Improved planarization for STI with fixed abrasive technology
06/01/2000
overview
Shallow trench chemical mechanical planarization is often the most difficult aspect of the shallow trench isolation module. This article discusses the use of fixed abrasive technology and compares the results to conventional and high-selectivity slurry processes. The planarization characteristics show that the fixed abrasive is selective to topography, not to nitride. Once the surface topography is planar, the overpolish rate is very low for both high-density-plasma fill oxide and the nitride polish stop. This allows for a large overpolish window, minimizes nitride loss and within-die uniformity, and reduces oxide loss in the wide trench features due to dishing. These advantages allow process simplification by minimizing the need for pattern fill structures (tiling) and reverse mask processing.
Tuyen Vo, Todd Buley*, Rodel Inc., Phoenix, Arizona
John J. Gagliardi, 3M Corp., St. Paul, Minnesota
*Additional authors are listed in the Acknowledgments
 Figure 1. a) Details of experimental wafer structures, and b) MIT 964 mask layout. |
As device geometries scale down to 0.35mm and lower, shallow trench isolation (STI) becomes a necessity, since it offers improved isolation between devices compared to the traditional approach of local oxidation of silicon [1]. The goal of STI chemical mechanical planarization (CMP) is to remove all of the oxide over the dense active areas and the low-density or isolated active area structures, while leaving the same residual nitride stop layer thickness across all feature sizes. The residual nitride thickness determines the step height between the active surface and the trench oxide surface. Ideally, this step height would be the same across all within-die pattern densities and across the wafer.
Different approaches are possible in order to achieve good planarization performance, including:
- a reverse etch-back process to pre-planarize the wafer by reactive-ion etching, where most of the oxide from the active area is removed while protecting the trench oxide [2];
- the use of pattern fill structures ("dummy" patterns) to optimize the overall pattern density to look uniform at the die level [3]; or
- the use of high-selectivity slurries that improve the ability to halt polishing on the nitride stop layer [4].
These approaches can be used independently or in combination to meet planarity requirements.
Fixed abrasive technology
A new approach to STI CMP has been developed through the use of 3M's Slurry-Free Fixed Abrasive technology [5-9]. A combination of microreplication technology and coated abrasive technology has resulted in a fixed abrasive matrix capable of CMP. Small composites (~200mm wide x 40mm high) of abrasive (CeO2) and resin binder are positioned on a polyester backing. The composites are precisely shaped and provide a third dimension of abrasive, as well as uniform space for chemical and by-product transport. Rather than continuously having to supply fresh abrasive to the pad via slurry and pad conditioning, the fixed abrasive matrix contains all the necessary abrasive with no conditioning required. The only chemistry needed is pH-adjusted water to meet polish rate requirements. The fixed abrasive is designed so that the topology of the wafers conditions the abrasive composites to expose fresh mineral during polishing.
A key advantage of CMP using fixed abrasive over conventional methods is its selectivity to topography (>100:1). The selectivity between oxide and nitride is on the order of 1:1. This provides a polishing system that rapidly planarizes and is not susceptible to dishing on overpolish.
 Figure 2. a) SEM micrograph of the fixed abrasive, and b) general diagram of the roll-to-roll polisher. |
Many of the concepts that are known and applied when using slurry-based STI CMP apply as well to fixed abrasive STI CMP. However, there are some fundamental differences between polishing with two-body (fixed abrasives) and three-body (slurry) systems. The use of abrasive slurries has essentially been the only technique for wafer CMP. In other types of polishing, such as glass, metal, or ceramic polishing, there are two competing material removal techniques: two-body and three-body abrasion [10, 11]. The relative effectiveness of these two competing methods appears to depend markedly on whether material removal occurs primarily by plastic or brittle deformation. In three-body systems, abrasive slurries are continuously fed to a platen. It has been suggested that the loose abrasive grain may become temporarily embedded in the slurry pad and thus be dragged in a fixed position across the work surfaces (two-body abrasion). Such phenomena are more likely to occur under heavy loads. More characteristic of three-body abrasion is material removal through impacting. In slurry systems, free abrasive grain is also available to remove material in low-lying areas on patterned wafers, which can reduce the planarization ratio. Three-body abrasion is generally preferred for brittle materials and, in general, two-body abrasive is favored for plastic deformation removal [10]. It has been theorized that oxide CMP mechanisms involve the plastic deformation and hydration of oxides [12]. It might therefore be deduced that CMP would be better accomplished through two-body polishing.
This article will present results of polishing Sematech/MIT 964 patterned wafers using a conventional interlevel dielectric (ILD) slurry process, a high-selectivity slurry process, and the fixed abrasive matrix process. The key outputs will be within-die range of trench oxide and active nitride, and planarity across the pattern density and line-pitch structures. Potential benefits of using the fixed abrasive matrix over conventional ILD slurry or high-selectivity slurry will be discussed in terms of process and integration simplification.
