Gupta D. (Ed.). Diffusion Processes in Advanced Technological Materials

Подождите немного. Документ загружается.

Figure 9.48 is a FIB image of the 0.21-mm-wide M2 line from the short-

lifetime group; it shows a void at V1M2 (open fail). The failure time of

64 hours at 298°C in the short-lifetime group is consistent with the time

to grow a 0.3-mm-long void at the cathode end of the 0.21-mm-wide line

with a blocking boundary as measured in CuSiO

structures. However, in

the case of the long-lifetime group, the voids were not observed at either

end of the 0.21-mm-wide M2 in most of the samples. No visible voids at

the V1 viaM1 interface are seen, as shown in Fig. 9.49(a), even though

the line was stressed for 585 hours. However, a void in the 3-mm-wide M1

short line is shown in Fig. 9.49(b). Voids in the 3-mm-wide M2 lines are

also often observed. For an electromigration drift velocity (Cu mass

migration rate) at 298°C of 0.006 mmh for the 0.21-mm-wide line, after

a lifetime of 600 hours, we would expect to see a 3.5-mm-void in the fine

M2 test line of Fig. 9.49(a) if the liner at the V1M1 interface were a

blocking boundary. In fact, no voids were observed near the M1V1

viaM2 interfaces in most of the tested samples. This result suggests that

the migration of Cu in the fine M2 line is replenished by the wide M1 line

through the V1 via. However, voids were found in the 3-mm-wide M1 and

M2, and some were observed at the end of the M1 section where there was

little or no electrical field, as shown in Fig. 9.49(b). Similar behavior has

been reported in AlSiO

[104, 117, 118]

and CuSiO

[40, 45]

During electromi-

gration, the Cu atoms will drift along the CuSiN

interface (the top Cu

surface), and the depletion of atoms is responsible for creating the voids.

Thus, voids would be expected to occur at the cathode ends of the fine M2

lines. However, a large portion of the tested samples showed no voids at

the fine M2V1 via interface in the long-lifetime group. The mass trans-

port of Cu toward the anode along the top M2 surface generates excess

472 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Figure 9.48 FIB image of a failed sample from the short-lifetime group, showing

an open failure at the M2/V1.

Chapter-09 11/29/04 6:51 PM Page 472

ELECTROMIGRATION IN CU THIN FILMS, HUETAL. 473

vacancies and tensile stress near the cathode. The vacancy concentration

and stress gradients generate a vacancy flux that moves to vacancy sinks

such as interfaces, grain boundaries, or the ends of the M1 and M2 lines

to form voids. Some vacancies pile up at the end of the M2 and some

migrate through to M1, resulting in short and long lifetimes, respectively.

This may be rationalized by the following discussion.

If the liners at the V1M1 interfaces were good blocking boundaries,

mass accumulation at the anode end of the M1 lines should occur. This

electromigration-induced mass accumulation would generate a stress gra-

dient in a line that can cancel out the electromigration driving force,

erj  Ω(∆sL

), where Ω, ∆s, and L

are the Cu atomic volume,

electromigration-induced stress, and critical length, respectively. No dam-

age should occur in a metal line length below L

. The estimated value of

is 20 mm for an unpassivated Cu line with j at 2.6 mAmm

(jL

530Acm).

[91]

However, the observation of voids in 12-mm-long M1 lines

with 2.6 mAmm

supports the notion that thin liners at M1V1 interfaces

Figure 9.49 FIB images of a line tested at T = 298°C for 585 hours. (a) M1/V1/M2

fine test line interface; (b) wide M2 and M1 lines.

Chapter-09 11/29/04 6:51 PM Page 473

used in the present test structure cannot withstand the incoming Cu elec-

tromigration flux and Cu migration becomes continuous through M1 to

V1. Then the failure mechanisms in this structure are similar to that of

testing a fine test line connecting two large reservoirs at both ends of the

line; that is, continuous or partial flow will enable a very long lifetime.

However, with the anode end of the line connected to a blocking

boundary, such as a thicker liner or W via, high threshold values of jL

8000 Acm,

[77]

4000 Acm,

[123]

3000 Acm,

[124]

and 7000 Acm

[125]

for

lines with electron flow to a blocking boundary have been obtained.

