Tungsten CMP process developed
04/01/1998
Tungsten CMP process developed
Kapila Wijekoon, Ronald Lin, Boris Fishkin, Susie Yang, Fritz Redeker,
Gregory Amico, Savitha Nanjangud, Applied Materials Inc., Santa Clara, California
Commercially available system hardware and consumables-sets produced well-characterized tungsten chemical mechanical planarization processes. Obtaining adequate tungsten removal rate selectivities to Ti/TiN layers and to oxides is critical for process integration. A marathon run showed that the process is reproducible.
Tungsten (W) chemical-mechanical planarization (CMP) is rapidly gaining acceptance as the process of choice for sub-0.35-?m W plug formation. Probe yield improvements of up to 15% compared to the W etchback process are reported [1]. W CMP is advantageous for sub-0.35-?m process integration, by minimizing plug recess and local step heights (Fig. 1).
Several key issues must be addressed, however, before implementing W CMP into production, including process repeatability, defects, process integration, ease of slurry use, and throughput. Each of these issues is affected by several interrelated CMP process parameters (Table 1). This article attempts to address the key issues and critical parameters required for a production-worthy W CMP process.
 |
Figure 1. W Plug formation using CMP: a) W deposition over barrier layers; b) W, Tin/Ti CMP leaves recessed plugs; c) oxide buff eliminates recess.
 |
Experiment
Experiments used a second-generation Applied Materials Mirra CMP system with a multiplaten, multihead configuration. The polishing head (Titan head) is a recent design with a flexible membrane that applies uniform pressure over the wafer backside while conforming to the wafer shape (Fig. 2). The head also has a retaining ring with independent pressure control to optimize removal rates at the edge of the wafer. A unique wafer backside pressure distribution profile improves system performance.
An in-situ, real-time endpoint monitoring system is located in the platen. This system uses laser interferometry to measure the reflectivity of the metal film on the polishing surface through a window in the pad. An optical in-situ rate monitor (ISRM) accurately signals the transition from metal film to oxide. The ISRM algorithm can be programmed to detect the sharp change in slope characteristic of the metal to oxide transition.
The sharpness of the transition is a function of the polish process uniformity. An inflection at the top of the trace corresponds to the end of the W polish step, and an inflection at the bottom of the trace indicates almost complete clearing of the barrier layer. The endpoint algorithm specified the over-polish required to ensure complete removal of the barrier layer. The ISRM controlled the endpoint and over-polish time on all the patterned wafers used in this study.
 |
Figure 2. Comparison of Titan Head to conventional rigid carriers.
A sequential two-platen polish, followed by either a DI rinse or oxide buff step on the third platen, generated the data. Platen speeds varied from 83 to 110 rpm, depending on the slurry used. The head speed was always slightly lower than the platen speed by about 3-6 rpm. An OnTrak DSS 200 cleaner performed the post-CMP cleaning with two double-sided roller brush scrubbers (using an ammonium hydroxide solution) and a spin-rinse dry station.
The polishing head membrane pressure on the wafer was 2.8-4.0 psi, with 3.5 psi as the optimum condition for most slurries. The retaining ring pressure was set to optimize the uniformity performance at 5-6 mm edge exclusion. Signals from the in situ endpoint system terminated polishing on the second platen. The buff time on the third platen was targeted to remove 300-400 ? of oxide. A diamond grid disk conditioned the pads prior to polish.
The within-wafer nonuniformity (WIWNU) marathon tests used 200-mm wafers with blanket coatings of 6000- to 8000-? CVD W, 800-? PVD TiN, and 300-? PVD Ti over plasma-enhanced tetraethylosilicate (PETEOS) oxide. The oxide erosion and surface morphology tests used 200-mm SEMATECH patterned wafers with ~8000-? CVD W, ~400-? PVD TiN, and 250-? PVD Ti. The patterned wafers had 0.5-?m plugs in 1-?m pitch dense arrays and 3-?m pitch sparse arrays. Defect tests were performed using 200-mm wafers coated with blanket PETEOS and thermal oxide.
Slurry/pad characterization
Process characterization involved testing three commercially available slurries with different oxidants. They included a potassium iodate oxidant with alumina abrasive particles; a ferric nitrate oxidant with alumina abrasive particles; and a hydrogen peroxide oxidant with fumed silica abrasive particles. A different pad was selected for each slurry to achieve optimum process performance (Table 2).
Table 3 compares the three slurry/pad combinations for process repeatability characteristics. For each slurry/pad combination, a different platen speed and membrane pressure is optimal for rate and uniformity results; the experiments used these previously determined optimum conditions. The table shows excellent WIW and wafer-to-wafer (WTW) uniformities for all three slurries. Removal rate of W is mainly controlled by platen speeds, membrane pressure, and chemical characteristic of the slurries.
