Advanced process control extends ECMP process consistency
02/01/2006
ECMP provides greater control capabilities compared to CMP processes, and a multizone cathode allows for precise control over within-wafer uniformities. Marathon runs show ability to extend the life of system consumable parts while maintaining process specifications. An end-point algorithm can accept feedforward data to adjust recipes and significantly reduce within-wafer topographic variation.
Antoine Manens, Paul Miller, Eashwer Kollata, Alain Duboust, Applied Materials Inc., Santa Clara, California
Process control is increasingly important to reduce copper line variations of on-chip interconnects for the 65nm manufacturing node of ICs and beyond. A recent IBM study [1] addressed the variables affecting the sheet resistance of damascene copper lines, and identified the need for increased process control of copper CMP. The same study also highlighted electrochemical mechanical planarization (ECMP) as a promising technology to increase process control relative to traditional CMP.
Advanced process control (APC) is necessary to maintain process consistency, especially across extended wafer runs where CMP consumable “pad” life becomes a limiting factor. The wear to the pad over time in conventional CMP causes not only profile variations but also more importantly an increase in defect counts. Conventionally, CMP process control relies on fixed settings, with inability to compensate for variations with incoming copper film thickness [2]. With ECMP technology, a different methodology allows for precise profile control on the wafer, without degradation across an extended pad lifetime.
ECMP technology
ECMP relies on a combination of mechanical and mostly electrochemical removal of the copper film on the wafer to achieve planarization. The planarization pad abrades the passivation layer that protects the elevated features of the wafer while leaving the recessed features passivated. The electrical potential difference between the anodic wafer and the cathode embedded behind the pad drives an electrochemical reaction that removes copper from the areas of the wafer that are not passivated. The net result is a high copper removal rate on the elevated areas only. This planarization mechanism can achieve a near perfect planarization efficiency on all types of structures [3].
Our experiments confirm that the removal rate increases with higher voltage between the wafer and the cathode. In accordance with Faraday’s law of electrolysis, the copper mass removed from the wafer surface is proportional to the charge flowing through the electrochemical cell. The linear relationship also agrees with the understanding of the ECMP electrochemical reaction: copper is oxidized into Cu2+ ions and the reaction releases two electrons for each atom removed from the wafer surface.
To better control the removal profile on the wafer, multizone cathodes were developed with each zone of the cathode independently biased by a separate power supply. A computer program simulating the motion of the wafer on the pad shows that each cathode has a different contribution to the removal of the wafer (Figure 1). While three zones are shown here, a five-zone cathode is used to provide greater control to meet 2mm-edge-exclusion specifications. The net removal is a linear combination of the removal from each zone. By modulating the voltage on each zone of the cathode, and therefore the removal rate, it is possible to modify the removal profile.
 Figure 1. a) Location of the wafer relative to the three-zone cathode and b) contribution from each zone to the removal profile on the wafer. |
Because of the linear relationship between the accumulated charge (i.e. current integrated over time) in a cathode zone and the copper removed, it is possible to calculate the charge required by each zone to achieve the desired removal profile. Modeling the contributions from each zone provides the foundation for a process end-point algorithm for ECMP.
Process control by charge quantization in ECMP is in principle very different from process control by end-point detection techniques in traditional CMP. Copper CMP process control usually relies on in situ sensors such as eddy current sensors [2] to directly measure the copper film and to modify the process inputs such as time and pressure accordingly. In contrast, ECMP process control relies on the simple and precise measurement of the energy input to the system for both end-point and removal profiles.
ECMP system
A three-platen planarization system was used, with ECMP on the first platen using an abrasive-free electrolyte solution, and conventional CMP slurries on the remaining two platens. Low ECMP planarization pressure protects the wafer from low-k damage while still providing a removal rate that would require high pressure in conventional CMP. Planarization pressure on platen one (P1) was 0.3psi, and 0.8-1.2psi pressures were used for P2 and P3. P1 removes the bulk copper film, P2 removes residual copper, and P3 removes the Ta/TaN barrier and additional dielectric material.
A three-zone cathode was used on P1 as shown in Figure 1. The drawing on the left of the figure shows the location of the wafer relative to the cathode zones. The graph on the right shows the normalized removal contribution from each of the three cathodes. To remove by-product build-up in the pad and maintain its properties, a low down-force (1.5 lb.), low abrasion in situ pad conditioning was used.
ECMP marathon run results
In the first experiment, a fixed charge/zone (no APC) was applied to the wafer to achieve a copper film removal of 10,000Å. More removal near the wafer edge was added to the planarization recipe to compensate for the typical thicker copper near the edge (Fig. 2).
 Figure 2. Copper removal profile of 18 monitor wafers over a 1500-wafer extended run. |
The removal profile was monitored on 18 wafers throughout a 1500-wafer run and the results are plotted in Figure 2. The wafer-to-wafer removal variation was <2% (3σ) at 3mm edge exclusion. The removal profile and removal amount were maintained well over time. Throughout the 1500-wafer run, the removal rate remained stable at around 6500Å/min with <4% (3σ) variation, and post-P1 defects also remained stable. These results demonstrate that controlling the charge through each zone of the cathode is an effective way to achieve a constant removal profile over the life of the pad. While very simple, this endpoint algorithm is efficient.