Electromigration in two-level Cu dual-damascene lines in a low-

dielectric-constant material (SiLK dielectric) has been discussed. The fail-

ure distribution cannot be fitted to a single log-normal distribution. At

least two distinguishable lifetime groups were observed. The short-

lifetime group is consistent with the time required to grow a line void at

the end of the 0.21-mm-wide test line. The results are also in a good agree-

ment with the lifetimes obtained from the CuSiO

system. The samples

with lifetimes in the long-lifetime group suggest that the migration of Cu

in the M2 is replenished by Cu from the M1 through the V1M1 interface.

These observed results suggest that the fundamental CuSiLK dielectric

electromigration failure modes are similar to CuSiO

and that the

observed bimodal distribution is not the result of using a low-dielectric-

constant material. A thin liner at the Cu viaCu line interface that cannot

withstand the incoming Cu fluxes can also eliminate the short-length

effect in a Cu line.

9.13 Reduced Cu Interface Diffusion

As discussed in Secs. 9.10 through 9.12, the electromigration Cu life-

time is most dependent on atomic transport at the Cu/dielectric interface.

Some investigators have, however, reported that the Cu/liner interface is

the dominant fast path. There may be significant differences in the inte-

gration processes and materials, leading to differences in the interfaces.

However, in both cases, the fraction of total Cu atoms situated on inter-

faces increases as the dimension of the interconnection is scaled down.

Accordingly, the Cu electromigration lifetime and the allowed current

density specification in Cu conductors may be reduced in every new gen-

eration of interconnect. The ability to significantly extend Cu conductor

technology to smaller dimensions can be realized by altering or modify-

ing the Cu/dielectric interface to reduce the Cu transport properties; for

example, 50-nm-thick CoP

[126]

or 100 to 200-nm-thick CoWP

[127, 128]

selectively deposited electroless films can serve as diffusion barrier lay-

ers. Recently, improved electromigration Cu lifetime was reported from

474 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Chapter-09 11/29/04 6:51 PM Page 474

ELECTROMIGRATION IN CU THIN FILMS, HUETAL. 475

testing Cu lines with a thin (10 to 20 nm) surface layer of electroless

CoWP, CoSnP, or Pd.

[129]

Figure 9.50 shows a TEM image of a 10-nm-

thick CoWP film on top of a Cu M1 line. The amorphous a-SiC

cap-

ping layer and a low-dielectric-constant (k  2.6) SiLK dielectric are

also shown. Asmooth, uniform layer of CoWP and large Cu grains were

observed. From this TEM sample, an x-ray energy dispersive spec-

troscopy (EDS) line scan was taken across the top Cu M1 surface.

Figure 9.51 shows a plot of x-ray intensity as a function of element and

probe position for the line scan. The scan started in the Cu line near the

top surface, proceeded across the CoWP and amorphous SiC

layers,

and ended in the SiLK dielectric. In Fig. 9.51, the boundaries of these

layers are labeled. The EDS line scan shows that the Cu signal extends

somewhat into the CoWPlayer, suggesting that Cu is mixed into the cap.

The Cu signal remains above background noise counts even in the amor-

phous SiC

and SiLK dielectric layers because the samples were

mounted on a membrane on a Cu grid. Spurious X-rays hitting the Cu

grid provided a background Cu signal. The CoWPgrain size was measured

Figure 9.50 TEM cross-section image of a Cu interconnect coated with CoWP

using a Fig. 9.3(d) test structure.

Chapter-09 11/29/04 6:51 PM Page 475

to be 15 to 20 nm from a series of plan-view TEM images. The RBS

analyses determined the composition of the CoWP metal cap to be

Co(3% W, 6% P). Electromigration was performed on test structures of

either single-damascene Cu lines in SiO

on W or dual-damascene Cu

lines in SiLK on W. The metal linewidths ranged from 0.1 to 2 mm. The

current densities in the via bottom and line were 120 and 35 mA/mm

respectively. Figure 9.52 shows typical dual-damascene Cu test line

resistance as a function of time for 2-mm-wide polycrystalline lines

tested with CoWPa-SiC

and a-SiC

-only thin cap layers in SiLK

dielectric at a sample temperature of 350°C. The data clearly show a

remarkable improvement in lifetime for samples with CoWP coatings.

Voids were observed in the via bottom and/or the line in dual-damascene

Cu lines with an a-SiC

-only cap. However, only via bottom voids

were observed in the case of a CoWPcap. Figure 9.53(b) shows a via bot-

tom void for a Cu dual-damascene line with a CoWP cap.