Experimental set-up
 Figure 3. Range of a) trench oxide loss and b) active nitride loss across the 0-90% pattern density features of the MIT 964 mask. |
The patterned test wafers used for this study were based on a Sematech/MIT 964 mask set filled with high-density-plasma chemical vapor deposition silicon oxide film (HDP oxide). The wafer structure and pattern layout are shown in Fig. 1. The trench depth difference for conventional slurry polishing and fixed abrasive matrix polishing (4500Å and 3500Å, respectively) was due to wafer availability. The HDP oxide overfill was 100Å on the wafers polished with the fixed abrasive due to the fixed abrasive's selectivity to topography. However, we polished wafers with an overfill thickness as high as 1100Å with minimal effects on the output response. The test-pattern structures had an active density of 0-90% fixed at a 100mm pitch, and 1-1000mm pitch fixed at a 50% density. Since we are comparing the planarization of wafers with different deposition thickness and trench depths, we will report the total trench oxide and active nitride loss with zero loss being ideal for all conditions.
Conventional polishing with high-selectivity and ILD slurry was completed using a Strasbaugh 6DS-SP polisher with process settings of 6psi down-force, 40rpm platen speed, 60rpm carrier speed, and 150ml/min flow. A Rodel 0.050-in.-thick IC1000 A3 k-groove pad was used for all runs. Fixed abrasive polishing was completed using the Obsidian Flatland 501 roll-to-roll polisher with process settings of 3.5psi down-force, 600mm/sec speed, 1-in. indexing between wafers, and the 3M Slurry-Free Fixed Abrasive matrix (see Fig. 2).
Trench oxide and active nitride measurements were taken using the Thermawave Optiprobe 2600. Step height measurements were taken using Tencor P-1 and P-22 profilometers. All wafers were polished until the oxide was just cleared from the active nitride features. The die that provided the best planarization performance was measured for the conventional slurry processes. This approach was used to eliminate any within-wafer nonuniformity effects. Wafer resources limited the amount of process optimization for the slurry processes.
Pattern density effect at 100mm line pitch
The active nitride loss and trench oxide loss were measured on each pattern density feature ranging from 0-90% for each process, and plotted as shown in Fig. 3 on p. 124. As expected, the ILD slurry showed the highest range of trench oxide loss (~3500Å) and active nitride loss (~1500Å) due to low selectivity to nitride. The high-selectivity slurry process showed good performance for the active nitride range (~410Å), excluding the 10% pattern density feature. However, the amount of trench oxide loss is not much improved over the ILD slurry. This suggests that the high selectivity to nitride does not prevent significant dishing in the trench oxide areas.
 Figure 4. a) Residual step height vs. pattern density, and b) time to planarity for the fixed abrasive matrix. |
The fixed abrasive process showed good performance for both the active nitride range (~490Å) and the trench oxide range (~1120Å). If the 0% density feature is excluded, the range of trench oxide loss is ~400Å.
Figure 4a shows step height vs. pattern density for the three processes. The high-selectivity slurry and ILD slurry both showed significant step heights resulting from dishing across the pattern density features, which explains the high range of trench oxide loss.
The fixed abrasive showed minimal step height (<50Å) across the pattern density features. This is explained by looking at the time-to-planarization curve shown in Fig. 4b. As the step height is reduced, the polish rate of the fixed abrasive drops below 500Å/min. The drop in rate, coupled with the fixed abrasive's selectivity to nitride (~1:1), results in a very planar surface with <50Å dishing.
Line pitch at 50% density
The line-pitch structure had a reduced effect on total loss and range for the trench oxide and active nitride for each process (Fig. 5). The ILD slurry-polishing results showed the most total oxide loss (~1700Å), with a range under 400Å. The total nitride loss was ~700Å, with a range of ~300Å. The high-selectivity slurry showed a large range of oxide loss (600-1600Å), but showed a small nitride loss (<200Å). Again, we see good polish stop on the active nitride, but continued erosion of the trench oxide as a strong function of line pitch. The fixed abrasive showed good performance with a very low total range of trench oxide loss (~200Å) and active nitride loss (~35Å).
Overpolish
 Figure 5. Range of a) trench oxide and, b) active nitride loss, across the 1-100mm line-pitch features of the MIT 964 mask. |
Nitride loss and trench oxide loss at the 50% density/100mm pitch feature was measured and plotted vs. polish time for each process (Fig. 6 on p. 128). The polish time was normalized to compare the three processes; 100% indicates the time at which the oxide is just cleared from all of the active nitride features. The ILD slurry showed a linear loss of active nitride and trench oxide with overpolish. The high-selectivity slurry showed a good overpolish response for nitride loss, losing ~500Å after 50% overpolish, but showed >1000Å loss of trench oxide for the same condition. In this case, the trench oxide polishes at a faster rate than the nitride, resulting in significant dishing. The fixed abrasive showed an excellent response to overpolish, losing <300Å of active nitride and trench oxide after 70% overpolish. The nearly flat overpolish response is explained by the fixed abrasive's high selectivity to topography (>100:1) and the near 1:1 selectivity of oxide to nitride, resulting in a planar surface.
