Shear thickening and defect formation of fumed silica CMP slurries

Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

Contents lists available at ScienceDirect

Colloids and Surfaces A: Physicochemical and Engineering Aspects journal homepage: www.elsevier.com/locate/colsurfa

Shear thickening and defect formation of fumed silica CMP slurries Nathan C. Crawford a , S. Kim R. Williams b , David Boldridge c , Matthew W. Liberatore a,∗ a

Colorado School of Mines, Department of Chemical and Biological Engineering, Golden, CO 80401, USA Colorado School of Mines, Department of Chemistry and Geochemistry, Golden, CO 80401, USA c Cabot Microelectronics Corporation, Aurora, IL 60504, USA b

h i g h l i g h t s

g r a p h i c a l

a b s t r a c t

• A novel technique that measures rheological behavior while polishing a semiconductor wafer was developed. • Surface scratch frequency increased when thickening of the slurry was observed. • The examined scratches are believed to be the result of shear-induced agglomerates. • These agglomerates are small in number and are found exclusively in the shear thickened sample.

a r t i c l e

i n f o

Article history: Received 10 April 2013 Received in revised form 25 May 2013 Accepted 5 June 2013 Available online 15 June 2013 Keywords: Chemical mechanical polishing Fumed silica Rheology Shear thickening High shear CMP defects

a b s t r a c t During the chemical mechanical polishing (CMP) process, it is believed that shear thickening of the slurry, caused by particle agglomeration, has the potential to result in a significant increase in particle-induced surface defects (i.e. scratches, gouges, pits, etc.). In this study, we have developed a methodology for the synchronized measurement of rheological behavior while polishing a semiconductor wafer, the first of its kind (a technique termed rheo-polishing). We investigate the shear thickening of a 25 wt% fumed silica slurry with 0.15 M added KCl and its impact on polishing performance and subsequent surface damage. The thickened slurry displays a ∼5-fold increase in viscosity with increasing shear rate. As the shear rate is reduced back to zero, the slurry continues to thicken showing a final viscosity that is ∼100× greater than the initial viscosity. Optical microscopy and non-contact profilometry were then utilized to directly link slurry thickening behavior to more severe surface scratching of “polished” TEOS wafers. The thickened slurry generated up to 7× more surface scratches than a non-thickened slurry. Both slurry thickening and surface scratching were associated with a dramatic increase in the population of “large” particles (≥300 nm) which were undetectable in the non-thickened slurry. These “large” and potentially scratch-generating particles are believed to instigate measurable surface damage. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Chemical mechanical polishing (CMP) has developed into the primary technique of the semiconductor processing industry for the local and global planarization of integrated circuits [1]. The

∗ Corresponding author. Tel.: +1 303 273 3531; fax: +1 303 273 3730. E-mail address: [email protected] (M.W. Liberatore). URL: http://rheology.mines.edu/ (M.W. Liberatore). 0927-7757/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.colsurfa.2013.06.003

removal of material is achieved through a synergistic combination of chemical and mechanical attributes of the slurry (hence the name of “chemical mechanical polishing”) and the interplay at the polishing pad-wafer interface [2–5]. Recently, great emphasis has been placed on reducing the size of microelectronic devices in order to produce faster and more powerful microprocessors [6]. As a result, the semiconductor industry must constantly improve the performance of the CMP process. Today’s advanced silicon chips contain over one billion transistors in one square centimeter of surface area [5,7]. Interconnecting

88

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

Fig. 1. Transmission electron microscopy image of fumed silica (provided by Cabot Microelectronics Corporation).

