The shrinking feature size, introduction of new materials and .... S. P. Murarka and R. J. Gutmann, Chemical Mechanical Planarization of Microelectronic.
Mechanical Aspects of CMP David A. Dornfeld
Abstract - CMP has become a leading planarization technique in the manufacture of advanced integrated circuits (IC) chips and is one of the key fabrication processes. The shrinking feature size, introduction of new materials and impressive requirements for surface planarity and quality push the limits of the process. There is still a need to better understand the physics of CMP from the mechanical as well as chemical point of view. This paper reviews research work at Berkeley on mechanical elements of CMP and the development of a comprehensive model integrating both mechanical and chemical effects in CMP with the goal of predicting feature pattern evolution, design of optimal process recipes and help to understand the CMP-imposed limits on the reproducibility of deep sub-micron features. Research on sensors for process monitoring, process behavior from a hydrodynamic viewpoint, synergy between chemical and mechanical removal, step height reduction and an architecture for an integrated model including abrasive size and characteristic will be reviewed. Introduction CMP has become a leading planarization technique in the manufacture of advanced integrated circuits (IC) chips and is one of the key fabrication processes. The shrinking feature size, introduction of new materials and impressive requirements for surface planarity and quality push the limits of the process. There is still a need to better understand the physics of CMP from the mechanical as well as chemical point of view. The literature is rich in reports of research on various aspects of CMP. Traditional sources include early VMIC and CMP-MIC conferences, MRS meetings, ACERS meetings, the excellent reference by Steigerwald et al [1], a new volume edited by Li and Miller [2], etc. – that is, too numerous to mention but all adding to the background. Most of this work has emphasized primarily the chemistry of the process. Notable exceptions are the work of Boning and colleagues at MIT, Danyluk at Georgia Tech, Cook at Rodel, and Murthy at Arizona, for example. The work at Berkeley hopes to add to this building on a work in the past on precision machining including single point diamond turning process monitoring and loose abrasive processes of lapping and texturing. That is, processes that are perceived as being primarily mechanical removal-based. This is not to say that there are no chemical effects but that they are less prominent than the mechanical effects. CMP, of course, is not such a process. But, the mechanical effects in CMP need to be better understood. This paper reviews work in the areas of i. sensor systems for CMP process monitoring, ii. effects of mechanical properties of pads and slurry film thickness on CMP performance and iii. initial efforts at building an integrated model of the chemical and mechanical elements of CMP. CMP Process Monitoring The material removal rate in CMP is usually in the range of 100nm - 800 nm/min in thickness, which is extremely small compared to conventional machining such as grinding or diamond turning. Monitoring the material removal process in CMP planarization is, therefore, a difficult task using traditional sensors such as current or force sensors. Fukuroda et al. [3] detected and analyzed the signals of small vibration of the polishing head and related the signals to surface planarization, surface non-uniformity, pad wear etc. Because the vibration signal is not directly from the material removal in CMP, the sensitivity of this method is limited. There have been several additional studies on this subject [4-7]. These techniques, however, have some limitations in sensitivity and are not closely dependent upon process mechanics or suitable for reliable process monitoring. Acoustic emission is proposed as a means for process monitoring. Acoustic emission is the class of phenomena where transient elastic waves are generated by the rapid release of energy from localized sources within a material. CMP planarization of interlayer dielectric (ILD), basically, is a combination of chemical reaction and free abrasive machining in which the abrasives are allowed to rotate between the ILD surface and polishing pad and remove material by micro indentation or three body abrasion. When an abrasive particle penetrates the pad surface, the abrasive can become embedded in the pad and remove material by micro scratching similar to that in fixed abrasive grinding or two body abrasion. Our research has
indicated that the acoustic emission is closely related to the material removal process in CMP and therefore is a good sensing method for CMP process monitoring [8-10], Figures 1 and 2. Acoustic emission energy and other signal features are a very sensitive indicator of the degree and nature of contact between surfaces and will be the basis for the monitoring of the CMP process.
Figure 1. Acoustic emission in CMP – process variation.
Figure 2. AE sensor integration for process monitoring. MECHANICAL EFFECTS ON CMP PERFORMANCE The mechanical effects in CMP are conveniently referenced to hydrodynamic bearing theory through, for example, the Stribeck curve [11], Figure 3. Studies have shown the influence on material removal using differing velocities to affect the slurry film thickness and noting the influence on friction force, planarization, surface roughness and presence of surface defects. The influence of velocity on removal rate is seen in Figure 4 and follows the predictions of the Stribeck curve. Experiments with a “abrasive-less” slurry and “chemical-less” slurry highlighted the synergistic nature of the chemical and mechanical elements of CMP, Figure 5.
Boundary layer lubrication Elastohydrodynamic lubrication Hydrodynamic lubrication
Figure 3. Stribeck curve – slurry film thickness.
Figure 4. Material removal per sliding distance.
Figure 5. Chemical-mechanical synergy effects.
The effects of Hersey number (reflecting velocity, viscosity and pressure) is clearly seen when focussing on profile development in CMP [12, 13]. A series of tests on patterned wafers demonstrated the coupling between Hersey number (primarily velocity in this study), pattern density and step height reduction, Figure 6. Figure 7 shows the influence of Hersey number on WIWNU.
Figure 6. Line density effect.
Figure 7. Effect of Hersey Number on WIWNU.
Figure 8. Architecture of integrated model. INTEGRATED MODEL OF CMP Compared to processes which are generally considered “mechanical” (lapping process, for example, in which material removal is mainly caused by scratch, fracture, and crack through direct indentations) the general mechanism of material removal in CMP is chemical etching and mechanical abrasion. A number of theories exist for the exact mechanism (see reference texts [1] and [2].) Any model, to be successful, needs to integrate the effects of abrasive, chemical slurry, hydrodynamics, contact pressure and surface hardness, and pressure and velocity. The
basis for many of the CMP process models is Preston's wear equation, which states that the volumetric removal rate on a work piece, due to the relative motion between surfaces, is proportional to the bearing load and the relative velocity [14]. However, Preston's equation does not take into account a lot of the critical elements except as they are reflected in an empirically determined “coefficient.” A comprehensive model of CMP is under development at Berkeley which will allow the integration of the elements identified above with sufficient detail to allow process design at the most fundamental level (slurry pH, collodial effects, abrasive size, shape and iso-electric pH/zeta potential, temperature, material properties such as hardness, stress, electrochemical/chemical dissolution and passivation of surface constituents, etc.) A high level architecture of the model under development is shown in Figure 8. The model is comprised of detailed sub-models. The potential for “tuning” the process, through for example adjustment of the effective Preston’s coefficient to optimize WIWNU or WIDNU exists. Similarly, there is the possibility of “designing” the abrasive, as in a “fixedabrasive” pad, to optimize the material removal characteristics of the process in the presence of a specific material, copper for example.
CONCLUSIONS The opportunities for understanding more about CMP are endless offering a lot of opportunities for many researchers. The goal of our research is to build a rigorous model of CMP by including the appropriate mechanical effects, hydrodynamic phenomenon and chemical effects to predict the evolution of arbitrary profiles in the submicron range. To that end, the research will benefit from the assistance of other UC faculty in metrology (Prof. Costas Spanos at Berkeley) and chemical engineering (Prof. Jan Talbot at San Diego). The impact of a process model integrating the subtle effects of process chemistry and mechanical effects for predicting the performance of polishing a range of materials with a range of profiles is substantial. This type of model will be useful for process optimization (recipe optimization), process planning and device design. ACKNOWLEDGEMENTS Research work on CMP is supported by National Science Foundation through award NSF DMI-9813039, University of California SMART program under contract 97-01, and the industrial affiliates of the Laboratory for Manufacturing Automation at UC Berkeley. The assistance of researchers Andrew Chang, Jianfeng Luo, Edward Hwang and Dr. Yongsik Moon is appreciated. Helpful conversations with Professors Costas Spanos (UC-Berkeley) and Jan Talbot (UC-San Diego) are acknowledged. References
1. J. M. Steigerwald, S. P. Murarka and R. J. Gutmann, Chemical Mechanical Planarization of Microelectronic Materials. New York, John Wiley & Sons., 1997.
2. S. H. Li and R. O. Miller, Chemical Mechanical Polishing in Silicon Processing, San Diego, Academic Press, 2000.
3. Fukuroda, A., et al., "In-situ CMP monitoring technology for multi-layer interconnection", IEDM, pp. 469-472 (1995).
4. Meikle, S., et al., "Chemical mechanical polishing technique and method of endpoint detection in chemical mechanical polishing processes", US patent 5,439,551, August 8, 1995.
5. Salugsugan, I., et al., "Audio endpoint detection for chemical mechanical polishing and method therefor", US patent 5,245,794, September 21, 1993.
6. Yu, C., et al., "Chemical mechanical planarization (CMP) of a semiconductor wafer using acoustic waves for insitu endpoint detection", US patent, 5,240,552, August 31, 1993.
7. Yu, C., et al., "Acoustic method and system for detecting and controlling chemical mechanical polishing (CMP) depths into layers of conductors, semiconductors, and dielectric materials", US patent, 5, 222,329, June 29, 1993. 8. Tang, J. S., Dornfeld, D. A., Pangrele, S. and Dangca, A., “In-Process Detection of Micro-scratching during CMP using Acoustic Emission Sensing Technology,” Proc. TMS Annual Meeting,, San Antonio TX, TMS, February, 1998, and J. Electronic Materials, 27, 10, 1998, pp. 1099-1103.
9. Tang, J., Unger, C. , Moon, Y. and Dornfeld, D. A. , “Low-k dielectric material chemical mechanical polishing (CMP) process monitoring using acoustic emission”, Proceedings of Low-dielectric constant materials and application in microelectronics, Material Research Society, 1997. 10. Moon, Y., Lee, Y. and Dornfeld, D. A. , “Study of Slurry Chemical Influence in Ductile/Brittle Transition Depth in Chemical Mechanical Polishing (CMP) using Acoustic Emission Sensor,” 1st International Conference of the European Society for Precision Engineering and Nanotechnology (EUSPEN), May 31st-June 4th, 1999, Bremen, Germany. 11. Moon, Y. “Mechanical aspects of the material removal mechanism in chemical mechanical polishing (CMP),” Ph. D. Thesis, Mechanical Engineering Department, University of California, Berkeley, CA, 1999. 12. Chang, A., “Profile development in CMP as influenced by Hersey Number,” Program Review Meeting- SFR, UC-Berkeley, April 20, 2000. 13. Moon, Y., Chang, A., and Dornfeld, D. A., “The Effect of Slurry Film Thickness Variation in Chemical Mechanical Polishing (CMP),” Proc. 102nd Annual Meeting and Exposition, American Ceramic Society, St. Louis MO, April 30-May 3, 2000. 14. Preston, F. W., "The theory and design of plate glass polishing machines," J. Soc. Glass Tech., vol. 11, pp. 214256 (1927). 15. Dornfeld, D. A., “Chemical Mechanical Polishing,” Program Review Meeting- SFR, UC Berkeley, April 20, 2000.
