resistance @J. The linear dependence between TI,, and power dissipation through & is measured during the initial phase of each test. Here only a mean temperature can be obtained. The current is found to vary less than 1% except of the last few seconds of the test. So we specify here only the starting current (Ism), which is applied to the sample at the beginning of the test after ramping (for details of the test procedure see [1],[3]). For the PL tests the temperature increase due to Joule heating was found to be neglectable.

Tablel Test conditions

Failure distributions: Figure 3 illustrates the failure distributions of SWEAT (solid

0-7803-6392~21001S10.00 '2000 IEEE 57

Zitzelsberger et al. Electromigration Testing on Via Line Structures wi th a SWEAT Method in Comparison t o Standard Package leve l Tests

2 -

1.5 -

0 o c * . m U c * . 0 n o 0 A C 0 n o 11 A h 0" D L

n 2 2

1 -___L. .!. . : - t . . . 0.5 -

z 0 -

-0.5 -

.* . l i i: i := 0 A . O A C 0 A 0 0 a k> . D n :,

I-. . . . = . . . * .

-1.5 - O A C

- z T ? :.; : : : i n * i i : ; : , t i ; : i : : i i i i

symbols) and PL (open symbols) tests.

97

87

97

87

1.5

1

0.5

50 P[%l ~P[%ol 2 0

-0.5

1

-1.5

-2

13 13

3.4 3.4

12 B - 11

10

0

C

Figure 3 Lognormal distribution of failure times PL and SWEAT

Figure 5 Lognormal distribution of failure times PL and SWEAT as well as projection to norm condition

n =1.3

I I

The cumulated distributions appear as straight lines indicating a lognormal failure distribution and only one failure Failureanalysis: mechanism for both tests at all stress conditions. The slope of the lines is almost the same for SWEAT and PL test, so sigma is independent of the stress conditions.

Failure analysis confirmes that in principle the same failure mechanism occures for both tests (Figure 6,7).

. Ictivatron energy and current density exponent For the SWEAT tcst the paramctcrs of Blacks equation 141 activation energy Ea and current density exponent n can only bc calculated iterativly For the PL test Ea and n are determined plotting ln(tso) versus l/kT respectivly ln(tso) vcrsus 1nO). Activation cnergy was found to be Ea = 0 95eV for both tests From PL tests n was determined to n = 1 3 (Figure 4). whcreas with the SWEAT test n was found to be n = 1 9

Figure 6: PFA of both upstream vias 1&3 and metal 2 after PL test

13 i I

I I

Figure 7: PFA of both upstream vias 1&3 and metal 2 after swAT test Figure 4: Determination of current density exponent n for PL

test

For PL and SWEAT the voiding is found directly above both upstream vias (via 1 and via 3 in ~i~~~ 1) and is found

structure if the current flows from metal 1 upstream into metal 2. The W-Via acts as a diffusion barrier for the Al-Ions and

Almost identical tSoNvaheS are obtained transforming the data

use of Black's equation (Figure 5). to conditions (T = 120°C and j = 0 .5wcm2) , with the in metal 2. This is the expected failure location for this test

2000 IRW FINAL REPORT 58

Electromigration Testing on Via Line Structures wi th a SWEAT Method in Comparison t o Standard Package Level Tests Zitzelsberger et al.

the line temperature in M2 is higher due to a worse heat dissipation. Additionally the M1 line is very short ( l o w ) so no failure in M1 is expected due to the Blech effect.

Discussion

An excellent agreement of t50~ values is found using the individually determined current density exponents for extrapolating to norm conditions. However the difference in current density exponent needs further investigation. It is often discussed in literature that the current density exponent is dependent on the stress conditions [5,6,7], which would indicate a different kinetic of the degradation mechanism at high accelerations. On the other hand one should first have a look on the various inaccuracies which can lead to an error in the determined parameters that are e.g. :

inaccuracies in the determination Tlke of M2 due to worse heat dissipation in M2, TI, is only an average temperature inaccuracy in j, due to variations in width and thickness of the line. This causes an additional inaccuracy in the line temperature due the different Joule heating and is therefore more critical for the SWEAT test.

The possibility of thermal gradients are neclected in this case since no current density exponent values of n > 2 are found. We will focus here on the exact determination of the temperature of the failed line. To get an idea of the “ r d line temperature we used the finite element method to simulate the temperature distribution in the teststructure.

Determination of Tline of M2 via simulation Figure 8 shows the simulation of the temperature distribution of a part of the test structure. As expected the line temperature in M2 is higher due to worse heat dissipation. We found a AT of approx. 5O-6O0C depending on the used current density. This means that the average temperature which was determined via the Rth is approx. 10°C to low. It is worth mentioning that no (or only a neglectable) temperature gradient is found in M2.