such a dense population of electrical elements requires multiple layers of wiring (over eight metal layers can be found in the most recent generation of logic devices) [5]. Each level of wiring involves a minimum of two CMP steps (one metallic CMP step and one dielectric CMP step). As the semiconductor industry moves towards smaller feature sizes and to more layers of wiring (thus, more CMP steps), CMP surface defects such as scratches, gouges, pits, and corrosion need to be reduced. Even mild surface defects on a microelectronic device can greatly diminish device performance [8], while more catastrophic defects, like the severing of a wire line or the fracture of a dielectric layer, can lead to complete device failure [7]. CMP-induced defects have been linked to the existence of small populations of large particles (typically > 500 nm) within the slurry [9,6,10]. These large particles can be impurities from slurry production or a consequence of the aggressive polishing environment. During the high speed CMP process, the slurry experiences shear rates in excess of 1,000,000 s−1 [11]. Recent high shear rheological studies have shown that silica CMP slurries will display shear thickening behavior under process relevant shear rates (≥10,000 s−1 ) [11–15]. Under this intense shearing, individual slurry particles are driven together to form large agglomerates, which trigger a spike in the slurry’s viscosity (termed shear thickening [16,17]). Even though CMP-induced defects have been definitively connected to the presence of large particles within the slurry, there has yet to be a direct correlation between slurry shear thickening and imperfections on a wafer surface. Here, a method to simultaneously monitor rheological behavior while polishing a semiconductor wafer is presented (deemed in situ rheo-polishing). The objective of this manuscript is to correlate changes in slurry viscosity, specifically shear thickening, with the formation of surface scratches during polishing. The ensuing surface defects are then correlated with changes in the slurries’ particle size distribution. 2. Experimental For this study, 25 wt% fumed silica slurries (d = 160±11 nm, determined via dynamic light scattering) with and without the addition of salt (0.15 M KCl) were employed. The fumed silica aggregates (Fig. 1) are made by fusing small (5–50 nm in diameter [18]), primary particles together under a high temperature flame (∼1700 ◦ C). The as received fumed silica aggregates are suspended in water and then electrostatically stabilized at pH 11 (well above silica’s isoelectric point of pH 2) through the addition of KOH. All slurry material was provided by Cabot Microelectronics Corporation (Aurora, IL) and was a simplified version of the commercial product; consisting of fumed silica, water, and KOH. Concentrated

Fig. 2. Rheo-polishing setup with TA Instruments disposable (a) Peltier plate fixture and (b) 40 mm upper plate and (c) 2 in. TEOS wafer, which is inserted into the disposable plate fixture to be “polished.” (d) Schematic of the rheo-polishing setup showing the cross-sectional view of the rheometer plate and TEOS wafer covered with fumed silica CMP slurry.

slurries (32 wt%; pH = 11) were diluted to a target solids fraction of 25% (final pH = 10.8) using ultrapure deionized (DI) water and a 0.6 M KCl solution (99% purity; Fisher Scientific, Pittsburgh, PA). The final KCl concentration (cKCl ) for the salt adjusted slurry was 0.15 M. After dilution, slurries were stored under ambient conditions for 24 h before commencing rheological tests. Slurry particle concentration (25 wt%) and ionic strength (0.15 M) remained high in order to induce thickening behavior under measurable shear rates (10,000–100,000 s−1 ; please see our previous publications for further details [14,15]). In situ rheo-polishing measurements were conducted using TA Instruments’ AR-G2 rheometer (New Castle, DE) with a parallelplate geometry. The top, rotating plate is constructed of disposable aluminum (40 mm in diameter). TA’s disposable plate fixture was used as the bottom, stationary plate (Fig. 2a–c). Inserted into the fixture housing (for “polishing”), was a 2 in. (51 mm) diameter silicon dioxide blanket wafer, where the silicon dioxide was produced