The Effect of Slurry Film Thickness Variation in Chemical Mechanical Polishing (CMP) of Patterned Oxide Wafers Andrew Chang Sponsored By: National Science Foundation (DMI-9813039) UC-SMART SM 97-01 Abstract— Material removal and step height reduction (SHR) of patterned oxide wafers from Chemical Mechanical Polishing (CMP) are significantly influenced by the wafer-pad contact mode determined by the slurry film thickness between the wafer and the polishing pad. The slurry film thickness is proportional to the Hersey number defined as relative velocity of wafer times slurry viscosity divided by normal pressure on the wafer. In the low Hersey number regime, patterned oxide wafers have high material removal and SHR and in the high Hersey number regime, they have lower material removal and SHR. It is believed that this phenomena results from the wafer-pad contact modes determined by the increasing slurry film thickness. It is also found that the SHR decreases with increasing pattern density of each die. Introduction Many researchers have acknowledged the significance of slurry film thickness in CMP. Previous research has been conducted to calculate the variation of slurry film thickness in terms of rotation speed and slurry viscosity [1]. It was determined that understanding the behavior of slurry film thickness in CMP will be fundamental in the investigation of material removal mechanism and the development of a process model for the CMP process [1-3]. In this study, the Hersey number [4] is considered as a main process variable to determine the slurry film thickness. Among the three parameters (velocity, viscosity, and pressure) in the Hersey number, the relative velocity and the normal pressure were varied so that wafers were polished under different relative Hersey numbers. To identify the influence of wafer-pad contact mode on wear rate, a material removal per unit sliding distance of the wafer was associated with a relative Hersey number. To determine the efficiency of planarization of the IC pattern structure, the step height reduction (SHR) was also correlated with the relative Hersey number. IC pattern density is known as one of the causes for non-uniform material removal rate of patterned wafer. In this study, different pattern densities were chosen from each die on the patterned oxide wafer, and SHR of each the pattern features was monitored with polishing time. 1. Characteristics of slurry film thickness It has been previously shown that the slurry film thickness increases with the relative velocity of the wafer, slurry viscosity, and decreasing normal pressure on the wafer (Figure 1). Thus, the Hersey number calculated from the process, defined as Hersey number = viscosity x velocity /pressure. is a useful indicator of slurry film thickness and, thus, it is a key parameter to characterize the wafer-pad contact mode (i.e. direct contact, semi-direct contact, and hydroplane sliding). The slurry film thickness is proportional to the square root of the wafer velocity and the Hersey number if pressure and viscosity are kept constant in the hydrodynamic lubrication regime [2][6] 2. Experimental Setup A Strasbaugh 6EC single-head CMP tool was used for all polishing and the experimental conditions are summarized in Table 1. To characterize the slurry film thickness, a relative Hersey number was calculated for each of the operating conditions as the velocity divided by pressure, and it assumes that the viscosity of the slurry remained constant for all polishing conditions. This parameter is used only as a comparative metric of slurry film thickness among the polishing conditions, with higher numbers assumed to correspond to larger film thickness. On each wafer, three dies (a center, intermediate, and edge die) were selected and monitored throughout the experiment using a step height profilometer. The data from each line density pattern was averaged over all three dies to decouple any effects from within-wafer nonuniformity. 3. Results Material removal The material removal of the “field oxide” thickness for a unit sliding distance was calculated with respect to the relative Hersey number to identify the dependency of material removal on the slurry film thickness, Figure 2.
Figure 1. The Stribeck curve and the variation of film thickness with the Hersey number.
Table 1. Experimental Conditions for experiment. Wafers Down Force Back Pressure Table / Carrier Speed Slurry Type Slurry Flowrate Elapsed Polish Time Relative Hersey Number
Five 4-inch wafers with oxide patterned (3, 5, 8) psi 0.5 psi (30/27, 50/45, 80/72) RPM Cabot Semi-sperse D7000 150 ml/min (15, 60) seconds (0.8, 1.2, 2.1, 3.5, 5.6)
The material removal per sliding distance decreased as the relative Hersey number (and thus the slurry film thickness) increased. The results indicate that polishing that occurred in the low relative Hersey number regime had a thin slurry film between the wafer and the pad. The material removal for a unit sliding distance is more aggressive in this regime due to the increased actual contact between the wafer and pad surface. It is also believed that as the relative Hersey number increases, the actual contact area between the wafer and pad decreases due to increasing slurry film thickness between the wafer and the pad.
Figure 2. Material removal per sliding distance with relative Hersey number
Figure 3. Definition of step height reduction and line density.
Planarization To quantify the effect of slurry film thickness on planarization, step height reduction (SHR) was calculated. SHR and line density are defined and illustrated in Figure 3. Ideally, SHR is 1 for the complete planarization of the pattern feature. The step heights were reduced with polishing time and the SHR decreased with increasing line density. The relationship between the SHR and line density is shown in Figure 4. It is believed that the line density effect results from the different local pressure applied on the features with different line density, Figure 5. For low density features, the local pressure of the pad (and thus the abrasive particles trapped between the pad and feature) is larger than the local pressure that is distributed over high density features. The increased local pressure on the low density feature is believed to cause increased SHR compared to the high density features.
Figure 5. Average SHR decreases with line density.
abrasive particles
Figure 4. Representative step height profiles for 3 features with different line densities (LD).
Figure 6. Local pressure variation between pad and feature for low density (left) and high density (right).
4. Conclusions The following conclusions can be drawn from this work: • The slurry film thickness between the wafer and the pad is proportional to the velocity of the wafer and the Hersey number. • The effect of slurry film thickness variation on the CMP process performance (material removal and planarization) is significant. • It may be useful to control and optimize the slurry film thickness to maintain process stability 5. References [1] S. Runnels, et al., “Tribology Analysis of Chemical Mechanical Polishing,” Journal of Electrochemical Society, Vol. 141, No. 6, pp. 1698-1701, June 1994. [2] S. Runnels, “Feature Scale Fluid-based Erosion Modeling for Chemical Mechanical Polishing,” Journal of Electrochemical Society, Vol. 141, No. 7, pp. 1900-1904, July 1994. [3] S. Runnels, “Advances in Physically Based Erosion Simulators for CMP,” Journal of Electronic Materials, Vol. 25, No. 10, pp 1574-1580, 1996. [4] M. D. Hersey, Theory and Research in Lubrication, John Wiley & Sons, 1966. [5] I. M. Hutchings, Tribology – Friction and Wear of Engineering Materials, CRC Press, 1992. [6] Y. Moon, “Modeling the Lapping Process Based on Hydrodynamic Effects,” Master degree thesis, UC Berkeley, May 1996. [7] D. Hetherington, et al., “Characterizing Variations in ILD CMP planarization Rates Using Atomic Force Microscopy,” Proceeding of CMP-MIC Conference, pp. 74-81, Feb 1996. [8] Y. Moon and D. A. Dornfeld, “Mechanical Properties and Relationship to Process Performance of the Polishing Pad in Chemical Mechanical Polishing(CMP) of Silicon,” Proc. of the ASPE, pp. 83-87, Monterey, April 1998.
Simulation Software Tools for CMP Modeling Andrew Chang, John Imamura, Shin Akiya Sponsored By: LMA Industrial Affiliates UC Undergraduate Research Opportunity Program Abstract— Current research efforts in the area of Chemical Mechanical Planarization have led to the development of various models relating to aspects of the process. The application of these models is unfortunately difficult to realize because of their complex form or necessity for in-depth understanding about subtle process characteristics in order to successfully apply the model. We propose here to develop a suite of software tools to help demonstrate these CMP models. While the software will be aimed to help researchers simulate and characterize the distinctions between various models, it will also provide “educational” value by allowing users without specific CMP domain knowledge, namely students and outside researchers, to interactively simulate process models.
Introduction Research on modeling and simulation of the Chemical Mechanical Planarization (CMP) process has become a primary research focus in the Laboratory for Manufacturing Automation (LMA). To date, several models have been developed by the LMA as well as other institutes to try to characterize and predict complicated chemical and mechanical interactions that occur during CMP. These models can vary dramatically from closed-form equations to probability-based predictions to numerical simulations. Every new evolution of these models may bring a subtle enhancement or improvement in the modeling capabilities. Unfortunately, modeling work is often difficult to evaluate, even at a high level, without an in-depth knowledge of process conditions and formulations. We propose to develop software for the evaluation of these CMP models. The major emphasis of the software will be to design and implement interfaces and modules that effectively demonstrate the usage and limitations of each of the models. User-centered Design The design of the user interface will be a primary focus for the development of the software. It will essentially determine the usability of the software and its ability to successfully communicate the subtlety of each process model. Figure 1 shows the overall development methodology that will be used to facilitate effective software design. A feasibility study and requirement analysis have already been completed. The design study, which is being planned, will be used to understand the habits and needs of the users. A variety of user interface design processes, such as task analysis, low-fi prototyping, user interviews, and surveys, will be used to identify what features and specifications will be necessary for the software to be effective. Furthermore, this information will allow the design of the user interfaces to be critiqued and modified early in the design process. Functional Specifications The functional specifications of the software tools have not yet been finalized. The design approach for this project will be focused on user-centered design, which necessitates flexibility in the development cycle. Initial design planning, however, has brought the attention to these requirements: • • • • • • •
Users should be able to select any model for simulation The user should be able to specify input parameters (such as operating conditions, wafer material, etc.) to the simulation Some output from the model should be displayed graphically in terms that can interpreted by the user The software should be usable under at least two different operating “modes”: educational and advanced, which will correspond to the complexity of the simulation procedures The software should be able to “save” and “recall” previous simulations The process modeling shall be confined to “modules” which must conform to an API (application protocol interface) that will allow the process models to be developed independently from the simulation software The software should be available on-line
Figure 1. Development process and methodology. Process Database Development The process database will contain information relating to nearly every aspect of the CMP process (Fig. 2). This information may include parameters and properties relating to consumables (slurry, pad, conditioning pad), the polishing tool (eccentricity of polishing head rotation, fluid nozzle location, slurry waste treatment data), etc. The purpose of the database will be to organize this information in a logical and hierarchical format allowing the software to query the database and collect whatever information is needed for the modeling. Sub-tasks in this development plan will include implementation of the process database using Java and Java Servlet languages.
Figure 2. Schematic diagram of the software architecture. Conclusion This project is supported in part by the Undergraduate Research Opportunity Program (URO) which enables undergraduates to participate in on-going research projects. Key sub-tasks and deliverables have been defined as part of the project and the functionality of the software will be demonstrated using simple modeling equations, such as the Preston Equation. The development of the other process models will be an on-going process in conjunction with other researchers in the LMA.
Fixed Abrasive Design
Author: Edward Hwang Sponsored By: UC SMART Abstract—A new technology for “Fixed Abrasive” for Chemical Mechanical Planarization (CMP) is discussed and a method for fabrication using MEMS techniques is presented. Introduction Fixed abrasive (FA) has become a new issue in the Chemical Mechanical Planarization (CMP) area. The speed of development in the semiconductor manufacturing industry moves quite rapidly, so that feature sizes in the sub-micron are needed. In order to address challenges, shallow trench isolation (STI) is used instead of local oxidation of silicon (LOCOS) for isolation purpose. It turns out that FA is very compatible with STI for the following reasons; (1) FA has a great selectivity to topography (~100:1) (2) FA has almost 1:1 selectivity between oxide and nitride. Another good feature of FA is that the only chemistry needed is pH-adjusted water, this is the potential to make the CMP process environmentally benign, even though there are still wear particles. It is also worthwhile to note that no conditioning is required. Another aspect of FA of interest is the point of material removal mechanism or wear. Abrasive wear can be defined as material removal by hard particles or asperities on a surface moving relative to another abraded surface due to contact. There are 2 kinds of abrasion: two-body abrasion and three-body abrasion. As known, in the former case, the abrading material is fixed to one of the surfaces and, in the latter case, it is loose between the two moving surfaces. FA is a typical case of two-body abrasion. The abrasive wear depends on a large number of parameters. Among them, the shape of the abrading material is a main concern in this work. Attack angle and dihedral angle dependence in abrasion are already well known in several investigations by single abrasive experiments. [1] Even though the possible geometries are limited because of the crystal structure of silicon itself, several distinct shapes of abrading material will be fabricated using the state of the art MEMS technique.
Fabrication Technique for Fixed Abrasive To produce the desired surfaces, (100) silicon wafers 4 inch in diameter were treated by a micro mechanical etching technique consisting of several steps. The manufacturing process involves the following major steps: Firstly, the wafers are wet oxidized to obtain a SiO2 layer of approximately 1µm. Secondly, a uniform pattern of oxide islands is produced by standard photolithographic techniques. This SiO2 spot pattern resists etching and thus determines the position of the individual tips and also the tip shape. Finally, the unprotected silicon is anisotropically etched, gradually undercutting the oxide discs which eventually fall off, leaving a uniform structure of silicon tips. Following is the fabrication technique for fixed abrasive based on MEMS technique. 1. 2. 3.