476 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Figure 9.51 Intensities of elements from EDS measurements.The electron probe

moved from the top surface of a Cu damascene line, through the CoWP and amor-

phous SiC

coating layers, and ended in the SiLK dielectric.The vertical dotted

lines indicate the boundaries between the layers.

Chapter-09 11/29/04 6:51 PM Page 476

ELECTROMIGRATION IN CU THIN FILMS, HUETAL. 477

Figure 9.54 shows an Arrehenius plot of the median lifetimes of Cu

damascene lines with various capping layers. For comparison, the data

points from the samples of 0.27-mm-wide single-damascene lines capped

with SiN

are also included in Fig. 9.54. The solid lines are the least-

squares fits. The value of electromigration activation energy for 2-mm-

wide dual-damascene CoWPcapped polycrystalline lines was found to be

1.0 ± 0.1 eV. In the case of lines with bamboo-like grains, the activation

energies of 0.90 ± 0.05 eV, 1.0 ± 0.1 eV, and 1.9 ± 0.2 eV were obtained

Figure 9.52 The resistance of a damascene Cu conductor, with and without a thin

metal film on the top surface, versus time.

Chapter-09 11/29/04 6:51 PM Page 477

for a 0.18-mm-wide a-SiC

capped dual-damascene line, 0.27-mm-wide

SiN

capped single-damascene line, and a 0.7-mm-wide CoWP capped

dual-damascene line, respectively. Significant lifetime improvement for

the samples with CoWP capping compared to SiN

or a-SiC

capping

suggests that CoWP coating significantly reduces Cu diffusion along the

478 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Figure 9.53 FIB images of Cu lines electromigration stressed with a bottom via

current density of 12  10

A/cm

and at a sample temperature of 280°C.

(a) 2.8 hours with SiC

; (b) 1100 hours with CoWP coating.

Chapter-09 11/29/04 6:51 PM Page 478

ELECTROMIGRATION IN CU THIN FILMS, HUETAL. 479

Cu/cap interface. The similar lifetimes obtained for 0.18-mm-wide

bamboo-like and 2-mm-wide polycrystalline lines with a SiC

cap

indicated that the dominant diffusion path in these lines is along the

CuSiC

interface, not grain boundaries. However, in the case of CoWP

capped samples, the measured activation energy of 1.0 eV for the 2-mm-

wide lines is in good agreement with the activation energy of Cu grain

boundary diffusion.

[81–83]

Furthermore, the data in Fig. 9.54 show a drastic

increase in Cu lifetime from the 2-mm-wide line to the 0.7-mm-wide

bamboo-like line. These results suggest CoWP capping caused a great

reduction in Cu interface diffusion and the Cu mass motion in CoWP

capped 2-mm-wide polycrystalline lines was primarily influenced by Cu

grain boundary diffusion. The 2-mm-wide CoWP line had an average

grain size d of ∼1.2 mm, while the 0.28-mm-wide silicon nitride capped

lines had a metal line thickness h of 0.25 mm. We would expect a factor

of 5 reduction in void growth rate from dh in Eq. (4) if the dominant

migration path were from the Cu/silicon nitride interface for the 0.28-mm-

wide uncapped line and from grain boundaries for the 2-mm-wide CoWP

line, even though similar activation energies were obtained in both cases.

Figure 9.54 Plot of median lifetime t

vs. 1/T for Cu lines with various capping

layers.

Chapter-09 11/29/04 6:51 PM Page 479

In addition, the existence of single-crystal grains capped with CoWP in

the 2-mm-wide line would create a blocking path for grain boundary dif-

fusion, which would further increase the lifetime for the samples with a

CoWP cap. Therefore, we would expect more than the factor of 5 in the

2-mm-wide CoWP capped line samples as compared with the 0.28-mm-

wide Cu/SiN

capped bamboo line. The observed activation energy of 1.9

eV for the bamboo-like lines capped with CoWP is close to the value of

2.07 to 2.2 eV for Cu bulk diffusion.

[72, 84]

For these data, it is not known

if the diffusion path in a bamboo-like grain line structure capped with

CoWP is along the CuTa, CuCoWP interfaces or in the bulk Cu. The

existence of voids at the bottom of the via could suggest that diffusion in

the Cu-Ta interface or in Cu bulk was dominant, but this voiding could

have been caused by top-interface Cu diffusion and vacancy migration to

the via bottom (vacancy sink).