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

by the oxidation of tetraethyl orthosilicate (TEOS; C8 H20 O4 Si). The gap spacing between the surface of the TEOS wafer and the top plate was 30 ␮m, which allowed for shear rates up to 200,000 s−1 to be explored. The entire experimental zone, 40 mm plate and exposed wafer surface, were completely submerged in 2 mL of slurry material (Fig. 2d). “Pooling” of the slurry eliminated sample drying along the edge of the rheometer plate, thereby reducing the probability of defect formation from dried slurry debris. All measurements were conducted at 25 ◦ C, with temperature control of ±0.1 ◦ C provided by the Peltier plate. A steady state shear rate ramp was employed to both “polish” the TEOS wafer and to probe for the shear thickening behavior of the fumed silica slurries. Shear rates ranging from 1000 to 100,000 s−1 were examined. Immediately following the shear rate ramping step, a shear rate reduction step tested the reversibility of the shear thickening demonstrated by the fumed silica slurries. Each rheopolishing test was repeated in triplicate (at minimum) to verify reproducibility. The duration of a full shear rate ramp and reduction experiment is around 10–12 min, which is significantly longer than a typical polishing event (∼1 min). Therefore, the surface damage endured during a rheo-polishing experiment should be more extensive than an equivalent polishing step. The rheo-polishing method, unlike a “true” polishing event, has no applied normal force. In order to collect usable rheological information, a defined gap spacing between the top and bottom plates must be established and maintained throughout the measurement. When a normal force is applied at zero-shear conditions, the gap between the top and bottom plates will be reduced until a nonzero normal force is registered, with the plates separated by only a single layer of particles. Under these circumstances, the experiment becomes a measurement of rolling or grinding friction rather than a measure of rheological behavior. While this has interesting potential, it would be incapable of determining “true” changes in fluid properties. Attempts to mimic the normal polishing process by affixing a pad to the upper, rotating rheometer plate were unsuccessful. With the pad attached, all fluids (including water) displayed pseudo shear thinning behavior. The apparent shear thinning was attributed to the deformation of the pad surface with increasing shear rate (resulting in a variable rheometer gap height) and padinduced transport of slurry along the pad interface (aided by the pores within the pad and the patterned grooves lying on the pad surface). In order to maintain a well-defined velocity profile, the bottom plate (i.e. wafer) remained stationary during experimentation. Having a stationary wafer is simpler than a true CMP process where the pad and wafer both rotate (in unison). Consequently, slurry material became trapped between the individual asperities from the pad texturing and the static wafer surface; etching deep, circular grooves into the wafer. These deep “trenches” were robust and indistinguishable from the shear-induced surface defects. As a result, all data reported here were obtained using a bare aluminum top plate. Immediately following a rheo-polishing experiment, the “polished” TEOS wafer was rinsed with DI water and then stored (completely submerged) in a sealed cup of fresh DI water. The wafers were cleaned using a zero-contact method (i.e. no scrubbing or wiping of the surface) to avoid generating further damage by dragging residual slurry material across the wafer surface. To determine the degree of scratching from our rheo-polishing experiments, optical microscopy and non-contact interferometry were employed. Quantitative scratch data was obtained using Zygo Corporation’s (Middlefield, CT) NewView 6300 non-contact, three dimensional white light interferometer. The NewView 6300 was equipped with a 20× objective that allowed for a vertical resolution of <0.1 m and a lateral resolution of 0.7 ␮m. The horizontal and vertical field of view dimensions were 350 ␮m by

89

260 ␮m, respectively. Images were captured using a 640 × 480 pixel camera. Interferometer surface scans were analyzed using the GPI application in Zygo’s MetroPro software. The GPI application allows for general surface topography analysis (in the x–y plane), as well as cross-sectional analysis (in the x–z or y–z planes) of surface roughness (such as scratches or gouges from polishing). For consistency, a scratch was defined as having a minimum depth of 0.1 ␮m (instrument limit) and a lateral aspect ratio (length:width) of at least 5. In order to eliminate analysis bias, interferometer images were divided into a 9-section grid and then numbered 1–9. For every image, a random number generator was used to determine which 3 out of 9 sections were to be analyzed. Scratch width and depth dimensions were then tabulated across at least two separate wafers for each set of polishing conditions. Slurry particle sizing, after a rheo-polishing event, was determined via dynamic light scattering (DLS). Before commencing DLS analysis, fully concentrated (25 wt%) slurries were diluted 1:104 (by volume) using a stock solution of KOH (pH = 11). Neat DI water was not used for dilution to avoid the induction of particle coagulation through the alteration of the particles’ surface charge. Malvern Instruments’ (UK) Zetasizer Nano ZS at a scattering angle of 90◦ was employed for DLS sizing.