Start with p type bare silicon Wet sink cleaning Thermal oxidation Oxide Si
4. 5.
Dehydrate wafer Place wafer in HMDS vapor
6.
PR coating PR Oxide Si
7. 8.
Soft bake Photolithography PR Oxide Si
9.
Develop PR Oxide Si
10. Hard bake 11. Anisotropic etching for SiO2
PR Oxide Si
12. Remove PR
Oxide Si
13. Silicon etching
Fig. 1 Process Flow of Designed Abraded Fabrication
dihedral angle
attack angle
Fig. 2 Schematic Diagram of Model Asperity
Fig. 3. Fabricated Abrasive (from ref. 10) Notes on the Fabrication Technique The most critical step is #13. As mentioned earlier, because of the crystal structure of silicon, if wet etching using KOH, EDP, or TMAH is done, a V-groove having only a limited angle (54.7 degrees) will be created. Another possibility is to use plasma-etching process. But, plasma etching will create more or less a very straight sidewall; it is also difficult to create arbitrary geometries. So, strategic combinations of the two etching processes will be adopted to generate arbitrary shapes.
Oxidation Sharpening Another way to control the abrasive shape is to use so called oxidation sharpening. The oxidation rate depends on the geometric shape of the silicon, exactly speaking; the oxide is thinner on both concave and convex corners than it is on flat regions because of stress due to volume expansion. So, after oxidation process, HF dip will strip the oxide, making the silicon structure more sharp, which will end up with making different attack angle.
Conclusions The fabrication process for various designed abrasive geometries will be simulated with a commercial simulation package (TSUPREM) as a fabrication feasibility check. Pin-on-disc tests using the fabricated designed abrasives (with varying geometries) will be conducted in order to investigate the relationship between the abrasive geometry and material removal mechanism. References [1] K. Kato, K. Hokkirigawa, T.Kataba, Y.Endo, “Three Dimensional Shape Effect of Abrasive Wear,” Journal of Tribology, Transactions of ASME, July 1986, Vol. 108 [2] Kenneth E. Bean, “Anisotropic Etching of Silicon,” IEEE Transactions on Electron Devices, Vol. ED-25, No. 10, October 1978 [3] D.B.Lee, “Anisotropic Etching of Silicon,” Journal of Applied Physics, Vol. 40, Number 11 October 1969 [4] Ylva Backlund and Lars Rosengren, “J. Micromech. Microeng. 2 (1992) [5] Carola Strandman, Lars Rosengren, Hakan G. A. Elderstig, and Ylva Backlund, “Fabrication of 45˚ Mirrors Together with Well-Defined V-Grooves Using Wet Anisotropic Etching of Silicon,” Journal of Microlectromechanical sustems, Vol. 4, No. 4, December 1995 [6] Wen-Tien Chang Chien, Liwei Lin, Yi-Chung Lo and Chia-Ou Chang, “Fabrication of 109.5˚ Micro V-Gr ooves Using a Two-Step Anisotropic Etching Technique,” DSC-Vol. 66, Micro-Electro-Mechanical Systems(MEMS) – 1998 [7] K.K.Chin and R.B.Marcus, “Field Emitter for Vacuum Microelectronic Devices,” J. Vac. Sci. Technol Vol. A8, 1990 [8] James D. Plummer, Michael D. Deal, and Peter B. Griffin, “Silicon VLSI Technology,” Prentice Hall, 2000 [9] Rickard Galin, Henrik Bjorkman, Pelle Rangsten, Staffan Jacobson, “Designed Abrasive Diamond Surfaces,” Wear 233-235 pp. 387-394, 1999
Investigation of Material Removal Mechanisms in Oxide CMP by Scratch Test
Author: Edward Hwang Sponsored By: UC SMART Abstract— At present, the material removal mechanism taking place on a wafer surface during CMP is not well understood. In order to make a breakthrough in the CMP area, fundamentals of material removal mechanism must be better understood. As a part of this, a scratch test was introduced. In this paper, using a scratch test, some characteristics of CMP treated wafer surfaces were studied. Introduction Chemical Mechanical Polishing (CMP), as it is, remains at an empirical stage of understanding, and is more like an art than a technology. The exact nature of the material removal mechanism taking place during a CMP process is the key to understanding the fundamental issues in CMP. Tseng et al [1] have proposed that the stress levels occurring on the wafer would be approximately 102 orders of magnitude less than the yield stress of the material removed. It is possible that some chemical interaction between the wafer and the slurry is responsible for this phenomenon. However, this reaction has yet to be successfully quantified. Trogolo et al [2] have demonstrated via TEM experiments that there exist two different layers on the wafer. The first layer is "a chemically weakened layer" typically on the order of a few nanometers in thickness. This layer is a highly hydrated and loosely bound network of silica that has a lower density. The second layer is "a plastically compressed layer" which is on the order of approximately 20-30 nanometers. Unlike the chemically weakened layer, this layer is characterized by a plastically compressed network of silica atoms, which corresponds to a higher density (see Fig. 1). This exactly matches the suggested oxide CMP mechanisms proposed by Tomozawa. In that paper, he claimed that the likely oxide CMP mechanism to be oxide hydration during the plastic deformation caused by abrasives. Plastic deformation is assisted by frictional heating, and the removal of the resulting softer hydrated surface layer takes place by the plowing action of the abrasive particles. In this paper, based on these early reports and the scratch test, it is proposed that understanding the characteristics of the chemically weakened layer can help in understanding CMP. To address these issues, a scratch test using a single point diamond turning (SPDT) machine set-up was used. A diamond tip contacts the CMP-processed wafer surface, causes indentation depending on the load and moves along the surface to scratch the material. The depth of scratch can be well adjusted to extract the characteristics of the layers. An AE sensor attached to the work material is used to monitor the scratch process and aid in the characterization. The acoustic emission (AE) sensor signal is analyzed to capture the features of each layer. In addition, as a reference, a silicon dioxide wafer before CMP (pre-CMP wafer) will be also scratched under the same conditions. This will give some insight to the material removal mechanism in CMP. bulk not affected by the process
Si
polishing pad
layer 2(order of 20 nm ) plastically compressed network – higher density
layer 1(order of a few nm ) highly hydrated, loosely bound network – lower density
Fig. 1 Cross Sectional View in Actual CMP Situations
Experiment Set-Up As mentioned above, a single point diamond turning (SPDT) machine set-up was used for this scratch test. There were 2 kinds of workpiece: pre-CMP wafers and post-CMP wafers. The sampling rate of the AE signal was 50KHz and tilt angle for the adaptor was 0.06 degrees.
Fig. 2 Single Point Diamond Turning Machine Set-Up for Scratch Test
Result with Single Diamond Turning Machine As seen in Fig. 3, when the diamond tool indented the pre-CMP wafers, no discontinuity in AE raw data was seen. The AE raw data was just proportional to the depth of scratch. On the other hand, when the post-CMP wafers were indented, a clear transition layer was seen which is due to the plastically compressed layer in Fig. 4. The SEM pictures for post-CMP wafers corresponding to a specific depth of scratch were also shown in Fig. 5. Based on these results, it was determined that there exist two different material layers during CMP. sensor 1.0 0.8
AE signals (volts)
0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
time(s)
Fig. 3 AE raw data vs. Depth of scratch for pre-CMP wafers sensor 1.0 0.8 0.6
AE Raw Signals (volts)
0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
time(s)
Air-cut + Region 1
Region
Bulk
Fig. 4 AE raw data vs. Depth of scratch for post-CMP wafers
sensor
Fig. 5 SEM pictures vs. Depth of Scratch for Post-CMP wafers Future Work There are two things that need to be done. First is to replicate this result with another machine, Kugler fly cutting machine, to show that the results shown above are independent of the machine. Second is to use an AFM as a scratch test machine to better represent the size of the abrasive process in CMP. In order to come closer to actual CMP situations, the scratch tool itself needs to have a very sharp tip. And, as well, a very stringent control of the motion is required. The AFM is expected to be an ideal instrument from this point of view.
References [1] Wei-Tsu Tseng, Chi-Wen Liu, Bau-Tong Dai, Ching-Fa Yeh, “Effects of Mechanical Characteristics on the Chemical-Mechanical Polishing of Dielectric Thin Films,” Thin Solid Films 290-291(1996) 458-463 [2] J. A. Trogolo, K. Rajan, “ Near Surface Modification of Silica Structure Induced by Chemical/Mechanical Polishing,” Journal of Materials Science 29(1994) 4554-4558 [3] Minoru Tomozawa, “ Oxide CMP Mechanisms,” Solid State Technology (Jul. 1997)169-175
Energy Consumption Model for semiconductor manufacturing equipment By: Paal Joergensen Henrik Passmann Sponsored by: Applied Materials
Abstract: The goal of this project is to build a model that Applied Materials and others in the semiconductor industry can use to identify the energy usage of fabrication. At the same time the model should be a tool for predicting future energy consumption in the semiconductor industry.
Background: Environmental aspects related to energy have been on the agenda for quite a while now. There have been research concerning energy and pollution from coal and fossil fuel and other topics. Although the total increase of hazardous emissions has gone down the last few years, the emissions are still high and affecting the global warming. In 1997 the total U.S greenhouse gas emissions were, according to United States Environmental Protection Agency, 1823.6 million metric tons of carbon equivalents. One issue that has not been fully studied is the possibility for energy usage reduction in the industry, and the environmental impacts of this reduction. Environmental factors have also become an emerging area of concern in semiconductor manufacturing. In the recent years the energy usage of the semiconductor industry has increased dramatically. The energy and water demands placed on natural resources in order to produce semiconductors are significant. As the complexity and size of the semiconductor facilities (fabs) have grown, so have these demands. New facilities can use up to 30 to 50 Megawatts of peak electrical capacity, enough to power a small city. So both the economical and environmental savings can be quite important if one can reduce this usage. The semiconductor industry is working on reducing their emissions, but the fact is that energy usage is also a big contributor to global warming. So in order to reduce the emissions even more, one have to look at the energy used in the production of semiconductors. It is important to keep in mind, however, that the efficiency savings available represent a very small share of production costs. One industry representative indicated that the cost savings achieved for one year’s worth of energy savings at a fabrication site is equivalent to the total cost of one day production. Put another way: energy accounts for about 1-2% of the costs of semiconductor production. Concerning the amount of money at stake, it is easy to understand industry reluctance to invest in unfamiliar efficiency technologies, which could require costly production downtime. Nevertheless, the relative amount of resources that the semiconductor industry uses should be kept in perspective for efficiency changes. Although the cost savings may not be noteworthy for the facility usage, the water and energy resources used are often equivalent to those used by small city or a number of towns. The overall impact on the local environment can be substantial, as can the efficiency changes made by the facilities. One small change or process improvement by a production unit can equal what it may take all other industries in a city or town to implement.
In many industries, including microelectronics, energy cost has not been separated from the general operating costs, making it harder to track both the costs and benefits of becoming more energy efficient. For the semiconductor industry, it has been estimated that as least half of the energy used is wasted. This waste could be converted into a profit with returns exceeding 30% return on investment (ROI). As an interesting savings/investment comparison, the risk of energy efficiency investments is slightly less than the risk involved in investing in long-term Treasury bonds. What does saving energy return to a facility? There are many benefits beyond saving money on electric bills. Energy efficiency can provide a fab with more flexibility than competitors who do not address efficiency issues, since they are less susceptible to price changes in resources. The bottom line is healthier; overhead is lower; and there is a lower overall demand for natural resources, making them more desired in communities. So a reduction in the energy consumption can reduce the global warming potential (GWP), and at the same time affect the cost of production. Together with Applied Materials, the green lab of LMA has established a research program for studying energy consumption in the semiconductor industry. The goal is to build a model that can be used by the producers of semiconductors to predict the energy usage of a new facility, but at the same time also identify where the potential for both energy and costs are substantial.