Low down-force conditioning may also contribute to the stable post-P1 defect count. Conditioning can prevent loading of the pad with by-products that generate defects on the wafer. Further, the low down force and low abrasion reduce excessive wear of the pad and extend the lifetime.
A second experiment was conducted to verify the APC capabilities. In this experiment, the incoming profiles were purposely chosen with various thickness and edge profiles. This simulates, in an extreme fashion, the natural variations that can be observed from wafer to wafer and lot to lot due to electrochemical deposition profile and thickness variations.
A noncontact, eddy-current based metrology unit was integrated with the factory interface of the CMP tool. This allowed incoming profile measurements for every wafer going through the tool. The copper film profile data was then fed-forward and used to adjust the multizone removal recipes to achieve a flat post-P1 profile at a target thickness.
The zone power supplies were then controlled by the endpoint algorithm to reach the desired charge by the end of planarization. The post-P1 profile was also tuned to match P2 removal profiles and prevent residues at the edge of the wafer (after P2). The removal rate was typically >6000Å /min.
The endpoint algorithm was set up to stop at a target thickness of 2400Å. The experimental results (Fig. 3) display the capabilities of the APC algorithm. The first observation is that the incoming wafer nonuniformity can be greatly reduced. In all cases, the thickness range was reduced from > 2000Å pre-planarization to < ~500Å post-P1. Thus, even with wide variations in the incoming film profiles, a consistently flat wafer with a target film thickness is achieved.
 Figure 3. Variations in the profiles of incoming wafers (solid lines) are reduced with excellent control after copper ECMP using feedforward APC (dotted lines) [4]. |
This outcome proves the capability of the multizone cathode to effectively control the removal profile. In conventional CMP, the control “knobs” available to modify the planarization profile include primarily down-force pressure modulation but also platen velocity and slurry flow [2]. At low pressures such as 0.3psi, profile control becomes more challenging because of the relative stiffness of the wafer and the non-Prestonian behavior of most slurries. With ECMP, the control knobs are the charges allowed to accumulate in each zone of the cathode. For example, by increasing the target charge on zone 1, an edge-fast planarization profile can be achieved.
A second observation is that a consistent post-P1 profile can be achieved, regardless of the incoming profile. Wafer-to-wafer nonuniformity was <4% post-P1, showing the ability of the endpoint algorithm to accurately calculate the required charge per zone based only on the incoming profile information. The approach is again dramatically different from conventional CMP, in which profile control requires real-time in situ thickness measurement to correct for profile errors [2].
These experimental results show the capability of ECMP to provide a consistent profile after P1. Therefore, the P2 planarization time will be more stable, potentially enabling a higher overall throughput. In addition, there will be a more uniform copper clearing across the wafer on P2 and less over-planarization, hence reduced dishing and erosion across the wafer.
A major benefit of extending pad life is a cost-of-ownership advantage for CMP, the most expensive step of the chip manufacturing process. Pad life is typically limited to ~1000 wafers. Such a short lifetime affects not only cost but also the availability of the tool because of frequent change and the need for requalification. With multizone APC ECMP and optimized pad conditioning, pad life can be extended to about 1500 wafers. In addition, ~30% slurry cost savings can be achieved as a result of ECMP’s use of an abrasive-free electrolyte instead of conventional CMP slurry.
Conclusion
Instead of relying on in situ film thickness measurements of the wafer, ECMP process control uses predetermined accumulated charge to adjust the film profile with a multizone cathode. The endpoint algorithm is rooted in a physical representation of the electrochemical cell based on Faraday’s law of electrolysis and the cathode’s geometry.
This process control method was validated through extensive wafer testing. Results show the capacity to maintain the removal profile and low defects across a 1500-wafer marathon without changing the planarization pad. Low within-wafer and wafer-to-wafer nonuniformities were also achieved, even with extreme incoming variations in film thickness and profile.
Acknowledgments
The authors thank Wei-Yung Hsu, Zihong Wang, and Yan Wang at Applied Materials for their contributions to these developments.
References
- D. Vogler, “Process Control Capability Needed to Take Advantage of Lower-k Materials,” Wafer News, July 4, 2005.
- D. Bennet, et al., “Real-time Profile Control for Improved Copper CMP,” Solid State Technology, June 2003.
- L. Economikos, et al., “Integrated Electrochemical Mechanical Planarization (ECMP) for Future Generation Device Technology,” IITC 2005.
- L. Economikos, et al., “Electropolish Techniques,” PacRim-CMP, November 2005.
Antoine Manens is an engineering manager in the ECMP division at Applied Materials, Santa Clara, CA.
Paul Miller is a product marketing manager in the CMP division at Applied Materials.
Eashwer Kollata is a mechanical engineer in the ECMP division at Applied Materials.
Alain Duboust is the technology manager for the copper ECMP process in the CMP division of Applied Materials.