3. Results and discussion 3.1. Monitoring viscosity during polishing The fumed silica slurries’ viscosity is highly dependent on the applied shear rate (Fig. 3a and b). The flow curve of the shear thickened slurry (represented as ST; Fig. 3b) is compared to two control experiments: (1) a 25 wt% no salt added slurry and (2) DI water (Fig. 3a). As expected, the viscosity of DI water is independent of shear rate. The average measured viscosity (across the full shear rate ramp and reduction experiment) of the DI water control was 8.6 ± 0.3 ×10−4 Pa s, which is within ∼3 % of the reported value for the viscosity of water at 25 ◦ C (8.9× 10−4 Pa s [19]), further validating the rheo-polishing technique. The silica slurry with no added salt (black triangles in Fig. 3a) is non-thickening (denoted as NT) and shear thins throughout the entire shear rate ramping step from 1000 to 100,000 s−1 , exhibiting a 40% decrease in viscosity. Shear thinning is commonly observed among suspensions of non-spherical colloids like fumed silica, because the particles have a preferential orientation under flow [20,21,16,17]. As the shear rate increases, the restoring effects of Brownian motion can no longer withstand the strength of the applied hydrodynamic force and the particles begin to align in the direction of flow. This orientation of particles lowers the suspension’s resistance to motion, resulting in an overall reduction of the system’s viscosity (i.e. shear thinning). In contrast, the slurry with 0.15 M KCl, transitions from slight shear thinning to shear thickening at ∼30,000 s−1 (0.15 M KCl (ST); red squares in Fig. 3b). From 30,000 to 100,000 s−1 the salt added slurry exhibits ∼5-fold increase in viscosity. However, if the ramping step is stopped at 30,000 s−1 (or below), no thickening is observed for the salt-adjusted slurry (0.15 M KCl (NT); blue squares in Fig. 3b). Therefore, any changes to the slurry’s viscosity are completely reversible until thickening commences. Recent work by Crawford et al. [14,15] and Amiri et al. [12,13] have examined the shear thickening response of these fumed silica slurries in great detail. In all of the previous studies, the slurries displayed discontinuous thickening behavior, where a distinct “jump” in viscosity is observed. Here, the slurry with 0.15 M KCl thickens continuously during the shear rate ramping step, without any sudden increases in viscosity.

90

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

Fig. 3. Steady state shear rate ramp (filled symbols) and reduction (open symbols) for (a) DI water (gray circles) and a 25 wt% silica slurry with no salt (black triangles) and (b) with an added KCl concentration of 0.15 M (red squares are the shear thickened sample (ST), while the blue squares are the non-shear thickened (NT) sample). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Currently, the dominant explanation holds that shear thickening is caused by the shear-induced coupling of particles through strong hydrodynamic lubrication forces (termed hydroclusters) [22–24,16]. The formation of hydroclusters necessitates the entrapment of liquid between particles, forcing the newly formed clusters to dilate (increasing the effective volume of the particle network) [25]. If the dilation of particle clusters originates near a system boundary (i.e. the air–liquid interface of a traditional parallel-plate setup or the inner cylinder walls of a Couette geometry) cluster growth will be retarded by the stiffness of the nearby boundary. As a result, the dilating cluster network “jams” the suspension microstructure and an abrupt increase in viscosity is observed [26–28]. Work by Brown and Jaeger [27] reported that in a conventional parallel-plate setup the surface tension around the outer edge of the plates is required to observe discontinuous shear thickening. For this study, the top plate of our parallel-plate geometry is completely encapsulated with slurry material, eliminating the relatively “rigid” air–liquid boundary around the outer edge of the rheometer plate (where the shear rate is highest). Also, the inner-edge of the bottom plate housing sits ∼2 mm beyond the outer-edge of the top plate, allowing for spatially unhindered growth of shear-induced structures and relatively small shear rates. By definition, shear thickening by hydrocluster formation is a completely reversible phenomenon and any increase in viscosity is relinquished upon the cessation of applied shear. Yet, the thickening behavior displayed by the salt-adjusted slurry is irreversible and its viscosity not only remains thickened, but continues to thicken during the shear rate reduction step (Fig. 3b). The irreversible thickening behavior observed here agrees with previous fumed silica slurry thickening studies [14,15,12,13] and most likely follows an entirely different mechanism than what is depicted in hydrocluster theory. Aqueous silica suspensions are believed to coagulate through the process known as hydroxo interparticle bridging, where acidic surface silanol groups (SiOH) are linked together by adjacent hydroxide ions (OH− ). In order for such bridges to develop, two criteria must be satisfied [29]: 1) the OH− concentration must be abundant (i.e. silica coagulation cannot occur at low pH) and 2) the surface of the silica particles must be covered with acidic hydrogens. However, under highly alkaline conditions (pH > 10), silica has a strong negatively charged surface and undergoes a deprotonation reaction (SiOH +OH− ↔ SiO− +H2 O), resulting in an almost completely deprotonated surface (SiO− ) [12,30]. In other words,