Process integration
 Figure 6. a) Oxide loss and b) nitride loss vs. normalized polish time for the 50% density/100mm pitch structure. |
The fixed abrasive CMP process shows the potential to simplify the integration of STI into the process flow. Currently, many integration techniques are used to improve process performance and increase the process margin for STI CMP. Reverse mask processing is one option, but this requires several additional process steps that increase both the cost and cycle time. The use of patterned fill structures (dummy patterns) is another option, but it can be difficult to implement and may require modeling and optimization to achieve the desired results.
The performance characteristics of the fixed abrasive may provide enough additional process margin so that direct STI polish is possible on a broader range of pattern density features, thus improving existing integration regimes. Minimizing HDP oxide overfill may be an additional benefit of having improved cycle time.
Conclusion
STI polishing using a fixed abrasive matrix was demonstrated and compared to conventional ILD slurry and high-selectivity slurry processes. With fixed abrasive technology, topography can be reduced from an initial step height of 4900Å to less than 50Å with minimal nitride and trench oxide loss and range, low dishing, and a forgiving overpolish window. It is important to note that high selectivity to topography, and near 1:1 selectivity to nitride, is key to achieving the level of performance demonstrated here. The improved polish performance using the fixed abrasive matrix also has the potential to simplify process integration by allowing direct STI polishing and, in doing so, to lower both cost and cycle time.
Acknowledgments
Additional authors of this work are Sharath Hosali and Tao Zhang of Rodel Inc. The authors would also like to recognize the contributions of Dale Hetherington and David Stein of Sandia National Laboratories, Mike Oliver of Rodel, and Christopher Loesch and Kenneth Bello of 3M.
References
- Joseph M. Steigerwald et al., Chemical Mechanical Planarization of Microelectronic Materials, John Wiley & Sons, 1997.
- S.S. Cooperman, A.I. Nasr, G.J. Grula, "Optimization of a Shallow Trench Isolation Process for Improved Planarization," J. Electochem. Soc., Vol. 142, No. 9, pp. 3180-3185, Sept. 1995.
- B. Withers, E. Zhao, R. Jairath, S. Hosali, "A Wide Margin CMP and Clean Process for Shallow Trench Isolation Applications," Proc. 3rd Int. CMP-MIC, pp. 319-327, 1998.
- Ki-Sik Choi et al., "Application of Ceria-based High-Selectivity Slurry to STI CMP for Sub-0.18mm CMOS Technologies," Proc. 4th Int. CMP-MIC, pp. 307-313, 1999.
- J.J. Gagliardi, T. Vo, "STI Polishing with 3M's Fixed Abrasive," 16th International VLSI Multilevel Interconnection Conf., pp. 223-228, 1999.
- J. Gagliardi, "3M Fixed Abrasive CMP Polishing of Semiconductor Oxide Films," 3M Internal Publication, Sept. 1999.
- E.F. Funkenbusch, "Slurry-free Oxide CMP Techniques," Semicon West, Educational Seminar Presentation, 1998.
- P. Van der Velden, "Chemical Mechanical Polishing Using Fixed Abrasives," Semi Europa, Educational Seminar Pres/, 1998.
- A. Sethuraman, W. Koutny, E. Shamble, "Comparison of Defectivity Between Slurry-based and Slurry-free Dielectric CMP," CMPUG, Dec. 3, 1998.
- E.J. Duwell, G.R. Abrahamson, "Performance of Coated Abrasive Lap Covers for the Grinding and Finishing of Ceramic Surfaces," Third International Grinding Conference, SME Tech. Paper MR88-598, 1988.
- E.J. Duwell, W.J. McDonald, G.R. Abrahamson, "Effectiveness of Aluminum Oxide and Silicon Carbide in 2-Body and 3-Body Abrasion of Ceramics and Metals," presented at the National Ceramic Soc. Meeting, April 1989.
- M. Tomozawa, "Oxide CMP Mechanisms," Solid State Technology, pp. 169-175, July 1997.
Tuyen Vo received his BS and MS in chemistry, and has been in the industry since 1987, working primarily in thin-film deposition and etch. He joined Rodel in 1998 as a CMP field applications engineer, supporting OEMs and semiconductor fabs on the West Coast. Rodel Inc., 725 San Juan Drive, #6, Sunnyvale, CA 94086; ph 408/245-8744, fax 408/245-8796, e-mail [email protected].
Todd Buley received his BS in electrical engineering technology, and has more than nine years of experience in the semiconductor industry, having held positions at Micron Semiconductor and Motorola. He joined Rodel in 1999 as CMP technical services engineer.
John J. Gagliardi received his BS in chemistry and MS in chemical engineering. He is a senior research specialist for 3M's Superabrasives and Microfinishing Systems Division, where he has 16 years of experience in material removal.