[11]

The mechanism of reduction of the interface diffusion by substitution

of Cu/metal for CuSiN

or Cu/amorphous SiC

interfaces is not totally

understood, although it is tempting to speculate that the Cu migration is

affected by the number of interfacial defects, the interface bond strength,

and/or the surface migration energy of Cu atoms directly in contact with

the cap material. Increased improvement in electromigration resistance is

expected to translate to exceptional flexibility for the circuit designers,

effectively removing electromigration as the limiting factor for use of

high currents. The results of the tests further support the hypothesis that

the uncoated surfaces or interfaces of Cu with the dielectric are the major

sources of electromigration and thus reliability degradation.

In summary, an investigation of Cu electromigration in Cu damascene

interconnections with and without thin CoWP coatings on the top surface

of the Cu line showed that electromigration failure lifetimes can be dras-

tically improved. The diffusion of Cu at the top surface of a Cu damascene

line was greatly reduced in the samples with CoWP cap so that the Cu

electromigration lifetime was markedly improved. The activation energy

for electromigration in Cu damascene lines capped with CoWPwas found

to be 1.9 and 1.0 eV for bamboo-like and polycrystalline grain structures,

respectively, and 0.90 eV for SiC

cap-only samples.

9.14 Conclusion

This chapter has discussed electromigration in three-level and two-

level Cu single- and/or dual-damascene lines. The mass transport in Cu

interconnects occurs mainly by interface diffusion. Fast diffusion along

either the Cu/metal liner or Cu/dielectric interface has been reported.

These results suggest that the fast diffusion paths in Cu interconnects

480 DIFFUSION PROCESSES IN ADVANCED TECHNOLOGICAL MATERIALS

Chapter-09 11/29/04 6:51 PM Page 480

ELECTROMIGRATION IN CU THIN FILMS, HUETAL. 481

are very sensitive to the nature of the interface, which is dependent on

the selected fabrication materials and processes. The main observations

follow:

1. Similar electromigration activation energy and void growth

rate were found for the via bottom and the line voids.

2. Line void growth rate is related to 1/(metal line thickness).

3. Bimodal failure distributions were found for Cu dual-

damascene lines on W.

4. Via bottom void growth rate is related to the via current

density.

5. A drastic reduction of the void growth rate found for a thin

metal (CoWP, CoSnP, Pd) layer capped on Cu line surfaces

with a bulk-like activation energy for a bamboo-like Cu line

suggests that the fast diffusion in Cu lines is along the Cu/

dielectric interface rather than the Cu/Ta interface.

Primarily, two types of void growth in Cu lines are observed, grain

thinning and edge displacement, even though mass transport is along the

top surface of the line. The grain thinning phenomenon is void growth by

thinning a single-crystal Cu grain from the top surface down, one grain at

a time, thereby removing the Cu grains in layers. Edge displacement void-

ing pertains to a vertical void growth where Cu atoms from the bottom of

the line/via edge feed Cu atoms drifted away from the top interface. Thus,

the Cu lifetime distribution for via-to-line current flow will differ for the

cases of via above and via below the Cu line. These differences can be

seen particularly for the case of a via fully landed inside a line with elec-

tron flow from via to line and a blocking boundary at the bottom of the

via. Grain thinning void growth can quickly separate the via from the line,

which will then cause a sharp resistance change at early failure lifetimes.

Edge-displacement voids will take more time to grow across the line

under the via to cause failure. Therefore, in this case, edge-displacement-

void growth will have a longer lifetime than grain-thinning-void growth.

On the other hand, for a Fig. 9.3(b) test structure, grain-thinning-void

growth would have a longer lifetime if the grain in the line on top of the

W via were larger than the W via diameter. The Cu lifetime distribution is

further complicated by partial blocking boundaries at the line/via inter-

face, such as the quality and thickness of the liner. Observations of very

long electromigration lifetime have suggested that a thin liner at the

via/line interface allowed Cu atoms to diffuse from one Cu line level to

the other level, despite a Cu via bamboo microstructure.

The fraction of the total Cu atoms present at interfaces increases as the

dimensions of the Cu interconnections are scaled down, which suggests

Chapter-09 11/29/04 6:51 PM Page 481