there are no hydrogens directly associated with silica’s surface to facilitate interparticle bridging. Conversely, at high pH (> 10) and in the presence of salt, silica’s surface will be entirely covered with hydrated cations [31,32]. The layers of hydration that surround the adsorbed cations can act as binding vehicles for the deprotonated surface silanols (SiO− ) [33,29]. Water molecules surrounding the adsorbed counterions are displaced by the dissociated silanols and an irreversible “bridge” is created between two silica particles [12,34,29]. The addition of KCl to the slurry can then lead to both charge screening between particles, reducing electrostatic repulsions, and irreversible particle agglomeration through cation enhanced bridging (commonly referred to as “shear-induced bridging” in polymer systems [35–38]). Therefore, the irreversible thickening behavior observed in Fig. 3 could be the result of irreversible shear-induced particle agglomeration. The potential shear-induced agglomerates formed during rheopolishing experiments should be held together more strongly than the reversible hydroclusters created in a traditional shear thickening system. Unlike hydroclusters, these rigid agglomerates should withstand the high frictional forces that exist along the wafer surface during CMP, making them more likely to cause a CMP-induced defect during the planarization process. Also, since these agglomerates are irreversible (i.e. stay intact after a polishing event), they should be quantifiable after the rheometer is stopped. In the following section, we investigate the polishing performance of the three rheo-polishing cases described in Fig. 3a and b (two non-thickening cases and one shear thickening case, denoted as NT and ST respectively). The three cases will allow us to indisputably connect the observed shear-induced thickening response to the presence of scratches on the wafer surface. We will address the argument that the scratches occur from simply exposing the wafer surface to such high shear rates (i.e. the no salt (NT) case) and that the scratches were generated by polishing with an unstable slurry (i.e. the 0.15 M KCl (NT) case). 3.2. Characterization of shear thickening-induced scratches During the fabrication of microelectronic devices, a wafer surface will have metallic and dielectric regions that are simultaneously exposed to the CMP slurry. The width and respective thickness of these regions is on the order of ∼100 nm for current integrated chips [7]. Thus, even minor surface scratching can cause complete device failure and having the ability to understand and

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

91

Fig. 4. Surface optical microscopy images of a 2 in. TEOS wafer “polished” with: (a) DI water and a 25 wt% silica slurry containing (b) no salt (NT), (c) 0.15 M KCl (NT), and (d) 0.15 M KCl (ST).

control surface imperfections is crucial to the success of the CMP process. A CMP-induced scratch is presumably generated when a hard object (slurry particle or polishing debris) is forced into the wafer and then dragged across the surface. Generally, large (≥500 nm) particle populations are believed to be the main culprit for the formation of scratches on the surface of polished wafers [9,39]. Optical microscopy was employed to qualitatively examine the formation of surface scratches during rheo-polishing (Fig. 4a–d). When “polishing” with DI water, little to no surface scratching was observed (Fig. 4a), even though the rheo-polishing experiments were not performed in a clean room environment. The minimal scratching that was observed in this control experiment is attributed to traces of dried slurry residue adhering to the rheometer tooling (and possible environmental contamination) even after thorough rinsing with DI water between rheological tests. Scratching intensity increases systematically as you move from image (a) through image (d) in Fig. 4. As expected, the wafer polished with DI water (Fig. 4a) has the smallest number of scratches. The unsalted slurry shows a few pronounced scratches (Fig. 4b), but still has regions of essentially scratch-free surface. More scratches are clearly evident in the 0.15 M KCl (NT) case (Fig. 4c) in comparison with the no salt (NT) case (Fig. 4b). The addition of salt to the slurry matrix reduces particle stability by screening electrostatic interactions [40] and eventually leads to particle agglomeration. Therefore, increased surface defects are expected when employing high salinity CMP slurries [1,41]. Previous work by Choi et al. [3] reported that silica slurries with 0.1 M KCl had a higher material removal rate and lower surface roughness when compared to slurries with other added salts at various concentrations. Material removal rates typically increase with increasing ionic strength due to enhanced frictional forces at the particle-wafer interface. However, inconsistent contact between agglomerated particles and the wafer surface, leads to decreased