Approx. 27OOC

Figure 8: Simulation of the the temperature distribution of a part of the test structure

2000 IRW FINAL REPORT

Extrapolation with “new n With this temperature correction a new current density exponent of n = 1.5 is obtained for the SWEAT tests. This value is very close to the one for PL. With the corrected n the deviation of the extrapolated t50N values between the two tests is smaller than one decade. Since we extrapolate over almost 7-10 decades this is still a reasonable result. ,

Conclusion/ Summary

A good correlation between conventional PL and SWEAT test is found not only for pure line structures but also for more product relevant via terminated structures. We found almost the same sigmas for both test types. Failure analysis shows the same failure location. Using Black’s equation similar t50N values are obtained extrapolating the data from both tests to norm conditions. The uncertainty of the line temperature determination is one reason for the differences in the current density exponent. With temperature simulations we estimate that the temperature has to be corrected by values of up to 10OC. With this the current density exponent for SWEAT tests becomes smaller (n= 1.5) and is very close to the value for PL. As a consequence the SWEAT method can not only be used for monitoring but also to quantify the reliability of metallization. The uncertainty in the value of n leads to a deviation of the extrapolated t50N values in the range of maximum one decade. Since we extrapolate over almost 7-10 decades this is still a reasonable result. SWEAT tests can be used for quantification of EM performance. However, the extrapolation parameters must be carefully determined and line temperature must be determined accurately. One should mention that the accelleration (in this case the stress current) should not be increased further. Thermal gradients at the line ends could also induce a change of the kinetics and therefore of the parameters of Black’s equation. If values for n much higher than 2 are obtained the data should not be used for life time extrapolation.

The stress conditions should be choosen in dependence on what to conclude from the data. It should be carefidly distinguish between a very fast test (tF< 100sec) that can be used for process monitoring and more moderate test (e.g.: 100<tF<2000sec) whch can be used for the electromigration life time prediction.

Future work: Although the data analysis looked very promising one has to improve the fast WLR testing. Therefore it is nessary to have a

0 better temperature simulation

more statistics optimization of test structures

detailed determination of n (small j intervals)

59

Zitzelsberger et al.

References:

Electromigration Testing on Via Line Structures with a SWEAT Method in Comparison to Standard Package Level Tests

QUESTIONS &ANSWERS

JEDEC Publication 1 19, A Procedure for Executing SWEAT, Sept. 1994, Solid State Products Engineering Council, 250 Wilson Blvd. Arlington, VA 2220 1-3824 S. Menon , J . Fazekas , J .v. Hagen , L. Head , C.H. Ellenwood, H. Schafft, Impact of Test-Structure De- sign and Test Methods for Electromigration Testing, I999 IEEE International Integrated Reliability Workshop Final Report, pp. 46-53, IEEE Catalog No. 99TH8460 J.v Hagen, G. Antonin, J. Fazekas, L. Head and H. Schafft, New SWEAT Method for Fast, Accurate and Stable Electromigrafion Testing on Wafer Level., submitted for 1RW2000. J.R. Black, IEEE Trans. Electron Devices; ED-16, 1969,

J.R. Lloyd, ESREF '97 tutorial notes S. Foley, A. Scorzoni , R. Balboni , M. Impronta, I . deMunari, A. Mathewsson and F. Fantini, A compari- son between normally and highly accelerated electromigration tests, 1998, Microelectronics Reliabil- ity 38, 1021-1027. F.W. d'Heurle, P.S. Ho, Electromigration in thin films 1978, in: Thin Films - Interdiffusion and Reactions, Ed.: J.M. Poate, K.N. Tu, J.W. Meyer, John Wiley & Sons, N.Y.

p338.

A: Q:

A: Q:

A:

Q:

A:

Q:

A:

Q: A:

No. the only visible voids were at the vias. Concerning the current density exponent, have you used 1.3 for the package-level results and 1.9 for the SWEAT-test results to do the extrapolation ? Yes. we did. The simulation showed a temperature difference of about 60°C. Why did you only add 10°C for the extrapolation. The temperature senses of the structure are at the very end. So you can only measure an average temperature between metal 1 and metal 2. If you take care about the different lengths of metal I , 1 Opm and 5optn. compared to metal 2. and you determine the average temperature, you will get a difference of 10°C. Why did you get a difference in the current density exponent between 1.3 and 1.5 ? This is what we will need to do for our future work. to get a better understanding on the different values. Did you recalculate the activation energy, when you determined the new current density exponent n for the SWEAT-test because of the temperature correction ? Yes. we determined a slightly higher activation energy of Ea = 0.98eV. What is the temperature in the vias ? It is quite close to the metal 2's temperature.

60 2000 IRW FINAL REPORT