The Model: The Energy Consumption Model gives a general view of how much energy that is consumed during production of semiconductor chips. The energy usage during the different steps of producing a semiconductor is measured, and the most energy consuming steps or processes, are included in the model. Measurements have been performed on different kinds of tools, manufactured by different vendors; therefore the model gives a general view. It identifies the processes in the production that consumes most energy. To get an understanding of own tools though, each manufacturer of semiconductors has to measure their own processes. This is because of all the different recipes that are used throughout the industry. Our model can be used as a guideline for identifying which tools consuming most energy from a general point of view, and to show where the possibilities for energy reduction are the greatest. Included in the model are also the energy rejected to the process cooling water (PCW) and the power factor (not shown here). Through the efficiency of energy rejected in the cooling water one can establish analyses of the amount of cooling water needed in the process. There are some quite interesting possibilities for cost and energy savings. By increasing the temperature difference in the cooling water one can reduce the amount of water needed, and therefor also the size of the pumps. The power factor is included because of the importance of energy efficiency. It shows how efficient the facilities use available power. This is quite interesting for customers buying the tools! The lower the power factor the greater is the out-of-phase component of reactive current. This cannot do work but requires sufficient capacity to carry it out through the distribution system. So even though the out-of-phase current cannot do work it has real costs. This is why it is so important for the power suppliers that the industry has a good power factor. In order to get the manufacturers to strive for a better power factor they have penalties for having a low power factor. For the semiconductor industry it is quite interesting to look at the power factor because of these costs. An increase in power factor can reduce the costs substantially. Therefore we have chosen to include the power factor in our model. It gives a view of how well the tool or system is performing. The energy model itself is presented in a spreadsheet. Important energy consuming process steps are listed with kilowatt-hour per wafer pass (kWh/wp). The total process power and the total idle power are listed, and then the total amount of kWh can be calculated.
The figure below is a sample from the model. The table shows the total and idle power consumed during production of one kind of chip:
(The numbers here are fixed) # Tool & Process Tools 18 13 3 8 13 7 8 10 7 21 6 2 3 23 54 2 23
Etch: DPS Etch: Metal Silicon Etch Implant, SWIFT CD- SEM DR SEM HDP-CVD RTP MCVD CMP PECVD SACVD EPI PVD Pegasys PFC Abatement TPU PFC & F2 Abatement TCS F2 Abatement
Average Process Power 160.0 kW 80.0 kW 90.0 kW 40.0 kW
Idle Idle Time Avg.Idle Power(KW) KW/Month
Process Power
120.0 kW 60.0 kW 67.5 kW 30.0 kW
40.0% 40.0% 40.0% 40.0%
864 312 81 96
1728 624 162 192
2.2 kW 2.1 kW 80.0 kW 55.0 kW 65.0 kW 8.0 kW 45.0 kW 38.0 kW 90.0 kW 65.0 kW 1.2 kW 2.1 kW
1.7 kW 1.6 kW 60.0 kW 41.3 kW 48.8 kW 6.0 kW 33.8 kW 28.5 kW 67.5 kW 48.8 kW 0.1 kW 2.1 kW
40.0% 40.0% 40.0% 40.0% 40.0% 40.0% 40.0% 40.0% 40.0% 40.0% 40.0% 0.0%
8.58 4.41 192 165 136.5 50.4 81 22.8 81 448.5 2.16 0
17.16 8.82 384 330 273 100.8 162 45.6 162 897 38.88 4.2
1.3 kW
1.3 kW
0.0%
0
29.9
Sub Total Idle Power: Generating Requirement Total Requirements: Total (Extrapolated)
2545.4 kW 2.55 MW 7.70 MW Fab: 19.26 MW
5159.4 kW 5.16 MW
The numbers in the model have to be obtained through measurements on live operating tools, using a meter that includes all the phases and currents appearing when the tool is operating. In this model we have used The Reliable Power Meter (RPM). It measures the consumption of power by the process tools. It measures real, reactive, apparent and harmonic power. These measurements are displayed as plots over time. The Reliable Software (RS) can then calculate the power factor.
Future Work: The hardest part of building a model is to decide what it should comprise. The development of this model is not an exception. Which steps to be included is not yet decided. The hope is to include the steps of semiconductor manufacturing that consumes the most energy for one type of chip. Then later expand the model to include different kinds of recipes used in the industry. The model is in its developing phase and will later also include the GWP impact of the energy usage (in MMTCE). The measurements used in this model is mostly done on tools with one process chamber, so it is not quite fair to tools which has more than one chamber. A future edition of the model will also comprise tools with more than one chamber.
References: Pasific Northwest Pollution Prevention Resource Cener, “Topological Reports, Energy and Water Efficiency for Semiconductor Maufacturing”, 1999, www.pprc.org/pprc/ pubs/topics/semicond/semicond.htm Bill Howe, Michael Shepard, Amory B. Lovins, Bristol L. Stickney and David Houghton, “Drivepower”, Technology atlas series 1996 EPA,
United
States
Environmental
Protection
www.epa.gov/globalwarming RPM, “The Reliable Power Meter handbook”
Agency,
“Global
warming”,
The Environmental Value Systems (EnV-S) Analysis Author: Nikhil Krishnan Sponsored by: Applied Materials, NSF/SRC Abstract The complex process sequences carried out in a modern semiconductor fabrication facility impede effective environmental decision making. Additionally, environmental issues are intertwined with cost, productivity and performance factors. A decision and design support tool can enlighten environmental decision-making and facilitate environmentally benign process and equipment design and selection. The Environmental Value Systems (EnV-S) Analysis (Figure 1) is a design tool under development, to evaluate cost, performance and ESH impacts of semiconductor manufacturing process sequences through an aggregated analysis of individual system components. In the past year of work, the approach behind the EnV-S has been detailed and the EnV-S is used to inform system selection decisions for copper Chemical Mechanical Planarization (CMP) effluent treatment and water recycling systems. Introduction Human industrial activities exact a staggering environmental and health burden. Manufacturing sector industries are typically large contributors to impacts, and have been among the first industrial sectors to come under regulatory scrutiny. In recent years, with rapid growth in the electronics and computer sectors, environmental and health issues associated with the semiconductor manufacturing industry are growing in importance. Semiconductor processes are typically highly resource intensive per dollar of output (facilities consume large quantities of water, energy, chemicals and materials) and also generate large quantities of waste (global warming gases, wastewater containing metals/solids/organics, hazardous solids and sludges, etc). A snapshot of the environmental performance of the industry in terms of a select few resource inputs and environmental impacts appears in Table 1. Metrics were selected to (a) represent a wide range of inputs and outputs; (b) provide simple measures of overall environmental burden; and (c) be based on available data (from EPA reporting requirements, public databases, etc.) Outputs from the Fab Source Inputs to the Fab Liquid Waste 75 Gal/in^2 [1] Water 30 gal/in^2 Hazardous Waste 0.1 Kg/in^2 [1] Electricity 10 KWhr/in^2 GWP 2.6 KgCE/in^2 [2, 3, 4] Chemicals 0.2 kg/in^2 PFC's 0.9 KgCE/in^2 [4] Toxic Releases 0.01 Kg/in^2 [5] Table 1. Selected Facility Environmental Metrics per Square Inch of Si
Source [3, 6, 2] [2, 3] [1, 3]
Beyond these generic impacts, specific reasons for analyzing the semiconductor industry appear in Table 2. Categories Specific Reasons for Environmental Concern Impacts on Different Temporal Rapid growth in the semiconductor industry scales Rapid changes in processes, chemicals, technology Evaluation of true societal environmental impacts, including complete lifecycle impacts. Increased importance to resource conservation and waste prevention/reduction Impacts on Different Spatial scales Local issues: Discharges (eg: Cu in the San Francisco Bay) Local issues: Availability (eg: Electricity in Japan) Cost: For facilities groups, environmental costs are significant. Less so for process groups (manufacturing costs dominate) Specific issues related to the Nature Large chemical inventories of the Industry (across temporal and Unknown toxicities spatial scales) Chemistries not completely characterized Chemical Exposures: Worker Health and Safety Table 2: Causes for Environmental and Health concerns in Semiconductor Processing
The complex and rapidly changing processes in semiconductor manufacturing, however, impede effective environmental decision making. Additionally, environmental issues are intertwined with cost, productivity and performance factors. A decision and design support tool can enlighten environmental decision-making and facilitate environmentally benign process and equipment design and selection. Architecture The EnV-S [6-8] consists of 3 shells to define, describe and characterize semiconductor manufacturing systems (Figure 1). The first shell is the Process Modeling Layer, and it serves as a warehouse for models that describe the operation of individual units (eg. filters, ion exchange beds) in a system. The second shell is a Sequencing Layer, in which the units are sequenced appropriately to describe a system. The final shell is a Decision Support Layer, which formulates an overall decision structure for a problem by enumerating decision nodes to develop a decision tree. Decision and Characterization Shell Sequence information
Strategic analysis
Sequencing Shell
Process Modeling Shell and Database
Figure 1: The Environmental Value Systems (EnV-S) Analysis The EnV-S supports equipment selection/design through two routes of analyses. The first involves a strategic analysis aimed at paring down a large design space, identifying problem areas and suggesting possible solutions. The second route of analysis is a detailed analysis, using all three layers of the EnV-S, to understand the cost, performance and environmental impacts of specific semiconductor manufacturing system configurations. Strategic Analysis - After reviewing the available technologies for removing copper and recycling water, a decision tree structure was developed to determine the most promising treatment options before modeling and characterizing the systems in more detail. Particular attention was given to the possibility of water recycling systems for copper CMP wastewater, to feed recycled water back into the process. The first decision (Figure 2) is whether to separate the drains on the CMP tool. Since the effluent stream during the polish cycle contains a high copper concentration and the stream from the rinse cycle is relatively pure, a valve can be added to the process tool to separate the polish and rinse waste streams. Next, the CMP streams can either be treated separately or combined with other CMP process effluents (oxide and tungsten). For each waste stream there then exists the possibility to treat and discharge, treat and recycle or treat and send to facilities gray water. Finally, treatment systems could treat waste streams at various stages – Point of Use (POU), Facility Scale, or Local (for 5 tools). The total number of options generated from these choices is approximately 200. A three layered approach is adopted to pare down this options space. First, a set of design rules are used to eliminate impractical or technically infeasible options. This reduces the number of options to 17 (Figure 3). Secondly, approximate costs for the 17 options can be generated based on direct lookup from information in the EnV-S process models. This further reduces the number of options to 4 (Figure 4). Third, a detailed EnV-S analysis can be run on the four final options, using the process modeling, system sequencing and the design characterization layers. From the decision tree analysis, we gather that facility-wide systems are generally favorable. Furthermore, all the final options resulting from this analysis recycle or reuse water in some way. Water costs are therefore a significant driver for copper CMP waste treatment decisions. It is also seen that (1) system
configurations involving treating copper CMP waste separately or combining the waste with other CMP waste streams, and (2) system configurations involving separating and not separating drains are both feasible.
Combine CMP waste with oxide waste? / what type?/ which CuCMP stream
CMP polish and rinse waste separate drains?