For the W CMP process, high TiN and Ti glue layer removal rates are essential to prevent oxide erosion and plug recessing. It is also necessary to have a high selectivity to the underlying oxide, to obtain local planarity over the plug surface, and to prevent excessive loss of the surrounding field oxide.
Selectivity comparisons for the three processes investigated (Table 4) show that slurries B and C have excellent selectivity to TiN and Ti, as well as good oxide selectivity.
Slurry selection for IC manufacturing applications is based not only on process performance but also on such attributes as ease of handling, pot life, and contamination (Table 5). Based on the process characterization, a new production-worthy W CMP process was developed with the hydrogen peroxide-based slurry with fumed silica abrasive to add to the previous industry standards that use potassium iodate and ferric nitrate-based slurries.
 |
Oxide erosion and plug recess
In regions of high feature density, there is a tendency for the oxide to over-etch below the surface of the field oxide. This effect of pattern density on oxide erosion is well known [2].
An additional oxide buff step is often incorporated after the polish step to planarize the elevated regions of the oxide. The target oxide removal is approximately 300 ?. In this investigation, pattern density effects were studied on 4 ? 4-mm arrays consisting of 0.5-?m plugs with 1-?m (dense) and 3-?m (sparse) pitch. The effect of the oxide buff step on erosion was also investigated.
For Process C, 103 rpm platen speed, 3.5 psi membrane pressure, and no oxide buff produced the data. The oxide erosion was ~1150 and ~700 ?, in the dense and sparse regions, respectively. Other authors reported similar oxide erosion [1, 3]. The addition of a buff step reduced these erosion levels to ~880 and ~620 ?.
The oxide erosion can be further decreased to <600 ? in dense plug arrays by reducing the membrane pressure. The removal rate can be maintained by increasing the platen speed. Further study is under way to reduce the erosion by optimizing the slurry formulation and other process parameters.
Over-polish is often required to ensure complete removal of the barrier layers, producing some additional recess of the W in the plugs. Endpoint detection minimizes the need for extended overpolish, thereby minimizing plug recess.
 |
Figure 3. Post-clean defect repeatability for Process C extended run at 110 rpm platen speed and 3.9 psi membrane pressure, taken using an oblique incidence defect metrology system (>0.2 ?m) at 6-mm edge exclusion.
Process C, at 103 rpm platen speed and 3.5 psi membrane pressure, produced an excellent plug recess of <200 ?. The oxide surface was very smooth with an RMS roughness value of 6 ?. The addition of a 24-sec. oxide buff step resulted in 77 ? of W plug protrusion, which is consistent with data reported elsewhere [1]. The plug morphology can be optimized by adjusting the oxide buff time. No coring was seen with the optimized process.
Marathon testing
An extended 500-wafer run of Process C evaluated its production-worthiness based on process stability. The wafers used for this test were supplied by a customer and were identical to those processed in the characterization phase.
The extended run demonstrated stable long-term removal rate and WIWNU. The average WIWNU was 3%, while the WTWNU was <5%. Post-clean defect levels on thermal oxide wafers at 0.2-?m detection threshold were also monitored during this extended run, showing excellent defect performance of <15 counts/wafer without an oxide buff step (Fig. 3).
Conclusion
A multiplaten, multihead CMP polishing tool with a flexible membrane carrier and in-situ endpoint monitoring produced an optimized W CMP process. The process consumable set was chosen from three commercially available slurry/pad combinations with demonstrated excellent within-wafer and wafer-to-wafer uniformity. The peroxide slurry process on a hard pad had improved defect performance and ease of handling.
After process optimization using one pad/slurry set, an extended run showed that the process is capable of rate stability and consistently low defect levels. With respect to yield issues, plug recess values were superior to those obtained by the conventional W etchback process. Oxide erosion for this process was consistent with current industry standards, and preliminary results show that tuning process parameters will significantly reduce erosion. When particle defects and ease of use are considered, this process appears to be an excellent choice for volume production.
References
1. V. Blaschke, et al., 1997 CMP-MIC Conference Proceedings, p. 219.
2. N. Elbel et al., 1997 CMP-MIC Conference Proceedings, p. 75.
3. Y. Shih, 1997 CMP-MIC Conference Proceedings, p. 117.
Acknowledgments
The authors would like to thank Bill Hartwig and Sen Ho Ko for their technical support and dedication in performing the process characterization described in this paper, and Gino Addiego for his management support.
Mirra, Titan Head, and In-Situ Rate Monitor are trademarks of Applied Materials.
For more information, contact Savitha Nanjangud, Applied Materials Inc., CMP Div., 3111 Coronado Dr., Santa Clara, CA 95054; ph 408/563-1419, e-mail [email protected].