Particle-Substrate Interactions
Although particle-particle interaction can be simulated by AFM to ensure adequate slurry stability—which translates into lower surface defectivity and fewer wafers lost to large-scale surface deformations (scratches)—particle-surface interactions drive the material removal and planarization processes. Hence the overall effectiveness of the CMP process largely depends on these interactions. Not surprisingly, the majority of AFM studies involving CMP focus on these interactions almost in entirety.
At the nanoscale, one can envision the CMP process as a single particle phenomenon. This is depicted in Fig. 5 where an abrasive particle is entrapped by a polymeric pad asperity as it transverses across the wafer surface. The particle-surface interactions incurred by the applied normal and lateral loads (FN and FL, respectively) result in the removal of the thin chemically modified layer. As the bare wafer surface is exposed to the solution environment, chemical reagents form a new chemically modified surface layer. This process repeats itself as subsequent particles are brought into contact with the wafer surface.
From this perspective, the CMP process can be effectively simulated by rastering an AFM probe against a wafer surface. Recently, several authors have performed this experiment to simulate the CMP of metal materials by scratching metal substrates with an unmodified AFM tip.[28-30] In metal CMP, a brittle oxide layer is formed on the more ductile metal surface, which is thought to be removed, almost in entirety, via the mechanical normal and shear stresses imposed by the abrasive particles.[3] Contrary to silica polishing, in these systems, an AFM tip may provide an adequate surrogate for an abrasive particle. However, it should be stressed that this assumption is speculation because no definitive studies have been conducted yet to verify the brittle fracture/film delamina-tion mechanisms for metal CMP.[3] Nonetheless, bare AFM tips have been reported to adequately simulate an abrasive particle in metal CMP.
Fig. 5 Schematic illustrating a single particle material removal event during CMP.
Deveecchio et al.[28] used a silicon AFM tip and a pure alumina substrate to simulate the polishing of a metallized wafer. By rastering the tip over a known area on the metal surface for a given time and applied load, they were able to derive nominal material removal rates from AFM experiments. The material volume losses from ”scratched” areas were determined by subsequent imaging at low applied loads. This value was normalized by the rastering time to obtain a nominal removal rate. Using this approach, the nominal material removal rates were studied under a range of conditions including varying tip/sample forces, solution pH, and electrode potentials (used to imitate an oxidizing agent). Baseline experiments performed in air had a negligible removal rate, illustrating that the observed phenomenon results from both the mechanical abrasion and the solution chemistry. As with ensemble CMP experiments, the removal rate was found to be strongly dependent on the solution pH, the relative amount of oxidation, and the applied load.
A standard silicon nitride AFM tip was used by Lim et al.[29] to measure frictional and topographic changes at the copper surface in the presence of nitric acid and neutral solution with benzotriazole (BTA), a corrosion inhibitor. No significant changes were found to occur in neutral solutions with BTA. However, under acidic conditions where isotropic dissolution of the copper surface is known to occur, BTA acted as a passivation agent. Moreover, the BTA-passivated chemically modified layers could be removed via abrasion with the AFM tip, resulting in preferential material removal at the contacting asperities. The presence of higher interfacial friction with BTA-passivated copper surfaces as well as tip-mediated, localized dissolution of copper interface was observed. Similarly, Berdyyeva et al.[30] also used a silicon nitrate tip to simulate the CMP of copper surfaces oxidized in aqueous solutions of peroxide and glycine at varying pH values. The material removal rate and observed indentation depth as a function of pH was consistent with the previous CMP results. The friction force between the tip and the surface was measured, and, in general, it was shown that the lower friction coefficients correspond to the slower removal rates.
Maw et al.[31] recently proposed an alternative approachg for studying CMP processes. Single surface asperity wear was simulated by measuring the dulling or material removal from a silicon nitride AFM tip rastered against a variety of substrates in aqueous solution. The chemical nature of the substrate was shown to play a critical role in the wear of dielectric oxide materials—significant material removal was observed only when the surface was populated with the appropriate metal-hydroxide bonds. These results suggest that pressure-induced bridging between oxide bonds on the substrate may assist in CMP processes.
Fig. 6 IC-1000 polishing pad surface as imaged by AFM (top image) and SEM (bottom image) showing the macroporosity and microporosity.