Copper, clean, oxide, total stream treatment methods
Place of Treatment
Treatment Methods 1. Treat and Discharge 2. Treat and Recycle 3. Treat and URW
Type of oxide 1. Total 2. Clean 3. Dirty
Place of Treatment 1. POU 2. Facility 3. Local
Figure 2: Treatment decision for copper CMP [7] Cost catergories of options
Combine CMP waste with oxide waste? / what type?/ which CuCMP stream
CMP polish and rinse Type of oxide waste 1. Total separate 2. Clean drains? 3. Dirty
$/wafer pass
Place of Treatment -0.3
-0.2
-0.1
0
0.1
0.2
0.3
Capital ($)
1
Split stream combinations 1. Copper & Dirty oxide 2. Clean & Clean oxide 3. Both
Cu and dirty 2
Place of Treatment 1. POU 2. Facility 3. Local
Treatment Methods 1. Treat and Discharge 2. Treat and Recycle 3. Treat and URW
3 4
Clean and clean
Split Stream Combination (1,2,3)
Combined w/ Oxide?
Separate Drains? no no Combined w/ Oxide?
Treat and Discharge Dirty/Copper Stream Treat and Discharge Dirty/Copper Stream
Clean Stream Treatment (2,3)
Clean Stream Treatment (2,3)
yes Total oxide
Facility Treatment
n=6
6 7
2 3
Treatment Location (1,2, or 3) Facility Treatment
POU
n =3
8
Facilities
9
n =1
10
Local Facilities
Treat and Discharge 11
Total Stream Treatment (1,3)
Facility Treatment
2 Treatment
no Total Stream Treatment (1,2,3)
Process DI water saved(gpm)
Footprint Backpad (ft ^2)
n =2 12
Treat and URW
13
POU
14
Facilities
15
Local
n =3
Location (1,2, or 3)
1, 3 Facility Treatment
Consummables and Utilities ($/hour)
Citywater Both saved(gpm)
5
yes
yes
Copper, clean, oxide, total stream treatment methods
n =2
16 17
System Options = 17
Figure 3: Quick-CoO analysis of 17 option [7]
Treat and Discharge Treat and URW
Footprint Subfab (ft^2)
Footprint ($)
Total cost of options
$/wafer pass -0.2
Combine clean copper with clean oxide
Separate Drains
-0.15
-0.1
-0.05
0
Treat and Recycle
Facilities system
3
Separate Drains
Combine clean copper with clean oxide and slurry rich copper with slurry rich oxide
Treat and Recycle
Facilities system
5
Do not Separate Drains
Combine copper with total oxide
Treat and URW
Facilities system
12
Do not Separate Drains
Do not combine copper and oxide waste
Treat and Recycle
Facilities system
14
Figure 4: Final selected options [7] Detailed System Analysis using the EnV-S - The three layers of the EnV-S (process modeling, system sequencing and design and characterization) can be used to model cost, performance and environmental impacts of specific systems. In this work, analyses have been set up for facility scale, local and POU systems that treat and recycling process/rinse water back to the copper CMP loop (Figures 5, 6, 7) Treatment process modeling is supported through data and information from Applied Materials, water treatment system suppliers, and the NSF/SRC Center for Environmentally Benign Semiconductor Manufacturing at the University of Arizona. Previous work in the NSF/SRC center in the effluent treatment area has focused on detailed modeling of individual treatment processes. Case studies with the EnV-S apply these models towards process sequences to characterize and improve the entire treatment system in terms of cost, performance and environmental impact metrics. Clean Water Back to Tools "Clean" Waste From Tools Local System
R/O Polish
Permeate
UV
Residue
Holding Tank
Facilities System Sulfuric Acid Copper Rich Waste From Tool
Ultrafilter
Activated C
Ion Exchanger Regenerate
Holding Tank Residue Dumped
Used C Dumped
Electrowin Cu
Figure 5: Local System (5 CMP tools treated)
AWN
Clean Water to Tool Recirculatation
Unsegreagated Waste From Tool
Filter
AC
Filter
IX
IX
UV
Filter
Holding Tank
Holding Tank
Residue Dumped POU system Facilities system
Overflow Electrowin
Figure 6: POU System (1 CMP tool treated) Clean Water Sulfuric Acid
Unsegregated Waste From Tool
Activated C
Ultrafilter
Back to Tools
Double Pass R/O
Ion Exchangers
UV
DI Tank
Holding Tank
Landfill
Electrowin
Used C Holding Tank
Cu Landfill
Figure 7: Facilties System (20 CMP tools treated) Overall cost comparisons from this analysis appear in Figure 8. It can be seen that the facilities system has a cost advantage compared to the local and POU systems. The performance of the three systems was comparable (Figure 9). Sensitivity analysis of various cost and performance outputs were performed, with respect to a set of 40 system parameters. Significant parameters that drive system cost differentials appear in Figure 10. Final output cost differentials were robust with respect to process parameter variations, with the facilities system consistently appearing to be the cheapest (Figure 11). $0.12
Maintenance
$/wafer pass
$0.10
Disposal
$0.08
Consummables
$0.06
Energy
$0.04
Water Usage Footprint
$0.02
Capital
$0.00 Facilities Treatment
Local Treatment
POU Treatment
Figure 8: Overall Cost Comparison for the three Systems Water Recycling Efficiency (%) Copper Removal Efficiency (%) Final Copper Concentrations (mg/L)
Facilities Treatment 72.7 99.5 0.002
Local Treatment 69.5 99.9 0.0005
POU Treatment 72.7 99.5 0.003
Figure 9: Comparison of Performance for the three Systems
dependence
No real dependence
Cost differential per wafer pass
Strong dependence
$0.03 $0.02 $0.01 $0.00 -$0.01 -$0.02 Local Facilities
POU Facilities
Local-POU
System Pair
Figure 10: Significant Parameter Summary
Figure 11: Output Cost Differentials
Summary - From this application of the EnV-S, it is therefore apparent, that the facility scale solution proposed here would be more effective than either the local or POU solutions. For a company like Applied Materials to consider a POU treatment solution, new/alternative technologies would perhaps have to be pursued. Furthermore, specific areas to focus system cost reduction efforts can be obtained (eg. Reverse Osmosis membrane life, efficiency). Finally, sensitivity analysis can indicate system parameters that most drive overall outputs. In this case, waste stream parameters (volume flow rate, rinse to polish volumes) were very significant. Efforts must therefore be made to improve environmental performance in the primary CMP process itself. Future work will therefore be directed towards environmental modeling of the CMP process and inclusion of that model into the analysis for CMP. References 1. Chepesiuk, R., Where the Chips Fall: Environmental Health in the Semiconductor Industry, Environmental Health Perspectives, Vol. 107/7, 1999, pp. A451-A457. 2. International Technology Roadmap for Semiconductors 1999 Edition, Environmental Safety and Health, Semiconductor Industry Association (SIA). 3. Thurwachter, S., “Environmental Value Analysis: Evaluating Manufacturing Product and Process Design Trade-Offs,” Dissertation, University of California, Berkeley, Spring 2000. 4. United States Environmental Protection Agency: Global Warming Information, 1997/98 http://www.epa.gov 5. Toxics Release Inventory (TRI): Available from the Right to Know Network, http://www.rtk.net 6. Krishnan, N., Thurwachter, S., Francis, T., Sheng, P., “The Environmental Value Systems (EnV-S) Analysis: Application to CMP Effluent Treatment Options,” Improving Environmental Performance of Wafer Manufacturing Processes, Proceedings of the Electrochemical Society (ECS), Toronto, Canada, May 2000. 7. Krishnan, N., Bauer, D., Thurwachter, S., Francis, T., Sheng, P, “Modular Environmental Design and Decision Making through the Environmental Value Systems Analysis (EnV-S): Evaluating CMP Wastewater Treatment Options,” EECA ISESH Conference Proceedings, Dresden, Germany, June 2629, 2000. 8. Krishnan, N., Thurwachter, S., Francis, T. Sheng, P., “A Modular Environmental Design and DecisionSupport Tool (EDDT) for Semiconductor Manufacturing,” Improving Environmental Performance of Wafer Manufacturing Processes, Proceedings, SEMICON West, 2000, 2000.
Integrated Model for the Chemical-Mechanical Polishing Based on A Comprehensive Material Removal Model Jianfeng Luo
Sponsors: UC SMART and NSF
ABSTRACT: The architecture of an integrated model for chemical mechanical polishing is proposed in this report. For details on the report, the readers are referred to J. Luo, D. Dornfeld, Z. Mao et. al. ‘ Integrated model for the chemical-mechanical polishing based on a comprehensive material removal model,’ Six International VMIC conference, March, 2001. I. INTRODUCTION The chemical-mechanical polishing (CMP) is a quite complicated process. The understanding of the different roles played by the input values and their interactions, is critical for the optimizations of cost, material removal rate, non-uniformity, micro-scratches and the control/design of CMP. Although various models have been developed to explain the fundamental mechanism of CMP from the different viewpoints of slurry flow, slurry abrasive abrasion and chemical etching, there are still needs to explain the roles and complicated interactions between the input values in a comprehensive way. Recently, a material removal model to describe these interactions, which are quite different from those in conventional polishing or lapping processes due to the small pad hardness and different size scales of the pad asperity and the polishing abrasives, has been developed by Luo and Dornfeld ([1]-[2]). There are several motivations to extend the material removal model into a more comprehensive, or namely, integrated CMP model. First, the current material removal model can only explain material removal process in solid-solid contact mode. An extension is needed to cover material removal process in other contact modes. Secondly, in the current model the synergy effect of the chemical and mechanical elements in the contact interface of abrasive particle and wafer surface is simply represented by a ‘dynamical’ hardness value Hw of wafer surface [1]. Understanding the basic electrochemical mechanism is needed to clarify what input values and how they influence this fitting parameter. Thirdly, other important output values such as the nonuniformity (NU), surface scratch, dishing and etching, which is related to the material removal mechanism as well as other parameter such as the pressure distribution, can be addressed in a more comprehensive model with the material removal mechanism clarified [9]. Integrating all the above parts into one model is not a simple task to be completed in short time. A framework describing the basic architecture of the integrated model will be a useful first step. In this report, we propose a framework for an integrated model. It is composed of layers of sub-models. Showing what, how and where these sub-models come into the integrated model, it can help clarify the most important features of the integrated modeling and simplify the modeling process. II. ARCHITECTURE OF THE INTEGRATED MODEL Based on the discussion in the introduction, the basic framework of the integrated model can be as in Fig. 1. There are basically five layers of sub-models in the integrated model. In the first layer of sub-models are simply the output values including the material removal rate, non-uniformity, dishing, etching and scratching. The formulation of material removal rate on each single location behaves as the inputs for the non-uniformity, etching and dishing. Following them is the second layer of sub-models including the material removal model and the pressure & velocity distribution modeling and pattern density characterization. Nonuniformity, etching, dishing and scratching are functions of the distribution of pressure times velocity as well as the material removal mechanism. The numerical methods including finite element method (FEM), boundary element method (BEM) and computational fluid dynamics (CFD) and dynamical analysis may be used to model the pressure and velocity distribution in the second layer [7]. The lowest layer of the input values representing the bulky material and geometry features of the pad, slurry and wafer, behaves as the input values for the pressure distribution model and the analysis of the velocity distribution. The material removal models are the core part of the integrated model. It can be separated into two sub-models in the third layer, one based on the solid-solid contact mode, which has almost been done in the comprehensive material removal model by Luo and Dornfeld [1], and the other based on the non-directly contact mode. The material removal mechanisms are quite different for these two contact modes, which influences significantly the roles played by the sub-models in the lower layers, such as chemical, fluid and abrasion models. The abrasion in the solid-solid contact mode is principally two-body abrasion and the material removal by three-body abrasions is negligible. Chemicals in this mode influence the material removal principally through their enhancing effects on the two-body abrasion. Their direct contribution on material removal, or namely, etching, are not apparent in comparison with the mechanical removal, except when the mechanical removal is small in situations of small abrasive concentration [4]. In the non-directly contact mode, the abrasion is principally three-body abrasion. The material removed due to the three-body abrasion is a function of the shear stress introduced by the fluid flow. (The shear stress and the fluid film thickness change with the slurry viscosity and velocity [8].) Chemical enhances the three-body abrasions by reducing the surface energy of the work. However, the etching of work and the following mass transport cannot be neglected. In summary, the solid-solid contact region can be considered as a mechanical dominant region where the chemical removal is negligible in comparison with the two-body abrasion. The non-directly contact region can be considered as a chemical dominant region where the chemical removal is significant in comparison with the mechanical removal. Realization and clarification of these differences help to apply correct strategy in the modeling process. In the default fifth layer are the input values including the slurry abrasive geometry, pad topography and so on. Two important issues exist with this layer. First it is needed to identify in this layer the most important features of consumables including pad, slurry abrasive and slurry chemicals. What features are important is determined by the two different contact modes. Features playing important roles in solid-solid contact mode may be negligible in the non-directly contact mode. Second issue is on how to model these features. Some of them can be based on physical evidence. For example, SEM pictures of slurry abrasives can be used to model the geometry of the abrasive [1]. The dynamical light scatting can be used to measure the distribution of the abrasive size. Other may be based on previous theoretical models.
