Stress Control in Multi-Layer Backside Metalization of Thinned Wafers

		At a Glance
	Modern trends in microelectronics require advanced backside metallization processes that work on thin and extremely thin substrates. This article describes some new approaches, using the Sputtered Films, Inc (SFI) Endeavor AT cluster PVD tool. Our goal was to develop a dc magnetron sputtering process which would allow the deposition of a low stress, three metal stack of titanium (3000 Å), nickel-vanadium (3000 Å), and silver (18500 Å) on the back sides of 200 mm diameter silicon wafers thinned to 200 m m. An inter-layer stress compensation approach was used. The stack, consisting of films strained in opposite directions, had residual stress (after self-relaxation period) of less than 2E8 dynes/cm2. This enabled high-quality metallization with very low wafer warpage.

The Semiconductor Industry needs to design new processes that can precisely control residual stresses in thin films. For example, stress control in backside and under bump metallurgy is very critical to reliable die attachment [1]. In applications on thinned wafers, low or even zero stress films are required to avoid 'potato-chipping' of the wafers.

The majority of vacuum deposited metal and alloy films have tensile stress unless special stress control methods are used. Highly efficient stress control methods developed in the last several years with magnetron sputtering systems allow creating plasma or low energy ion bombardment conditions on the substrate during film growth. Depending on the film material, different approaches are necessary to ensure an effective stress reduction. Some of the modern stress control methods are described in [2].

In backside metallization processes, multi-layer metal film stacks are typically employed, for instance, a stack of titanium (Ti), nickel-vanadium (Ni-V), and silver (Ag) films. Ti and Ni-V are materials in which tensile stress can be decreased significantly or even converted to compressive stress by means of deposition with a negative substrate bias. At the same time, magnetron sputtered Ag is a very difficult material for stress control. It is possible to utilize the good stress controllability of some materials in a multi-layer stack, to compensate for the influence of other layers upon the total residual stress in the stack.

Therefore, we created a low stress sandwich consisting of films strained in opposite directions. This inter-layer stress compensation approach has been successfully employed in previous SFI work on 150 mm-diameter wafers. For larger size, thinned wafers, fine tuning the stress is more complicated. Here, one of the most important issues is the ability to ensure high stress uniformity. An erratic stress distribution in any layer can lead to the disturbance of the inter-layer stress compensation principle. As a result, the wafer can become essentially non-flat or even be curled up.

Our goal was to develop a dc magnetron sputtering process which would allow the deposition of a low stress, three metal stack of Ti (3000 Å), Ni-V (3000 Å), and Ag (18500 Å) on the back sides of 200 mm diameter silicon wafers thinned to 200 m m.

All deposition experiments were done in the Endeavor AT cluster PVD tool equipped with series IV S-Guns [3]. The series IV S-Gun is a cylindrical dual concentric ring cathode sputter source. The two independently controlled cathodes (targets) and a central bias-able anode allow depositing highly uniform films. RF power can be applied to the substrate holder creating a negative bias on the wafer. By varying the substrate bias power it is possible to influence film properties, such as morphology and stress.

We used standard 0.7 mm thick non-doped silicon (Si) wafers for stress investigation and process development. The process parameters were then adjusted for wafers thinned to 0.35 and 0.2 mm. Film thickness was measured using a Dektak-3ST Surface Profiler;

thickness uniformity was evaluated with a CDE ResMap Automatic 4 Point Probe. For a quality evaluation of the stack adhesion a simple scratch test was employed. Wafer curvature radius and height measurements were taken by an FSM 128 Thin Film Stress and Flatness System.

Investigation results and discussion

Technology of pre-deposition surface treatment

In accordance with specifications, the technological process included wafer etching, which removed 50 Å before metallization. RF plasma etching was performed using different levels of RF power (300-700W) and argon gas flow (5-15 sccm). It was found that the three-metal stacks had poor adhesion; at the same time, the titanium layer itself had good adhesion. An additional wafer degas (300C for 40 seconds) followed by etching did not prevent stack delamination in the scratch test. The most probable reason for this is the dilution of the substrate-coating bonds conditioned by residual tensile stresses in silver.

We then employed a high-temperature wafer pre-heat followed immediately by Ti sputtering in the same process module to stimulate a diffusion interaction between first film and substrate material. Experiments have shown that for the standard 0.7-mm thick wafers, the thermal treatment at 400C ensured an excellent cohesion.

