Planarized surfaces have become key to the success of advanced semiconductor devises/circuits/chips. The planarization, achieved by the use of chemical mechanical means, has enabled the interconnection of ever increasing number of devices and also the use of lower resistivity copper as the interconnect material for such devices. Chemical mechanical planarization (CMIP) has now found application at several different stages of semiconductor chip fabrication and many other microelectronic applications. However, there remain a large number of nuances and effects e.g. pattern, chemical, and pad dependencies and scratching, that need to be carefully studied, evaluated and eliminated if we want to continue to progress in sub 0.1 µm (minimum feature size) regime, where the amounts of material to be removed will be small, surfaces will dominate the performance, and margin of error extremely small and unforgiving. This presentation will discuss the future needs, the CMP variables, the relationship of these variables to CMP behavior and planarity, scratch-free CMP, and size-impact on CMP outcome. A new set of goals will be presented and discussed.

Planarization of dielectric films, silicon dioxide, silicon nitride, etc. by Chemical- Mechanical Polishing (CMP) is regulated and moderated by the interaction of the abrasive particles and chemicals in solution with the film surface through complex chemical and physical processes. Changes in the slurry properties have a profound effect on the polishing chemistry and relative removal rates of dielectric films. Common slurry properties include pH, temperature, abrasive particle composition, its size and shape, degree of agglomeration, and weight percent, and chemical composition. While the slurry vendor has control over most slurry properties, the pH and temperature can be controlled during the polishing process by the user and can have a strong influence. Data are presented highlighting the influence of pH and temperature on the CMP of both blanket and patterned silicon dioxide and silicon nitride films.

Oxide planarization using a specially fabricated “grindstone” was investigated. Using the grindstone, better planarity compared with the conventional CMP technique was demonstrated. Interestingly, it was also found that the oxide removal rate became very slow after the oxide surface became planar. This self-stop was thought to be influenced by the isolated abrasives from the grindstone. The dependence on the tool structure and the conditioning was investigated to prove the model. Also the defect issue is presented in this paper.

In this study, we have characterized the effects of abrasive properties, primarily particle size, on the Chemical Mechanical Polishing (CMP) of oxide films. Sol-gel silica particles with very narrow size distributions were used for preparing the polishing slurries. The results indicate that as particle size increases, there is a transition in the mechanism of material removal from a surface area based mechanism to an indentation-based mechanism. In addition, the surface morphology of the polished samples was characterized, with the results showing that particles larger than 0.5 μm are detrimental to the quality of the SiO2 surface.

In this work, we propose a new equation that predicts the planarity as a function of active pattern density, initial step height, selectivity between gapfilled oxide and silicon nitride and over CMP amounts. In order to achieve highly planarized STI surface, uniform active density, reduced initial step height, minimization of over CMP amounts and high selective slurry were required. Our new equation was applied to the 0.18um graded CPU devices’ STI CMP to enhance planarity and these parameters were evaluated quantitatively. It is concluded that the model suggested is useful in predicting CMP planarity

One of the major problems with oxide-CMP is the oxide thickness variation after CMP within one die, the socalled Within Die Non-Uniformity (WIDNU). The variations in pattern density of the design layout causes different local removal rates across the die resulting in the WIDNU. In this paper we shown that this is influenced by the pad stack. Depending on the thickness of the top pad the WIDNU can be reduced from about 470 nm until almost zero. This will be related to the bending of the top pad. The modelling will focuss on the two extreme cases of perfect pad bending and no pad bending.

Chemical-Mechanical Planarization with structured abrasive uses a subpad to manage the pressure variations due to loading over a range of length scales. The effect of subpad construction on pressure responses related to those scales is illustrated.

A minimum length scale for the effect of the subpad is established via contact mechanics. Differences between one- and two-layer subpads are shown. Uniform compression, point loading, and edge exclusion are considered briefly. A model of the subpad as a plate on an elastic foundation is applied to the problem of die doming. The roles of process pressure, die size, and subpad construction are illustrated. Planarization at the intra-die, die, and wafer scales are related to the subpad construction.