The investigation described above further illustrates the importance of the selection of an appropriate probe material when attempting to simulate a CMP process via AFM. A bare AFM tip should be used only to simulate another material on which the probe is not expected to significantly react with the solution environment, wafer surface, or otherwise facilitate the adsorption of solution species that would interfere with the measurement process. These criteria may not be applicable in silica CMP especially if the abrasive particle material is also silica. Although a significant fraction of the silicon nitride surface may contain silica because of hydrolysis in water, the surfaces are nevertheless different in charge, hydroxyl site density, and modulus.[31] Under such situations, an attached larger-sized particle (colloidal probe) with nearly identical surface chemistry with respect to the primary slurry particles may provide the best simulation. For reasons of sensitivity, reproducibility, and scalability with radius, smooth particles are generally preferred unless an adequate surface morphology match can be made or specific morphological features are investigated.
The range of normal loading forces required for simulation must also be known. This requires an approximation of the pressure per particle experienced in the CMP process. The applied tool pressures in CMP operations can range from less than 1 to over 7 psi. Because of the corregation (Fig. 6) of the polishing pad, only a fraction of the pad surface transfers this loading force. Hence the localized forces transferred by pad asperities can be appreciably higher, and the pressure per particle will depend on the pad characteristics such as its effective surface morphology and elastic modulus.
Recently, FTIR-ATR was used to measure the effective pad contact area as a function of applied pressure.[9] The IR spectra of the IC 1000 pad sample were collected with applied download forces ranging from 23 to 79 kPa. As the applied load was increased, the signal intensity also increased, indicating that more pad contact was achieved with the ATR crystal. To obtain the percent pad coverage, the same analyses were also conducted on a piece of polyurethane without corregation, identical in composition to the pad. The results represented in Fig. 7a and b evidence that only a fraction of the pad surface was in contact with the wafer surface or with the ATR crystal. Indeed, the percent pad contact values varied between 0.25% and 0.5% in the selected pressure range. These results agree with the literature findings, which have reported the pad contacts between 0.05% (hard pad with elastic modulus value E =100 MPa) and 0.54% (soft pad with E =10 MPa) at 7 psi.[32] Through scanning electron microscope (SEM) and subsequent image analysis of the particle-laden pad after polishing, the number of 200-nm particles in contact with the pad at 12 wt.% solids loading was estimated to be 42 x 106 per in.2, and the interpreted applied load was then found to be 750 nN per particle. Through these approximations, it became evident that the steric force barriers imposed by micellar aggregates on the silica surface in Fig. 3 were exceeded during the polishing process. Therefore micelles were not present on the wafer surface during CMP and could not be responsible for the lack of material removal. Hence the presence of surfactant monomer was found to impede the polishing process.
Fig. 7 Percent IC-1000 pad contact as a function of applied load.
To investigate this anomalous polishing behavior, rectangular tipless cantilevers (MikroMasch) were selected with a normal spring constant in the range of Kn=2.8±0.3 N/m to ensure that normal forces comparable to those experienced in CMP could be accessed. Sol-gel silica particles (7.5 mm) were attached to the end of the cantilevers using a small amount of high-temperature melting epoxy (Shell Epikote 1009) to ideally simulate the slurry particles. As with the previous normal force measurements, the frequency method[33] was used to determine the exact values of the normal spring constants, and forces normal to the flat surface were measured according to the method introduced by Ducker et al.[13,14] As will become evident, the utility of colloidal probes and AFM, in general, for CMP simulations offers multiple routes to probe the fundamental interactions that define the CMP process.
Lateral Forces and Surfactant Adsorption
Atomic force microscopy lateral force measurements have been used to investigate the loss of material attrition in the surfactant-based slurry system. The lateral interaction force between a silica colloidal particle and an atomically smooth flat silica surface was measured to simulate particle-wafer system interactions. Figure 8 illustrates representative lateral force vs. loading force curves for the case of silica surfaces interacting across pure water and a 32-mM solution of C12TAB. Data from several experiments are presented to illustrate the reproducibility of the measurements. Each experimental series was taken at a fixed position on the sample. Subsequent to each increase in the normal force, 10 friction force cycles were allowed to transpire to allow the lateral force magnitude to stabilize prior to taking a reading. Consecutive runs taken over different areas show good reproducibility with a maximum deviation of about ±5%. This suggests that there was no significant wearing of the probe during measurement. It is important to mention that the experimental results from different probes have similar features with regards to lateral and adhesive force interactions under the given solution conditions; however, slight differences in magnitude might appear because of variations in the local roughness of the probe surface.[34] For consistency, the data presented have been obtained using the same colloidal probe, unless otherwise noted. The lateral force measurements presented were also taken at a constant lateral scan velocity of 2 mm/sec. However, it should be noted that contrary to studies on chemisorbed self-assembled monolayers1-35-37-1 and adsorbed polyelec-trolytes,[38] no significant dependence of the frictional force was found when the scan rate was varied between 1 and 20 mm/sec. The measurements in surfactant solutions were systematically performed 30 min after the solution injection to prevent variations in the extent of consolidation or the packing and morphological state of the adsorbed layers.[39,40]
Fig. 8 Lateral force as a function of the applied load for a 7.5-mm silica sphere interacting with a smooth silica surface in pure water and in solution 32 mM C12TAB. Data from several runs are presented.