Titanium films

The process module for Ti deposition had an etch configuration which enabled extremely deep stress control possibilities for that material. Fig. 1 shows that Ti films become compressive with the application of a comparatively low RF bias power (40W) and can reach a stress level of about -1E10 dynes/cm² with increasing power. Stress in the Ti film does not depend as much on argon gas flow and deposition temperature as some other materials. (For example, Ni-V films demonstrate significantly higher tensile stress for higher temperature and gas flow.)

With a goal to ensure a certain compressive stress value, we split the Ti film deposition into two steps: (1) the 'hot' step, which improved adhesion properties, and (2) the 'cold' step, which gave an overall deeper compressive level. These two sub-layers, when deposited on standard 0.7-mm thick wafers, had the following properties:

1. The 'hot' layer, deposited on Si wafer with pre-heat 400 C for 50 seconds:
   stress = -8E8 dynes/cm²,
   bulk resistance = 75 m Ohm-cm;
2. The 'cold' layer, deposited on Ti under-layer without heat:
   stress = -(6-8)E9 dynes/cm² ,
   bulk resistance = 96 m Ohm-cm.

This data corresponds to deposition conditions, optimized to get 3000 Å film having thickness uniformity 3s < 5% and total stress = . 3E9 dynes/cm².

Fig. 1 Stress in Ti films vs. substrate RF bias power.

Nickel-vanadium films

The nickel-vanadium 3500Å films were deposited in a process module with a standard series IV S-Gun. Distinguishing differences in stress control were observed between the 200-mm wafers and the smaller size wafers during the process investigation.

The films deposited on 100-mm wafers (with Ti under-layer) were under high tensile stress if no substrate bias was used. Stress magnitude continuously diminished in the

range from tensile +1E10 to compressive -5E9 dynes/cm² with RF power varying in the range 0-200W (Fig. 2), then, stress reached its saturation limit and didn. t change significantly with applied RF power up to 500W. A point of stress conversion from tensile to compressive on the stress versus RF power curve moved in higher RF power direction with increased argon gas flow. The same recipes, employed for 200-mm wafer deposition, did not enable the stress conversion. Moreover, tensile stress increased after it reached a minimum value at RF power of about 300W (Fig. 2).

Fig. 2 Stress in NiV films vs. substrate RF bias power.

We found that low cathode power and low argon gas flow are necessary conditions for low stress Ni-V film formation on 200-mm wafers (see Table 1). Note that the data for 200-mm wafers in Fig. 2 corresponds to those optimal sputtering conditions; a minimal stress point is higher if higher values of power and gas flow are used.

Stress has essentially non-uniform distribution on the 200-mm wafer surface; it has more tensile value in the center (the opposite is true for smaller wafers). For example, the film (last sample in Table 1) had an average compressive stress of -2.7E8 dynes/cm², but the stress range ran from compressive -1.2e9 to tensile +2e9 dynes/cm². To improve stress uniformity we employed a multi-step deposition method, developed previously by SFI for highly uniform low stress chromium (Cr) film deposition [4]. Dividing the deposition process into several steps allows us to change zones of RF plasma localization on the wafer surface, increasing the uniformity of charged particle bombardment on the growing film. Fig. 3 illustrates the stress distribution for both the continuous process (a) and the multi-step process (b). Refer also to the data in Table 2.

(a)

(b)

Fig. 3 3500A films deposited under low cathode power and argon gas flow onto 0.7-mm thick Si wafers with Ti under layer: a) continuous process, b) multi-step deposition process.

Stress in Ni-V films was found to be very sensitive to deposition temperature¾ elevated temperature caused a displacement in the tensile direction. For example, film deposited on Ti under-layer by the recipe corresponding to the minimum stress point in Fig. 2, had a tensile stress shift from -1.3E8 to +2.3E9 dynes/cm² if the wafer was heated for 20 seconds before deposition (approximately 200C). So, it may be useful to introduce a wait step into the deposition flow between the Ti layer and the Ni-V layer if precise stress control is needed.

No significant changes in bulk resistance were observed for films deposited with the various process parameters or with the different thermal conditions (approximately, R_B = 77 mOhm-cm). This indicates that the film has uniform physical properties (foremost, density) across the wafer surface.

Silver films

Silver sputtered films reveal two essential features: first, it is difficult to get films with compressive or close to zero residual stress, and second, an initial stress can relax significantly within a few hours after deposition.