(and irregular) polishing rates. Work by Basim and Moudgil [39] demonstrated that salt-induced agglomeration under high NaCl concentrations (> 0.2 M) greatly depreciated the surface finish of polished wafers. More recent work by Chang et al. [42] showed that alkaline slurries with no added salt had fewer stress-induced agglomerates and fewer particle-induced defects in comparison to salt-adjusted slurries. Therefore, more surface scratches should be expected when using a slurry with an elevated ionic strength, even if no noticeable change in viscosity is observed (i.e. no thickening). On the other hand, when thickening is observed during the rheopolishing experiments (Fig. 4d), scratching on the wafer surface is more frequent and severe. Here, the observed thickening response is most likely due to an irreversible shear-induced agglomeration process (hence the irreversible thickening reported in Fig. 3b). The shear-induced agglomerates formed in this study appear to be stabler and more rigid than hydroclusters and consequently, are more likely to cause surface defects. Therefore, the scratches observed in Fig. 4d are a result of both naturally occurring salt-induced structures and durable shear-induced agglomerates. Optically, shear-induced thickening leads to increased scratching on the wafer surface during polishing. However, the microscopy images from Fig. 4a–d only give a 2D glimpse of the polished wafer surface and cannot distinguish between dried-on slurry material and “true” shear-induced defects. Therefore, non-contact white light interferometry was employed (Fig. 5a–d) to obtain more quantifiable scratch data. 3D surface scans of the NT cases, with and without added salt (Fig. 5b and a, respectively) display little to no surface scratching (similar to the results shown in Fig. 4). Conversely, the shear thickening case (Fig. 5c) has a much higher number of surface imperfections. The use of the surface profiler application (the cross-sectional view, in the y–z plane, from the vertical red line in Fig. 5c is shown in Fig. 5d) allowed for not only the quantification of scratch frequency, but also the width and height

92

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

Fig. 5. Interferometer surface scans of a 2 in. TEOS wafer polished with a 25 wt% silica slurry containing (a) no salt (NT), (b) 0.15 M added KCl (NT), and (c) 0.15 M added KCl (ST). The interferometer images are 350 ␮m by 260 ␮m, width by height, respectively. (d) Cross-sectional view in the y–z plane of the vertical slice from image (c) indicated by the red line in image (c). (e) Normalized number of surface scratches per image examined with the profilometer for the three different polishing cases (NT with and without added KCl and with observed ST). The displayed p-values are from a two sample t-test comparing the three polishing cases with one another. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

of each analyzed scratch (a full analysis procedure was detailed in Section 2). Shear thickening led to 7× and 3.5× more scratches per unit area of analyzed wafer surface in comparison with the no salt (NT) case and the 0.15 M KCl (NT) case, respectively (normalized scratch counts per image found in Fig. 5e). A total of 218 scratches were counted on the surface of the wafer polished under thickening conditions (across 10 analyzed images), while only 69 scratches were enumerated for the 0.15 M KCl (NT) case (across 11 images; Table 1). The no salt (NT) case showed only a total of 32 scratches (across 10 images). A two sample t-test analysis was used (at 95% confidence) to determine if the ST case was significantly different than the two NT cases (with and without added KCl). The p-values from both t-tests were much less than 0.001 (∼10−9 ; pvalues inset in Fig. 5e), indicating that the elevated scratch counts from the ST case did not occur by chance and that these results

are highly significant. The two non-thickened samples were statistically distinguishable (p-value of ∼0.01), but at a much lower level of statistical confidence. Comparing the shear thickened and non-shear thickened salt treated samples, the much higher number of scratches and very high level of statistical significance clearly shows that shear-induced agglomeration leads to surface damage at a rate not experienced by salt-induced agglomeration alone. More scratches are observed for the shear thickening case while the depth and width of the analyzed scratching events are equivalent for the ST and NT (with and without added salt) rheo-polishing cases (Table 1). All three cases have reported scratches that are approximately 0.3 ␮m deep and 3 ␮m wide (on average), indicating that the scratch-generating particles are similar in size. In order to estimate the size of a scratch-causing structure (i.e. “problem particle”) we assumed that a single, spherically shaped particle created the surface defect in question. For this analysis, the chord length