1
II. 1. ARCHITECTURE OF THE COMPREHENSIVE MATERIAL REMOVAL MODEL BASED ON SOLID-SOLID CONTACT MODE The portion in the dashed line box in Fig. 1 epresents the developed comprehensive material removal model for solidsolid contact mode. The detailed framework of the model is shown in Fig. 2 An important feature of the model is that the material removal can be separated into two parts, one the material removed by a single abrasive, and the other the number of active abrasives. Two sub-models evaluating these two parts comprise of the highest layer in Fig. 2 The product of the active abrasive number and the volume removed by a single active abrasive is equal to the total material removal at the wafer surface. II. 2. ARCHITECTURE OF THE MATERIAL REMOVAL MODEL BASED ON NON-DIRECT CONTACT MODE Up to now, there is no comprehensive material removal model developed for the non-direct contact mode. Based on the literature, the framework of the material removal in the non-direct contact mode can be proposed as in Fig. 3. III. CONCLUSION In this report, architecture of an integrated CMP model is proposed. Basically, the integrated model is composed of layers of submodels, explaining the material removal mechanism. Showing what, how and where the sub-models come into the integrated model, it helps to clarify the most important features of the integrated modeling and simplify the modeling process. IV. REFERENCES [1]. J. F. Luo and D. A. Dornfeld, “Material removal mechanism in chemical mechanical polishing, theory and modeling,” IEEE Transaction: Semiconductor Manufacturing, in press, 2001. [2]. J. F. Luo, “Material removal mechanism in chemical mechanical polishing,” Annual Research Report of LMA, University of California at Berkeley, Berkeley, CA, U.S.A., 1999. [3]. J. F. Luo, “Effect of particle size distribution in chemical mechanical polishing: modeling and verification,” ESRC Report (2000-09), University of California at Berkeley, Berkeley, CA, 2000. [4]. J. F. Luo, “Material removal saturation in chemical mechanical polishing with the abrasive weight concentration: effects of abrasive size and wafer-pad contact area,” this report, 2000. [5]. Y. Moon, “Mechanical aspects of the material removal mechanism in chemical mechanical polishing (CMP),” Ph.D. Thesis, Department of Mechanical Engineering, University of California at Berkeley, Berkeley, CA, U. S. A., 1999. [7]. D. Wang, J. Lee, K. Holland, T. Bibby, S. Beaudoin and T. Cale, “Von mises stress in chemical-mechanical polishing processes,” Journal of Electrochem. Soc., Vol. 144, pp. 1121-1127, 1997. [8].Y-S. Su, “Investigation of removal rate properties of a floating polishing process,” Journal of Electrochem. Soc.,Vol.147, pp. 2290-2296, 2000. [9]. J. F. Luo, “Improvement of non-uniformity in chemical mechanical polishing from the viewpoint of consumable effects based on a developed material removal model: a research proposal,” this report, 2000.
Figure 1 Basic framework of the integrated model Figure 2. Framework of the comprehensive material for solid solid contact mode
Figure 3. Framework of the comprehensive material removal model in solid-solid contact mode
2
Material Removal Saturation in Chemical Mechanical Polishing with Abrasive Weight Concentration: Effects of Abrasive Size and Wafer-Pad Contact Area Jianfeng Luo
Sponsors: NSF and UC SMART
ABSTRACT----The material removal rate as a function of abrasive weight concentration has been proposed. With the increase of the concentration, three regions of material removal exist, first the chemical dominant region, where the abrasive weight concentration is quite small, second the mechanical dominant region, where the material removal increases linearly with the weight concentration, and third the mechanical dominant saturation region, where the material removal no longer increases with the weight concentration because the contact area is fully occupied by the abrasives. In the model, a fitting parameter is used to represent the effect of chemical dominant region. The slope of material removal increase in the linear region is a function of abrasive size distribution. The saturation removal rate in the saturation region is a function of abrasive size distribution too based on our model. The verification of the MRR formulation in these two regions clarifies the roles of contact area and abrasive size distribution in the developed material removal model. I.
INTRODUCTION Usually the material removal rate in the solid-solid contact mode of CMP increases linearly with the abrasive weight concentration. It is observed that when the concentration of abrasives is larger than a special value, however, the material removal rate will stop increasing. We call this special value of concentration as saturation concentration Cs and this phenomenon as material removal saturation. A qualitative explanation of this phenomenon is that the total contact area between the wafer and pad surface has been occupied by the abrasives. The further increase of the concentration cannot increase the number of abrasives on the contact area any more, leading to the saturation of the material removal rate. This explanation is shown in Fig. 1 schematically. The material removal model developed by Luo and Dornfeld [1] supports this qualitative explanation. Moreover, it can explain this phenomenon in a quantitative way, pointing out that the saturation concentration is a function of abrasive size as well as the contact area. Based on the model, three regions exists with the increase of the weight concentration, one the chemical dominant region, second the mechanical dominant linear region and third the mechanical dominant saturation region. The slope in the linear region and the saturation material removal rate are functions of the abrasive size distribution. In this report, we discuss the material removal formulation from the viewpoint of the weight concentration and how the model predictions correlate with the experimental results. II. MATERIAL REMOVAL MODEL AND MATERIAL REMOVAL RATE AS A FUNCTION OF WEIGHT CONCENTRATION AND ABRASIVE SIZE DISTRIBUTION Before the saturation, the material removal formulation as a function of abrasive size distribution has been developed in [2] as follows: (1) 2 , MRR
X avg + 3 σ C 5 = 1 − Φ 3 − C 6 3 X avg σ , part
1
part
2
X avg + 3 σ σ p 3 − C 6 σ X avg + X avg + 3 σ Φ 3 − C 6 σ part
3= X
2 avg
− a
where C5 is a function of the weight concentration C, relative velocity V, and other consumable parameters, and C6 is a function of the pad hardness, pad topography, and down pressure [2]. According to our model, the C5 is linearly related to the weight concentration, so we can write C5 as C5= h(C+b) where h is a parameter related with the pad topography, abrasive density, slurry dilution ratio, abrasive geometry and chemical enhancing effect [1]. Note here we assume that the chemical enhancing effect is independent of the weight concentration. A little difference between C5 here and that in [1] is that we introduce a parameter b into C5. This parameter is introduced into the model in consideration that when the weight concentration is very small, say, close to zero, the material removal is not mechanical dominant any more (it is possible for the chemical removal such as etching to be compatible to the mechanical removal even in solid-solid contact mode when the number of cutting tools is small.). Only when the concentration is larger than some value, say, C1, which may be a function of the slurry chemicals, the material removal becomes mechanical dominant and increases linearly with the concentration. We don't know exactly what is the threshold concentration. The parameter b is simply introduced into the model as a function of the chemicals for considering this effect. We can take the b as independent of abrasive size considering the perfect correlation between the experimental results and model prediction in [2]. There the value of C5 is considered as independent of abrasive size. 3 We know that h(C+b)/Xavg is equal to the number n of abrasives on the contact area A. When the weight concentration is small, most part of contact between the wafer and pad is direct contact between the wafer and pad asperities, and the contact area A is dependent on the down pressure, pad material and pad topography but independent of the abrasive geometry and abrasive size [1], Fig. 2 (a). When the area is totally occupied by abrasives, however, abrasives behave as an interfacial layer between the wafer and pad asperities, Fig. 2 (b). In this case, we can consider the pad asperities having higher effective Young’s modulus. When the abrasives are larger, the effective Young’s modulus of the pad is larger, leading to smaller contact area A’. We consider that when saturation happens, the contact area has been totally occupied by active abrasives. (The abrasives smaller than these active abrasives are rolling down the asperities.) The area of each single abrasive on the contact area 2 is equal to 0.25πXavg-a , where Xavg-a is the size of the active abrasives. So when the contact area A’ is totally occupied by the abrasives, there should approximately be 2 3 ’ (2) [h(Cs+b)/(VXavg )]× X avg + 3 σ [0.25π Xavg-a ] =A =>
1 − Φ 3 − C
6
σ
(3)
Cs= 3 A 'V X avg , part
0 . 25 π h X − a 2 avg
part
3
1
X avg + 3 σ 1 − Φ 3 − C 6 σ part
− b
2
1
From Eq. 3, we can see that Cs is a function of abrasive size distribution. Part 2 in Eq. 3 is usually close to each other for different abrasive size distributions [2]. Part 3 is the size of active abrasives and it is usually close to kXavg where k can be considered close to a constant independent of abrasive size [2]. A’, the contact area at saturation, is dependent of the abrasive size, but does not change too much for different abrasive sizes, as to be shown later. Therefore, the Cs should be approximately linearly related to the abrasive size, indicating the material removal saturates earlier for smaller abrasive size. For abrasive concentration larger than Cs, the material removal will keep constant since the number of abrasive on the contact area cannot further increase. Based on the above discussions, the material removal can be written as a function of abrasive size distribution and concentration as follows: MRR= Eq. 1when C≤ Cs and MRR= MRRs (4) when C ≥ Cs Substituting Cs into the material removal rate function, we obtain the saturation material removal rate as
MRR
s
=
A 'V , 0 . 25 π
(5)
which is a function of contact area at saturation and relative velocity. If the slope of the linear region is S, which is a function of abrasive size distribution as shown above, then Cs can be written as (6) Cs= MRRs/S-b The contact area A’ at saturation is a function of abrasive size, or exactly the active abrasive size, down pressure and pad topography. For smaller abrasive size, the contact area A’ is larger. Based on contact mechanics, an approximate relationship between the abrasive size Xavg-a and the contact area A’ can be obtained as (7) A’∝ (1+m1Xavg-a)-2, where m1 is a constant related to the pad topography [2] [4]. Details about how to obtain this formulation will not be discussed here. Figure 3 shows schematically the plot of material removal as a function of abrasive weight concentration based on above discussions. III. EXPERIMENTAL VERIFICATION The experimental results from Bielmann et. al. are used to verify the material removal formulations [3] as shown in Fig. 4. IV. CONCLUSION The material removal rate as a function of abrasive weight concentration has been proposed. With the increase of the concentration, three regions of material removal exist. The verification of the MRR formulation clarifies the roles of contact area and abrasive size distribution in the developed material removal model. REFERENCES [1]. J. F. Luo and D. A. Dornfeld, “Material removal mechanism in chemical mechanical polishing, theory and modeling,” IEEE Transaction: Semiconductor Manufacturing, in review, 2000. [2]. J. F. Luo, “The Effects of abrasive size distribution in chemical-mechanical polishing: modeling and verification,” Annual Research Report of LMA, Department of Mechanical Engineering, University of California at Berkeley, Berkeley, CA, U. S. A., 2000. [3]. M. Bielmann, U. Mahajan and R. K. Singh, “Effect of particle size during tungsten chemical mechanical polishing,” Electrochemical and Solid-State Letters, Vol. 2, pp. 401- 403, 1999. [4]. K. L. Johnson, Contact Mechanics. Cambridge, Cambridge University Press, 1985.