It was determined that the traditional stress control method by means of substrate bias could not decrease stress in Ag films. The opposite was true ¾ RF bias power exerted a negative role, increasing stress. Other process parameters had little influence on the film stress.

It is known that residual stress in Ag sputtered films has a mostly thermal component (tensile) [5], hence in this case an ion bombardment is not an effective instrument in decreasing stress. To lower the stress, it is necessary to ensure low temperature film growth conditions. Our experiments affirmed that periodic pauses in the process allowed us to obtain less stress. Nevertheless, as-deposited pure Ag films had tensile stress of about 2-3E9 dynes/cm². This means that the wafer gets a huge initial warpage: an average curvature radius for a 0.7-mm thickness 200-mm diameter wafer is about 25 - 30 m, wafer height in the center is more than -150 mm. Obviously, this is not acceptable for thinned wafers. Consequently, we had to search for an alternative approach, which would enable stress control in Ag.

Sputtered Films has undertaken considerable investigations of Cr and Ni-V sputtered films. We have found that their physical properties and stress characteristics were essentially changed by an addition of a small amount of nitrogen (N₂) into the gas flow during sputtering. This approach was successfully employed for stress reduction in Ag films. By evaluating the sheet resistance data, we determined that there was no chemical interaction between Ag and N₂ (even if nitrogen flow was higher than argon) when the process temperature was relatively low (less than 200C, approximately). The 'cold' deposited Ag + N₂ films had a shiny color and their bulk resistance had almost the same value as pure Ag films (2 mOhm-cm). To reach low temperature conditions these films were sputtered in several short steps with wait periods for wafer cooling. The 'hot' Ag + N₂ films (deposited by continuous process) had a milky color and elevated sheet resistance. The 'cold' recipes, optimized in terms of film uniformity and stress, yielded films with about two times less initial stress 1E9 dynes/cm² (curvature radius = 60-70 m).

Table 1 Stress in Ni-V Films deposited on Ti covered 200-mm diameter wafers
Cathode power, W			Argon flow, sccm		RF power, W	Stress, dyness/cm2
6000			4.5		250	7E9
2000			3.0		300	2E8
1000			3.0		300	-2.7E8

Post-deposition stress relaxation was the most essential difference between Ag sputtered films and Ti or Ni-V films. This process needs to be investigated more carefully, but some observations seem to be important:

1. Stress relaxation can continue for about 1 - 2 days under normal atmospheric conditions. Its rate depends on sputter parameters (i.e., temperature). The relaxation process has the highest rate during the first few hours.
2. Shiny pure Ag films decrease their stress level more significantly than milky films, so 'cold' films have more potential for relaxation than 'hot' films. For example, a 'cold' film, deposited using a recipe with wait steps under low cathode power = 500W, had initial stress 1.5E9 dynes/cm², which decreased to 7E8 dynes/cm² by the next day. A 'hot' film, continuously deposited with cathode power = 7000W, had initial stress 1.6E9 dynes/cm²; which decreased only to 1.2E9 dynes/cm² by the next day (Fig. 4).
3. Shiny Ag + N₂ films have the lowest level of initial stress and the lowest final stress after relaxation. Milky films have less stress shift. Films deposited under elevated temperatures (250-280C) had almost no stress relaxation (Fig. 4). This probably relates to activation of a chemical interaction between Ag and N₂ and the formation of any compound having a more stable structure and hardness.
4. Total post-deposition stress shift for shiny low stress 18500A thick Ag films sputtered at relatively low temperatures with the addition of nitrogen was
   about (6 - 8) E8 dynes/cm².

   Fig. 4 Stress relaxation in Ag films.

Design of optimal stress compensated Ti / Ni-V / Ag stack

To create a low stress Ti / Ni-V / Ag stack it is necessary to:

1. choose the initial stress level in Ag top layer,
2. define the stress values of the Ti and Ni-V under-layers based on the Ag stress level,
3. evaluate stress evolution in the stack due to the relaxation processes.

Note that stresses in Ti and Ni-V films do not change their values with time. So stress relaxation in the stack primarily results from the self-relaxation processes in the Ag layer and probably the relaxation processes activated in the Ag layer by distending forces from the compressive stressed Ti layer. Possibly the wafer. s elastic deformation, induced by metallization, also assists in Ag film stress relaxation.