This paper reports a technological advancement in developing and implementing a novel retaining ring of advanced edge performance (AEPTM ring) for an advanced polishing head design. The AEP ring has been successfully used for significantly improved CMP performance in different CMP applications: oxide (PMD and ILD), shallow trench isolation (STI), polysilicon, metal (W and Cu), silicon-on-insulator (SOI), and silicon CMP. Robust processes have been developed using AEP ring along with many hardware upgrades for each application with extended runs to meet requirements of advanced IC device fabrication.

Chemical Mechanical Polishing (CMP) as a semiconductor polishing technology has grown dramatically during the past decade. It has been a key enabling technology, facilitating the development of high density multilevel interconnects. Its widespread application has exceeded the growth of the scientific understanding.

Models for silicon dioxide polishing mechanisms have built upon the work originally done for glass polishing. Recent work has augmented our insight, but our understanding is far from complete. A quantitative picture of the basic interaction mechanisms for silicon dioxide polishing does not yet exist. The models for metal polishing are now an active area of investigation. Recent work has demonstrated that more work is need to adequate explain the CMP of metals. The metal CMP mechanisms appear to be substantially more complex than originally assumed.

The challenges for CMP have been to improve the performance of the technology, and increasingly this needs a foundation of scientific understanding to achieve the needed gains. We can expect that, as the scientific foundation grows, it will contribute significantly to manufacturing improvements as has been the case in other areas of semiconductor technology.

A production worthy, Tungsten Chemical Mechanical Polish (CMP) process using a commercially available K103 slurry was developed, characterized, and tested for sub-0.35μm multilevel interconnect fabrication. The effects of pre-tungsten CMP process on tungsten polish are reported in detail. A head-to-head comparison of the optimized KIO3 process with the standard Fe(NO3)3 process is described. Critical CMP tool parameters (process and hardware) were flexed using statistically valid experimental designs. The advantages and disadvantages of a post tungsten polish, oxide buff, are discussed. Across-wafer non-uniformity, specifically the enhanced polish rate of tungsten at the wafer edge, was significantly reduced with the optimized process parameters and hardware setup. Also, an automated endpoint system was utilized and a set of robust endpoint algorithms were developed to minimize the amount of oxide loss during tungsten CMP processing. Finally, the positive effects of the optimized KIO3 tungsten CMP process on interconnect integration and die yield are reported.

Electrochemical interaction between the oxidizer and the metal is believed to play a key role in material removal in tungsten CMP. In this study, we use X-ray Photoelectron Spectroscopy (XPS) in conjunction with electrochemical measurements in both in-situ polishing conditions as well as in static solutions, to identify the passivation and dissolution modes of tungsten. Dissolution of tungsten oxides was found to be the primary non-mechanical tungsten removal mechanism in CMP.

Tungsten CMP involves a synergistic interaction of electrochemical and tribological (wear) phenomena. So far, numerous studies have been conducted using static electrochemical measurements as well as some polishing experiments. In this study, we present some results obtained from carrying out potentiodynamic measurements and tribological experiments in a reciprocating sphere-on-plate tribometer, which allowed a precise control of mechanical and electrochemical conditions. In addition, anodic current-time transient measurements were also used to characterize the kinetics of tungsten passivation reaction. These results indicate that the presence of an passive film is essential for wear of tungsten to take place.

Abrasive particle size plays a critical role in controlling the polishing rate and the surface roughness during chemical mechanical polishing (CMP) of interconnect materials during semiconductor processing. Earlier reports on the effect of particle size on polishing of silica show contradictory conclusions. We have conducted controlled measurements to determine the effect of alumina particle size during polishing of tungsten. Alumina particles of similar phase and shape with size varying from 0.1 μm to 10 μm diameter have been used in these experiments. The polishing experiments showed that the local roughness of the polished tungsten surfaces was insensitive to alumina particle size. The tungsten removal rate was found to increase with decreasing particle size and increased solids loading. These results suggest that the removal rate mechanism is not a scratching type process, but may be related to the contact surface area between particles and polished surface controlling the reaction rate. The concept developed in our work showing that the removal rate is controlled by the contact surface area between particles and polished surface is in agreement with the different explanations for tungsten removal.

