Real-time profile control for improved copper CMP
06/01/2003
Overview
Maintaining constant line or sheet resistance across the wafer requires greater control of remaining copper thickness during the CMP process. Incorporating an in situ thickness sensor into the CMP process has yielded information not previously available — the actual evolution of the copper film thickness profile during polishing. With the knowledge of dynamic profile evolution, it becomes possible to achieve a closed-loop control of the remaining film thickness uniformity in real time, which leads to tighter distribution of metal parametrics for copper damascene structures across the wafer and wafer-to-wafer.
Previously, real-time profile control was not available for CMP, forcing engineers to rely on the initial settings in a copper CMP process to deliver uniform copper lines. Active control of the polishing profile is required to achieve target line resistance uniformity across the wafer and to compensate for variations in both the initial electrochemical copper deposition thickness profile and the CMP copper removal profile.
The 300mm CMP tool used for the evaluation has three independent polishing platens that perform a three-step process: bulk copper removal is done at a high polishing rate on platen 1; complete copper removal down to the thin barrier layer is done at a low rate on platen 2; and the barrier layer is removed on platen 3. For maximum throughput, the removal rate on platen 1 is adjusted to be as high as sustainable, without compromising within-wafer removal uniformity and pad life. The removal rate on platen 2 must be low, to allow for accurate detection of the endpoint at the thin barrier layer and to minimize dishing of the copper structures.
It is preferable to leave no more than ~0.2–0.3µm of copper on the wafer before starting the second polishing step. This has traditionally been achieved by using a qual wafer to determine the copper removal rate for step one and estimating the polishing time required to reach the target thickness. Recently, a system called iScan was introduced that performs continuous in situ monitoring of the average remaining copper thickness [1]. It provides for the endpointing of step one at target thickness regardless of incoming copper film and average removal rate fluctuations.
High-resolution thickness monitoring
The use of only average film thickness data for controlling the process has limitations. To achieve closed-loop uniformity control, real-time, high-resolution, spatially resolved thickness data are needed. Based on previous work, a high-resolution, in situ eddy current resistivity sensor has been developed that measures the copper resistance (and indirectly, the copper thickness and removal rate) across the wafer during polishing [1]. A small AC probe coil beneath the wafer induces an eddy current in the conductive copper film and the film thickness variation is derived from changes in the eddy current, which depends on the average film conductivity. The new sensor provides mm-scale resolution of the copper thickness from the center to the edge of the wafer.
null
The sensor design and placement were optimized for maximum signal/noise and spatial resolution especially within the wafer periphery. The spatial resolution of the sensor across the wafer radius is shown in Fig. 1, where readings made on a continuous sheet of copper film are compared to measurements made with the previous low-resolution sensor and an off-line four-point probe. The readings of the sensor track the four-point probe readings all the way to the edge of the copper layer. By contrast, the thickness drop-off starting 25mm from the wafer edge is not detected by the low-resolution sensor.
Closed-loop profile control
Incorporating the thickness sensor into the CMP tool yielded information not previously available — the actual evolution of the copper film thickness profile during polishing. An analysis of these data shows that just knowing incoming film profile (first scan), and final profile (last scan) would lead to erroneous assumptions regarding the CMP polishing rate, and possibly to incorrect settings of the polishing parameters. Without the knowledge of dynamic profile evolution, the engineer would have to assume that the profile remains stable throughout the polish. The greatest value of the sensor lies in using the real-time in situ measurements for closed-loop profile control.
The CMP profile control "knobs" available to control polishing rate include down force pressure, platen velocity, and slurry flow. In this work, profile control is implemented using down force pressure modulation with a multizone polishing head Titan Profiler, where the rate profile of copper polishing is controlled via input pressures to different annular regions behind the wafer. The real-time copper profile data are utilized by software that incorporates a profile control model. This model is constructed from a physical model of the hardware components together with the experimentally determined response sensitivities of the copper CMP process to variations in input parameters.
During polishing, the software model instantaneously analyzes the copper thickness profile and determines the areas of the profile that deviate from the target profile. The model then generates incremental modifications to the polishing down force in individual pressure zones of the head to correct for profile errors by enhancing or reducing removal rate for each of the individual concentric zones of the wafer. The control system generates these changes to the CMP system at discrete time intervals during the polish process.