The first step in designing an optimal stress compensated stack depends on the kind of silver layer we employ¾ either pure Ag (as-deposited stress is in the range of 2 - 3E9 dynes/cm²) or Ag with N₂ (1E9 dynes/cm²).

Table 2 Comparison of stress uniformity of Ni-V films deposited by means of continuous and multi-step recipes
Deposition method			Cathode power, W		Argon flow, sccm	RF power, W	Average stress, dynes/cm2	Stress range, dynes/cm2
Continuous			1000		3.0	300	-2.7E8	-1.2E9, +2E9
7-steps			1000		3.0	300	-1.2E8	-5E8, +5E8
14-steps			500		3.0	300	-1.5E8	-6E8, +1E8

In the case of the pure Ag layer, our experiments have shown that the as-deposited stress in the stack must be about 1E9 dynes/cm², which will be decreased to a value close to zero during the day (Fig. 5). This means we have to deposit the Ti layer with deep compressive stress of about -8E9 dynes/cm² (we consider that Ni-V layer has low tensile or low compressive stress).

In the case of Ag deposition with addition of N₂, the optimal stress compensated stack can be created with the following initial stress values (Fig. 6):

1. compressive Ti layer = -(1 to 1.5)E9 dynes/cm²;
2. compressive or tensile Ni-V layer = -1E9 to +1E9 dynes/cm²;
3. tensile Ag layer = +1E9 dynes/cm².

Process qualification for thinned wafer treatment

The backside metallization process described above, was first investigated and developed on standard thick wafers (0.7 mm). A low stress stack, designed with the employment of an inter-layer stress self-compensation approach, was realized with good repeatability.

Fig. 5 Post-deposition stress relaxation in Ti-NiV-Ag metallization

Fig. 6 Post-deposition stress relaxation in Ti-Ni-Ag metallization (silver sputtering with the addition of nitrogen)

Next, the process was adjusted for thinner, 0.35-mm thick wafers. We found that the temperature during Ni-V and, especially, Ag film formation, was a very important factor in defining the stress level in the stack. For thinner wafers, it must be ensured that a lower level of heat flows to the growing films during the deposition process in order to reach the same positive result. The optimal process for 0.35-mm wafer treatment enabled the following stack characteristics: residual stress after relaxation period was less than 2E8 dynes/cm², wafer curvature radius >100 m, and wafer height in the center < 50 mm.

Unfortunately, we could not measure stress directly on the 0.2-mm thick product (patterned) wafers because of their high initial curvature radius. However, because the lower heat capacity of the product wafers (as compared to the monitor wafers), should lead to the development of higher process temperatures and, consequently, a total stress shift in the tensile direction; we decreased the pre-heat temperature to 370C and increased the wait step duration during the Ag deposition. Post-deposition manufacturing operations with these samples were successful and confirmed that wafers did not receive any additional warpage.

Conclusion

We have shown that the stress levels in Ti and Ni-V films can be varied from compressive to tensile and therefore a range of near zero stress films can be produced. An inter-layer stress compensation approach has been developed for the sputter deposition of 3-layer low stress backside metallization of thinned to 0.2-mm thick, 200-mm diameter wafers. The Ti / Ni-V / Ag stack, consisting of films strained in opposite directions, had residual stress (after the self-relaxation period) of less than 2E8 dynes/cm², which enabled very low wafer warpage.

Acknowledgements

The technical staff of Sputtered Films, Inc. is gratefully acknowledged for the efficient support with the sputter tool qualification for the thinned wafer treatment. Particular thanks are due to Mr. Andrew Clarke for helpful discussions.

References

1. J. H. Lau 'Flip Chip Technologies', McGraw-Hill, New York, 1995.
2. J. Ullmann, A. J. Kellock, J.E. E. Baglin 'Reduction of Intrinsic Stress in Cubic Boron Nitride Films', Thin Solid Films, 341 (1999) 238-245.
3. P. J. Clarke, U.S. Patent # 3711398 (1974).
4. V. V. Felmetsger 'Tailoring Sputtered Cr Films on Large Wafers', Solid State Technology, July 1999.
5. N. Marechal, E. Quesnel, Y. Pauleau 'Characterization of Silver Films Deposited by Radio Frequency Magnetron', J. Vac. Sci. Technol., A 12(3) (1994) 707-713.

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

• Topics
• Author

Related Content

 

By This Author

• Controlled Sputtering Enables Better SiCr Films

SPONSORED LINKS