Conventional methods of controlling tungsten Chemical Mechanical Polishing (WCMP) processes have included blanket wafer polish rate and platen temperature monitoring. In order to enhance process control of WCMP, a second generation motor-current sense endpoint system was chosen as the evaluation tool. The endpoint detection system (Luxtron 9300) was installed on a WCMP polisher (IPEC 472) at the Motorola MOS6 facility. Endpoint algorithms were evaluated under various polishing conditions during time periods between pad changes. The data collected over 3 months led to a reliable endpoint recipe for both the contact and via WCMP process. The endpoint data compared with the conventional timed recipe, indicated that a time reduction at WCMP could be achieved, with a 10% increase in throughput. In addition, contact and via W-plug endpoint data showed a strong signal versus pattern density for given device types. In summary, the motor current endpoint system tested verified reliable endpoint, process control, and characterization capabilities.

In this work, we investigated the dependence of the removal rate upon the oxidizer (peroxide) addition into commercially available slurries for a variety of films such as aluminum, titanium, titanium nitride and oxide. We found that the barrier layer materials were extremely sensitive to the peroxide addition while the removal rate varied only slightly for aluminum and oxide. The selectivity to titanium and titanium nitride drops from as high as 1000 to almost close to 1 as the mixture ratio (peroxide : slurry) increases. We proposed that the barrier layer be used to protect the oxide from being over-exposed and suppress the erosion eventually. This can be easily realized by dividing the process into two steps with each step being run at a specific peroxide mixture ratio. The experimental result unambiguously proved, for the first time, the effectiveness of this approach.

Nanoindentation techniques were used to determine the hardness of Cu, Ta & W metal discs and thin films on silicon substrates as a function of load or indentation depth. Cu films exposed to oxidizing solutions containing H202 exhibited a higher hardness at the surface while no such change was observed for W exposed to ferric nitrate. The implication of these measurements and their relationship to chemical-mechanical polishing rates are discussed.

Hydroxyl radical generation has been observed during Cu CMP using hydrogen peroxide-glycine based slurries. While the Cu dissolution/polish rates increased with increasing glycine concentration, the copper dissolution rate decreased with increasing peroxide concentration indicating the occurrence of both dissolution and passive film formation during CMP. This is further confirmed by both in situ and ex situ electrochemical experiments.

The polishing of copper and examination of the polished surfaces were carried out with surfactant based alumina slurries to yield interesting results. Contrary to our expectation and previously reported research, some of the surfactant based alumina slurries resulted in higher copper polish rates when compared to the control. Of the nonionic surfactants, BrijR 35 was overall the most effective in both acidic and basic media. Ionics were effective at the pH for the appropriate charge type. For the range of surfactants studied, polish rates correlated with the HLB of the nonionic surfactants. The Hydrophile-Lipophile Balance (HLB) is related to the solubility of the surfactant, with higher number corresponding to increased water dispersibility. The surfactant BrijR 35, with the nonionic composition polyoxyethylene(23) lauryl ether, resulted in a dramatic improvement in the average surface uniformity when compared with the control at pH 2, and Sodium Dodecyl Sulfate produced even greater uniformity. Additionally, the effect of BrijR 35 surfactant was maintained with change in abrasive size, pad and polishing tool. In order to insure that surfactants are compatible with the chemical reagents contained in the commercial slurries, two chemistries (ferric nitrate and hydrogen peroxide) were employed to test the efficiency of the selected surfactants in their presence. The results showed that the effect of surfactant on stability and removal rate is not influenced by the presence of the chemicals. Preliminary results indicate that surfactants can have a beneficial effect on both defects and post polish clean.

CMP removal rate (RR) of electrodeposited Cu film was found to increase by 35% over time after plating. The RR increase was attributed to Cu film hardness reduction of 43% and grain growth from the initial 0.1urn at as-deposit to lum at the final stage at room temperature. The removal rate increase will translate to variations in manufacturing environment and are therefore unacceptable. It was found that annealing at ∼100C for 5 minutes in inert gas will stabilize Cu films and provide consistent CMP removal rate.

The effect of pressure and velocity on the polish rates of copper was determined in DI water and in the presence of ferric nitrate, H202/glycine, and NH4OH with alumina particles as the abrasives. The polish rate shows a stronger dependence on velocity than that predicted by the Preston equation in the case of ferric nitrate, a highly reactive chemical. The velocity dependence is weaker for the other two less reactive chemicals, and is the same as that predicted by Preston equation for DI water. Our earlier empirical model, R = KPV + BV + Rc, where K, B, and Rc are constants, describes all the polish rate data satisfactorily.