2
30 Saturation MRR
Normalized MRR
25 15 10 0 -10
Slope K
5
-b -5
Saturation Region
Mechanical Dominant Linear Region
20
Chemical Region
1 0
Cs
C1
5
10
15
20
25
30
35
Weight Concentration C (%)
Figure 1. The contact area with abrasives before and after saturation 800
Material Removal Rate (nm/min)
700
Figure 2. Schematic of two contact modes with different abrasive weight concentrations
Experimental Xavg=2um Experimental Xavg=0.88um Experimental Xavg=0.6um Experimental Xavg=0.38um Experimental Xavg=0.29um Prediction Xavg=0.29um Prediction Xavg=0.38um Prediction Xavg=0.6um Prediction Xavg=0.88um Prediction Xavg=2um
600 500 400 300 200 100 0 -10
-5
0
5
10
15
20
25
Concentration (% )
Figure 3. Three regions with the increase of weight concentration
3
The Effects of Abrasive Size Distribution in Chemical-Mechanical Polishing: Modeling and Verification Jianfeng Luo
Sponsors: NSF and UC SMART
ABSTRACT: The abrasive size distribution plays an important role in the material removal in CMP. Basically, they influence the material removal from two aspects, one the number of active abrasives, and the other the size of the active abrasives. In this report, a model with material removal rate formulation explaining the effects of abrasive size distribution on the CMP material removal has been proposed and verified. I. INTRODUCTION The understanding of the material removal mechanism should be based on understanding the roles of the cutting tools, or the abrasives in CMP and their interactions with other important input values such as the pad, chemical and wafer materials. The effect of abrasive size distribution in the chemical-mechanical polishing has long been observed ([1]-[4]). Experimental results show that there is a reverse proportional relationship between the abrasive size and the material removal rate when the abrasive weight concentration is kept constant ([2]–[3]). Connections between the size distribution and the scratching of wafer surface have also been observed and reported [4]. Although qualitative explanations of the roles of abrasive size distribution in CMP have been proposed, however, to clarify the roles of abrasives in the CMP process, a quantitative explanation is needed. Recently, a comprehensive model to explain the fundamental mechanism and interactions ([5]-[6]) of consumable parameters in solid-solid contact of CMP, including the abrasive size and size distribution, has been developed by Luo and Dornfeld [5]. Material removal formulations as functions of the process parameters including the down pressure and velocity as well as consumable parameters including the abrasive size distribution have been proposed. It is found that this model correlates the experimental MRR as a function of abrasive size distribution perfectly. In this report, we discuss this material removal model from the viewpoint of abrasive size distribution and how the model prediction correlates with the experiment results. II. MATERIAL REMOVAL MODEL AND MATERIAL REMOVAL RATE AS A FUNCTION OF ABRASIVE SIZE DISTRIBUTION Based on the material removal model developed, the material removal rate can be written as a function of the abrasive size and size distribution before the material removal saturation [7], Eq. 1. In the earlier version, they were written as a function of down pressure and velocity [5-6]. , (1) 2 MRR
=
X avg + 3 σ C 5 1 − Φ 3 − C 6 3 σ X avg , L O O O O O M O O O O part
1
part
2
X avg + 3 σ σ p 3 − C 6 σ + X avg X avg + 3 σ Φ 3 − C 6 ON σ N L O O O O O O M O O O O O O part
3
where the C5 and C6 are two parameters representing the effects of other consumable parameters such as the weight concentration of the abrasives in the slurry, and process parameters including down pressure and velocity. The first part of the equation represents the effect of the abrasive size on the total number of abrasives on the contact area. Note that the Xavg3 is proportional to the average volume of a single abrasive and C5 includes a term of the weight concentration of abrasives in the slurry. Apparently, for the same weight concentration, when the average abrasive size is larger, the total number of abrasive on the contact area is small, leading to a small material removal rate. Note that there is an upper limit for the weight concentration. When the weight concentration is so large that the total contact areas are occupied by the abrasives, the material removal will saturate and the above formulation needs to be revised. This will be discussed in detail in [7]. The second part is a proportion function and represents the effect of abrasive size distribution on the number of the active abrasives. Apparently, the value of the proportion is smaller than 1, indicating that only part of the abrasives on contact area is active. The third part represents the effect of the average size of active abrasives on the material removal by a single abrasive (Note the material removal is proportional to the square of the average size of abrasives. This square term is introduced by the indentation model [5-6].) The average size of the active abrasives is larger than the average size Xavg of all abrasives, including both active and inactive abrasives. Note that if we did not consider the size distribution function of abrasive size, the MRR can be written as a function of the average size Xavg: (2) MRR= C5/Xavg=C5Xavg-1. Here, the part 2 in Eq. (1) is equal to one, and the part 3 is equal to Xavg2, indicating all abrasives on the contact area are involved in material removal process. III. EXPERIMENTAL VERIFICATION Buria et. al. [3] did Tungsten CMP experiment using five different distributions of abrasives. The size distribution is measured using dynamical light scattering technology and it is shown schematically in Figure 1 [3]. The average sizes of the five kinds of abrasives were provided by Buria et. al. [3] as 0.29µm, 0.38µm, 0.60µm, 0.88µm and 2µm, respectively. Assuming that the size distribution satisfies a normal distribution, we obtained the standard deviation of the size distribution using the measured data in Figure 1. Buria et. al. changed the weight concentrations of abrasives from 2% to 15% and obtained the material removal rate as a function of median abrasive size under four concentrations, as shown in Figure 2. Figure 3 shows the perfect correlations between the experimental results (average values of the experimental data under the three concentrations 2%, 5% and 10%) and the model predictions using Eq. 1. We found that with the increase of the abrasive size, the decreasing of material removal rate can be fitted with a power function. From Eq. 1 we can see that the power is actually determined by –3 in the term Xavg3 of part 1 and coefficient C6 in parts 2 and 3. The C6 value is determined by consumable values including pad hardness, pad topography and down pressure [5-6] but independent of the weight concentration. This implies that the weight concentration does not influence size dependency of material removal rate. Experimental results have supported this conclusion: three values of power– -0.6764, 0.6853 and –0.7756, which are close to each other, were obtained for the slurry concentrations 10%, 5% and 2%. When the weight concentration is 15%, the value –0.4985 of the power is much larger. This is because the C5 representing the effect of weight concentration is not a constant any more. Its value in concentration 15% is smaller for the smaller abrasive sizes (0.29, 0.38 and 0.6µm) than that for the larger abrasives (0.88 and 2µm)[7]. This variance contributes to the increase of the fitted power value. Moreover, it is noted that the distribution function of abrasives does play a significant role in the material removal from the fitted power values. Otherwise, they should be equal or close to –1 as indicated by Eq. 2, instead of –0.67~-0.78. From the data fitted from the equation, it is found that only a small portion of abrasives are involved in material removal, and the order of the portion (0.1% - 0.4%) is the same as that fitted from the earlier experimental data [5-6]. It is mentioned in [5-6] that this portion contributes significantly to the order of the material removal rate. Otherwise, the order of material removal prediction (compare Eq. 2 and Eq. 1) will be much larger than the real value, based on an estimation of the material removal rate [5-6].
1
V. CONCLUSION AND FUTURE WORK The abrasive size distribution plays an important role in the material removal in CMP. Basically, they influence the material removal from two aspects, one the number of active abrasives, and the other the size of the active abrasives. In this report, a model with material removal rate formulation explaining the effects of abrasive size distribution on the CMP material removal has been proposed and verified. In the future, the application of the model on the process optimization, for example, improving the non-uniformity by changing the standard deviation [8] or obtaining minimum surface scratching and large enough material removal rate by using optimal standard deviation, will be attempted. VI. REFERENCES [1] M. C. Pohl and D. A. Griffith, “The importance of particle size to the performance of abrasive particles in the CMP processes,” Journal of Electronic Material, Vol. 25, pp. 1612- 1616, 1996. [2] R. Xu, G. Smart and M. Zhang, “Particle characteristics and removal rate in CMP process”, Proceedings of Fourth International ChemicalMechanical Planarization for ULSI Multilevel Interconnection Conference, Santa Clara, CA, U.S. A., pp. 253- 255, Feb. 11-12, 1999. [3] M. Bielmann, U. Mahajan and R. K. Singh, “Effect of particle size during tungsten chemical mechanical polishing,” Electrochemical and Solid-State Letters, Vol. 2, pp. 401- 403, 1999. [4] J. Huang, H. C. Chen, J. Y. Wu and W. Lur, “Investigation of CMP micro-scratch in the fabrication of sub-quarter micron VLSI circuits,” Proceedings of Fourth International Chemical-Mechanical Planarization for ULSI Multilevel Interconnection Conference, Santa Clara, CA, U.S. A., pp. 77- 79, Feb. 11-12, 1999. [5] J. F. Luo and D. A. Dornfeld, “Material removal mechanism in chemical mechanical polishing, theory and modeling,” IEEE Transaction: Semiconductor Manufacturing, in press, 2001. [6] J. F. Luo, “Material removal mechanism in chemical mechanical polishing,” Annual Research Report of LMA, University of California at Berkeley, Berkeley, CA, U.S.A., 1999. [7] J. F. Luo and D. A. Dornfeld, “Material removal rate saturation in chemical mechanical polishing with the weight concentration: effects of abrasive size and wafer-pad contact area,” ESRC Reports, UC Bekeley, 2000. [8] J. F. Luo and D. A. Dornfeld, “Improvement of non-uniformity in chemical mechanical polishing from the viewpoint of consumable effects based on a developed material removal model: a research proposal,” ESRC Reports, UC Berkeley, 2000.