The lateral force measurements were performed using the ”friction-force” mode of the AFM, in which the colloidal probe is pressed against the substrate at a constant applied load while the substrate slides horizontally underneath the cantilever. Further details of the measurement procedure are given elsewhere.[41-47] | In lateral force microscopy (LFM), a micron-sized colloidal probe significantly improves the reproducibility of lateral force measurements by reducing the relative impact of probe wear on the experimental results; surprisingly, however, attached particles have only been incorporated in a few AFM/LFM studies.[38,40,41] The friction force, as measured between bare silica surfaces in deionized water, was found to increase linearly with the applied load in agreement with the Amontons’ law. Accordingly, the data for these conditions can be fitted to a linear function of the type FL=a+mFN, where FN is the applied load, m is the frictional coefficient, and a is the adhesive force contribution. The average friction coefficient obtained for pure water was found to be m=0.22 and is in the range of the literature values (0.1-1.0) measured between quartz surfaces.[48]
It has been established that CH3-CH3 group interactions result in a significant reduction in friction when compared with the interactions of higher-energy functional groups[49,50] from previous studies of monomolecular films adsorbed at interfaces. The presence of surfactant at the solid-liquid interface is thought to prevent material removal by inhibiting bare surface contact or by acting as boundary layer lubricants. The interactions between intervening surfactant hydrocarbon tails and the silica surface replace the pure SiOH and SiO~ interactions found in the pure water. From this perspective, the extent of lubrication should therefore depend on the number of surfactant residing between the interface, which was found to be the case, as illustrated in Fig. 9. As free surfactant concentration increased, the lubrication between the interfaces also increases. The maximum amount of lubrication is a plateau value that occurs at and above the bulk CMC (16 mM), representing an order-of-magnitude decrease in the frictional interactions. These measurements confirm that surfactant monomer actively participates in mitigating the particle-wafer surface engagement. As noted earlier, the experiments were preformed at high loading forces at which the micellar structures have already been broken.
The enhanced lubrication at and above the CMC may be caused by the relative number of surfactant in the local vicinity of the interface. This is indicated in Fig. 10, where the extents of the repulsive force barrier at the surfactant concentrations of 16 and 32 mM are shown to be nearly identical, indicating that saturation adsorption has been reached as previously demonstrated through experimental adsorption isotherms.[21,25] Hence under these conditions, the probable number of entrapped surfactant monomers should be the highest and identical, resulting in equivalent frictional interactions. Similarly, at lower free surfactant concentrations, the extent of lubrication should be lower, resulting from fewer surfactant residing between the interfaces. These trends have been verified in Fig. 9, which illustrates that the lubrication efficacy of surfactant solution increases with free surfactant concentration until the bulk CMC is reached. Both electrostatically bound and hydrophobically associated surfactant monomers are thought to participate in the surfactant-mediated lubrication phenomena. The lowest concentration of amine shown in Fig. 9, 1 mM C12TAB, corresponds to the point of zeta-potential reversal (PZR) at which the surfactant molecules are primarily electrostatically bound to the interface. The surfactant adsorption is still as low as 20-25% of that of the saturation value at the PZR.[51,52] At higher concentrations, adsorption primarily takes place through hydrophobic association. These additional hydrophobically bound surfactant monomers participate in a further decrease of the frictional force until saturation adsorption.
Fig. 9 Lateral force as a function of the applied load for a 7.5-mm silica sphere interacting with a smooth silica surface in pure water and C12TAB solutions of various concentration.
Fig. 10 Repulsive force barrier and pull-off force as measured between a 7.5-mm silica sphere and a flat silica surface as function of the C12TAB concentration.