Fixed-process vs. closed-loop
The ability of closed-loop thickness control to adjust for incoming across-wafer copper nonuniformity is shown in Fig. 2 for CMP of two wafers, one with fixed process parameters and one with closed-loop profile control. The pre-CMP copper thickness profile was similar for both wafers (green profile data) and the target final copper thickness was set at 3000Å for both wafers. The results for the two control methods differ significantly in the region beyond ~130mm from the wafer center. The total thickness variation (max.–min.) for this outer region is about 3000Å for the fixed-process wafer, but only 900Å for the closed-loop control wafer. This portion of the wafer is significant. On a 300mm product wafer, it contains the last complete device chips; a 20mm die may contain 30–35 devices. With a fixed CMP process, those chips will have nonuniform interconnect parameters and much thinner lines. That may lead to either product yield degradation or bin-sorting issues for the final IC product.
 Figure 2. Final thickness profiles after CMP for fixed-process (red) vs. closed-loop (blue) profile control for two wafers with an edge-thin incoming profile (green). |
null
To evaluate the capability of the closed-loop CMP control system to compensate for normal process variations, the copper removal on one 300mm Reflexion CMP tool was monitored on a daily basis over a three-month period. Figure 3 summarizes the data as histograms of the percentage within-wafer nonuniformity (WIWNU) or copper removal for both fixed-process parameters and closed-loop control. Closed-loop control resulted in a tighter distribution.
null
Profile Tuning
In some cases, it may be advantageous for CMP process engineers to require the copper film profiles to deviate from flatness during the bulk copper polishing. The subsequent CMP steps, the complete copper removal or barrier removal, may require predetermined thickness deviations. To provide this capability, the system was designed to let the user input the specific offsets for copper film thickness in specific areas of the wafer. Figure 4 shows data for several wafers run with a variety of final copper profile offsets in the wafer periphery. The system followed the offset settings and by modulating the appropriate pressure zones in the polishing head, modified the final film profile accordingly.
 Figure 4. Closed-loop control with user-programmable thickness offsets. |
null
Line resistance uniformity
The ultimate proof of the process is the variation in final line resistance in a patterned wafer. Figure 5 shows the effect of implementing closed-loop control at the bulk copper CMP step on the line resistance for the interconnect structures in the die. The graph compares two wafers, one run with the fixed optimized set of CMP parameters, and the other run with the system programmed to modulate parameters to maintain a flat final profile. Each data point is a copper line thickness measurement for the same structure within the die.
 Figure 5. Final copper line thickness across a wafer. |
null
Fourteen die across the 300mm wafer diameter were measured. The total variation of the line thickness
esistance exceeded 25% for the fixed-process wafer and was about 9% for the closed-loop wafer. At electrical test, the interconnect performance on the wafer processed with closed-loop CMP control system will be very uniform die-to-die, while for the fixed-process wafer, the center dice will have higher line resistance, and may exhibit higher RC-constant values, slowing chip speed.
Conclusion
A real-time, in situ, closed-loop profile control system for 300mm copper CMP that can track and control the evolution of the copper thickness profile has been designed and implemented. The closed-loop bulk copper CMP control system provides a more robust process with tighter control over on-wafer interconnect performance. Closed-loop control also maximizes the efficiency of the bulk copper-polishing step and increases throughput. Process data for open- and closed-loop CMP are presented for both blanket and patterned wafers, including post-CMP film profiles and line thickness data.
Doyle Bennett, Bogdan Swedek, Jeff David, Jimin Zhang, Konstantin Smekalin, Applied Materials, Santa Clara, California
Acknowledgments
The authors would like to acknowledge the following individuals for their contributions: Manush Birang from the CMP division at Applied Materials, Dick de Roover from SC Solutions Inc., and Laurie Miller. Titan Profiler, iSCAN, and Reflexion are trademarks of Applied Materials.
Reference
1. K. Smekalin, B. Swedek, R. Bajaj, A. Zutshi, F. Redeker, M. Birang, "High Throughput Copper CMP Process with Real-Time In Situ Bulk Copper Thickness Monitoring," Proceedings of the 2001 CMP-MIC Conference, pp.163–166, 2001.
For more information, contact Konstantin Smekalin at Applied Materials, 3111 Coronado Dr. M/S 1512, Santa Clara, CA 95054; [email protected].