Figure 1. Abrasive size distribution (from [3])
Figure 2. Material removal rate as a function of abrasive size distribution Figure 3. Material removal prediction as a function of abrasive size distribution
2
Improvement of Non-Uniformity (NU) in Chemical-Mechanical Polishing (CMP) from the Viewpoint of Consumable Effects Based on a Developed Material Removal Model Jianfeng Luo
Sponsors: NSF and UC SMART
ABSTRACT--- Controlling/designing the consumable parameters is a possible way to optimize the non-uniformity in CMP, and may be a better choice than the conventional method of adding dummy structures with the decrease of the feature size in ICs and higher requirements on the circuit performance. In this report, the motivation for improving NU by adjusting consumable parameters are discussed. Details on this report can be found in J. Luo and D. A. Dornfeld, “Improvement of NU in CMP from the viewpoint of consumable effects based on a developed material removal model,” ESRC reports, UC Berkeley, 2000. I. MOTIVATION The non-uniformity (NU) in chemical-mechanical polishing (CMP) process is an issue associated with the non-uniform material removal rate over the wafer surface due to the uneven distribution of pressure times velocity over the surface. Basically, two kinds of non-uniformities exist in the CMP process, one is on the wafer scale, or namely the with-in wafer non-uniformity (WIWNU), the other on the die scale, or namely the with-in die nonuniformity (WIDNU) and pattern density effect. The WIWNU is related to the elastic deformation of the polishing pad and the velocity distribution over the wafer-pad interface. The WIDNU is related to the pressure difference over the different density area. Due to the non-uniform material removal rate over different locations at the wafer surface, the even-thickness surface before polishing will become non-planrized. The degree of non-planarity (NP) is defined as the height difference over the surface after polishing and determined by the non-uniformity of material removal rate and the polishing time/the average height to be polished. The non-uniformity (NU) of material removal rate can be expressed as follows once the maximum, minimum, and average material removal rate is known: NU= MRR max − MRR min × 100 % (1) MRR
avg
Apparently, the NU is a parameter without unit. As to be shown later, the MRRmax in Eq. 1 is the material removal rate on the location with the maximum pressure times velocity over the wafer surface, the MRRmin is the material removal rate on the location with the minimum pressure times velocity over the wafer surface, and the MRRavg is the average material removal rate. For WIDNU or pattern density effect, the MRRmax is the material removal rate on the low density area where the pressure is larger, MRRmin is the material removal rate on the high density area where the pressure is smaller and MRRavg is the mean material removal rate over the whole die area. The non-planarity after the polishing can be written as: (2) NP= Hmax-Hmin= (MRRmax-MRRmin)×T= (MRRmax-MRRmin)×H/MRRavg= NU×H, where Hmax is the height of the highest point on the wafer/die, Hmin is the height of the lowest point on the wafer/die, T the polishing time and H the average height to be removed. Since H is a constant specified before polishing, the non-uniformity (NU) of material removal rate is linearly related to the final non-planarization (NP). Experimental MRR data shows that the MRR satisfies the Preston's equation MRR= KpP0V + MRR0 (3) when the P0V is large enough, where MRR is the material removal rate, P0 the down pressure, V the relative velocity of wafer, Kp and MRR0 two constant values independent of P0V representing the effect of consumable parameters. A linear relationship exists between the P0V and the MRR according to Eq. 3. For different consumable combinations/recipes, the values of Kp and MRR0 may be different. Fig. 1 shows the three possible pressure times velocity dependences of material removal rate for three different consumable combinations, with different values of Kp and MRRo. Substituting Eq. 3 into Eq. 1, with the consideration that MRR is maximum when the P0V is maximum and MRR is minimum when the P0V is minimum, we have NU= K p ( P 0 V ) max + MRR 0 − K p ( P 0 V ) min − MRR 0 (4) K p [( P 0 V ) max − ( P o V ) min ] . K
p
( P 0 V ) avg
+ MRR
=
0
MRR
avg
To reduce NU, from Eq. 4 there are basically two methods. Method 1 is to reduce the non-uniformity of the distribution of the pressure times the velocity, or reduce the value of (P0V)max-(P0V)min. For WIWNU, this can be realized through the optimization of polishing head and platen design. Numerical methods including finite element method (FEM) and boundary element method (BEM), and dynamics and fluid flow analysis may be helpful for analyzing and reducing the non-uniformity of the pressure and velocity distribution on this scale. Based on a similar idea, adding dummy patterns in the low-density area may help to reduce the pressure difference and therefore the NU in the die level. This method is simple. However, eventually adding the dummy features in the die level will create circuit performance difficulties, due to extra parasitic capacitance and other factors. An alternative method to reduce the NU, according to Eq. 4, is to reduce
1
the sensitivity of material removal rate on the distribution of pressure times velocity. This can be realized through the adjustment of the Preston’s coefficient Kp and the average material removal rate MRRavg. The production rate cannot be decreased. Therefore, the average material removal rate should be constant or increased after the adjustment. To satisfy the above NU and MRR requirements simultaneously, there are basically 4 choices. They are listed in Table 1 with symbols ‘↑’ indicating ‘increases’, ‘↓’ indicating ‘decreases’ and ‘→’ indicating ‘keeps constant’. This can be realized based on the understanding of the material removal mechanism. A detailed model have been developed to explain this mechanism [1-3]. The details on the improvement of NU using this model can be found in [4].
II. REFERENCES [1] J. F. Luo and D. A. Dornfeld, ‘Material removal mechanism in chemical mechanical polishing: theory and modeling,’ IEEE Transaction: Semiconductor Manufacturing, in press, 2000. [2]. J. F. Luo, “Effect of particle size distribution in chemical mechanical polishing: modeling and verification,” Annual Research Report of LMA, Department of Mechanical Engineering, University of California at Berkeley, Berkeley, CA, U.S.A., 2000. [3]. J. F. Luo, “Material removal saturation in chemical mechanical polishing with the abrasive weight concentration: effects of abrasive size and wafer-pad contact area,” Annual Research Report of LMA, Department of Mechanical Engineering, University of California at Berkeley, Berkeley, CA, U.S.A., 2000. [4]. J. Luo and D. A. Dornfeld, “Improvement of NU in CMP from the viewpoint of consumable effects based on a developed material removal model,” ESRC reports, UC Berkeley, 2000.
1200 Consumable Combination 1
Material Removal Rate (nm/min)
1000
Consumable Combination 2 Consumable Combination 3
800
600
400 K p =80
200
MRR 0
K p =100
0 0
1
2
3
4
5
6
7
8
9
10
11
Normalized Pressure times Velocity (PV )
Figure 1. The material removal rate equation for different consumable combinations Choices i ii iii iv
MRRavgg ↑ ↑ ↑ →
Kp ↓ → ↑ ↓
Table 1. Four choices to improve the NU through Preston’s Equation adjustment
2
On the Abrasive Particle Settling Mechanisms in Chemical Mechanical Polishing (CMP) Author Zhoujie Mao Supported by LMA
Abstract – In a CMP process, the slurry is used both as a chemical reaction agent and as a carrier of abrasive particles. This research focuses on the behavior of the abrasive particles in the slurry as a first step to better understand the role of slurry in the material removal process and the efficiency of slurry usage in CMP. Introduction It is believed that 2-body abrasive process is the dominant material removal mechanism in a CMP process. Various material removal models based on such abrasive mechanism have been developed [2]. It is assumed that the material removal rate (MRR) is proportional to the number of the abrasive particles engaged in the material removal process. However little research has been conducted on how the abrasive particles are engaged into the material removal process and why 2-body abrasive process dominates. In this research, the abrasive particle settling mechanism in the slurry will be investigated, and the effect of the slurry delivery position on the material removal rate will be discussed. Abrasive Particle Settling Mechanisms in CMP Usually slurry is delivered onto polishing pad at some specific flow rate, the slurry then flow away from the delivery point. The abrasive particles in the slurry will move either individually or as a group depending on the abrasive particle concentration. In a typical CMP process, the abrasive particle concentration in the slurry is usually between 3 –5 %, which is low enough for the interaction among particles to be ignored. The following analysis is thus based on individual abrasive particle movement. Assume the initial concentration of the abrasive particles in the slurry is n0 , the velocity of the abrasive particle is v0 , the terminal velocity of the abrasive particle can be found via simple Newtonian motion analysis of abrasive particle as follows 2r 2 ρg Vs = 9η Where r is the radius of the particle, ρ is the density of the particle, η is the viscosity of the slurry. Let w( x0 , x, t )dx denote the probability of finding the particle between x and x+dx after time t, then it can be found that w( x0 , x, t ) satisfies the following equation in general ∂w = −div (V w) + D∆w ∂x with ∫ wdx = 1 , w( x0 , x0 ,0) = 1 , and w( x0 , x,0) = 0 if x = x0 . It can be calculated then
that the probability a particle at a distance h from a wall with a constant velocity Vs reaches the wall at the rate of
(h − Vs t ) 2 exp − 4 Dt 4πDt 3 and then the settling rate of abrasive particles from the slurry on the polishing pad can be found as h
I (t ) = ∫
∞
0
D −Vs2t (h − Vs t ) 2 Vs Vs2 t 4D exp − e ) + (1 + erf dh = n0 4 Dt 2 4 D πt 4πDt 3 h
Where D is the diffusion coefficient of abrasive particle in the slurry, and t is the time taken by the abrasive particle to settle. Given slurry supply rate, the slurry delivery position, the abrasive particle settling rate, the number of abrasive particles deposited on the polishing pad, can then be calculated. Result and Future Work Figure 1 shows how the settling rate changes when the slurry is delivered onto the center of the polishing pad with different slurry supply rate. The settling rate decreases when the slurry is moving away from the point of delivery, and increases with the increase of the slurry supply rate. Notice how rapidly the settling rate decreases along the radius. Figure 2 shows the average abrasive particle settles underneath the wafer when the slurry is delivered at different location. It is interesting to notice first that the settling rate does not change monotonically with the point of delivery. Secondly the settling rate difference across the wafer varies greatly with the point of delivery, which implies that the slurry delivery position not only affects the material removal at each point on the wafer, but also the overall non-uniformity across the wafer. This is experimentally verified by Chou [3]. Future research will focus on integrating the above analysis with material removal model to better understand the material removal mechanism and non-uniformity issue in CMP. References 1. Fuchs, N. A. “The Mechanics of Aerosols,” Dover Publication, Inc. New York, 1989 2. Nanz, G and Camilletri, L.E., “Modeling of Chemical-Mechanical Polishing: A Review,” IEEE Transactions on Semiconductor Manufacturing. Vol. 8, No. 4, Nov. 1995 3. Chou, F., Fu, M., and Wang, M., “A General Optimization for Slurry Injection during Chemical Mechanical Polishing,” Journal of the Electrochemical Society, Vol. 147, No. 10, 2000
8
x 10
15
Q=50ml/min Q=100ml/min Q=150ml/min
Rate of deposition (n/m2/sec)
7 6 5 4 3 2 1 0 0
20
40
60
80 100 120 Radius (mm)
140
160
180
200
Figure 1 Abrasive particle settling rate along the radius of the polishing pad with different slurry supply rate
8
x 10
16
Average Settling Rate Underneath W afer e=0mm e=100mm e=200mm
Average Settling Rate (n/m2/s)
7
6
5
4
3
2
1 100
120
140
160
180 200 220 Raidial Position (mm)
240
260
280
300
Figure 2 Average abrasive particle settling rate across the wafer with different slurry delivery position
Surface Texture Monitoring in Ultraprecision Cutting of OFHC Copper Giuseppe Mastrolilli Sponsored By: affiliates of LMA Abstract—The texture of diamond-machined optical-quality surfaces on polycrystalline metals is strongly influenced by the anisotropic elastic and plastic properties of the work material. Microscratching tests were carried out on coarse-grained OFHC copper specimens to examine the correlation between surface topography and process forces. In agreement with previous work performed by other researchers [1] the grain boundary has been found to be a potential source of surface defects. When the force reaches a high level it’s possible to detect rough ares starting from the grain boundary. Experiment The experiment was performed on a Pneumo precision lathe, with diamond tools and coarse-grain OFHC copper workpieces. The depth of cut is constant and the scratch starts at the edge of the specimen and ends at an intentional stop.The cutting force was measured with a load cell fixed on the tool-holder. Several experiments have been done with different tools and different depth of cut,with the intent to control the average cutting force by the material removal. Depth of cut and tool nose radius have been kept to values low enough to cut primaraly on a single grain at the time. A sample is shown in Fig. 1. Rough areas are easily correlated to high forces and distinct grains; wide smooth areas are clearly associated with lower forces.
process continous cut c.speed : 0.3 mm/s depth of cut : 6.5 um
6 Cutting
Force[N] 4 2
tool monocryst.diamond nose radius : 1.414 mm
A B
0
workpiece OFHC copper av.grain size : 300 um 0
2
4
6
8
Time[s]
Figure 1. Different grain orientations lead to different cutting forces and different surface textures. The tool structure accumulates energy when traversing a grain with unfavorable crystallographic orientation.The accumulation of energy is characterized by unstable cutting, and large areas with high roughness are found. When decreasing the depth of cut the surface texture tends to be more homogeneous. This is in accord with the theory that damage occurs only when the tool structure reaches certain energy levels.
Fig. 2 shows a magnified views of a rough and smooth area. The picture to the left is from point “A” in Fig. 1. The force varies from 2.9 to 3.8 N at this point, up to 6.6 N within the same grain. The area around point “B” is magnified in the right image in Fig. 2. The force here is only about 1.6 N. A microscopic waviness is still detected.
Figure 2. Conclusions Scratching tests were performed on OFHC copper. The machined surfaces exhibit different characteristic microtopographies. Force measurements demonstrate that high force levels due to passing over a grain with different orientation can be the cause of increasing surface roughness. The rough areas start at the grain boundary, and their surface areas are variable. Experiments demonstrate the feasibility of using force as a means for process monitoring, and explanation of the phenomena can contribute to the modelling of ultraprecision cutting processes. References [1] E. Brinksmeier and J. Schmutz,”Generation and texture of surfaces in ultraprecision cutting of copper” Machining Science and Technology, Vol.1, No.2, pp.185-193, 1997.