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

93

Table 1 Scratch counts and dimensions. Parameter

No salt (NT)

0.15 M KCl (NT)

0.15 M KCl (ST)

Image count Scratch count Scratch depth (␮m) Scratch width (␮m) “Problem particle” diameter (␮m)

10 32 0.20 ± 0.09 3.2 ± 1.1 13.0

11 69 0.37 ± 0.16 3.4 ± 1.1 8.2

10 218 0.25 ± 0.14 3.4 ± 1.4 11.8

and no salt (NT) cases have intensity average particle diameters of 194 nm and 175 nm, respectively. The increase in average size and a slight broadening of the distribution for the non-thickened salt added case, is most likely caused by salt-induced agglomeration (an anticipated result for such a high salinity sample). The shear thickened sample, on the other hand, has a much broader size distribution than both the NT samples (Fig. 7c) with an inten-

Fig. 6. Cross-sectional view of a “problem particle” with radius r, creating a scratch of depth d, and width w, on the surface of a TEOS wafer.

length of a sphere is equivalent to the scratch width (w) and can be calculated with the expression in Fig. 6. Using the scratch depth (d) and width values from Table 1, allowed indirect estimate of the diameter of a “problem particle” for all three polishing cases. According to these calculations, the agglomerates responsible for the measured surface damage are between 8 to 13 ␮m in diameter (assuming a spherical geometry). Particles that are multiple microns in size, let alone > 10 ␮m, are rarely found within the slurry matrix. Work by Ring and coworkers [7] characterized surface defects from both metallic and dielectric CMP and used V-shaped indentations to size impurity slurry particles. The dimensions of the impurity particles were found to lie between 1 to 4 ␮m. Ring and coworkers also reported that the resultant debris caused by material fracture at the wafer surface was much larger than the indenting particles (i.e. impurity particles). An indenting particle leads to material fracture at the wafer surface and the large fracture debris is then expected to generate even larger scratches; rapidly amplifying the number and size of surface imperfections [7]. Therefore, the size of a defect-instigating particle is not necessarily equivalent to the dimensions of the resultant surface damage. 3.3. Connecting surface scratches to particle agglomeration Current thinking within the CMP community, is that small populations (≤1 ppm) of “large” particles (≥500 nm) are the chief instigators of CMP-induced defect formation [9,39,6,43]. These large particles can be impurities from slurry production or shearinduced agglomerates originating during the high shear polishing process. Due to their small population and relatively small size (i.e. sub-micron), these damage causing particles commonly escape detection [6,9,39,10]. It seems obvious that scratches would be directly associated with the presence of large, rigid particles, however linking the two effects is quite difficult [6,10,41,44]. Previous efforts to connect surface scratching with infrequent large particles mainly stems from the limitations of particle measurement devices and the inherent evolution of the slurry system post-polishing. For this study, we employed dynamic light scattering (DLS) to explore shear-induced changes to the particle size distributions of the slurries. The particle size distributions of the non-thickened samples (Fig. 7a and b) are narrow, with polydispersity indices near 0.1 (indicating nearly monodisperse samples). The 0.15 M KCl (NT)

Fig. 7. Particle size distribution for the (a) no salt (NT), (b) 0.15 M KCl (NT) and (c) 0.15 M KCl (ST)cases. The solid lines represent standard Gaussian fits (a) and (b), and an Edgeworth–Cramer dual-peak function (c). Intensity average particle diameters are reported for their respective samples.

94

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

Table 2 Number average (aNA ) and intensity average (aIA ) particle diameters, and polydispersity idices (PDI) measured using DLS. Parameter

No salt (NT)

0.15 M KCl (NT)

0.15 M KCl (ST)

aNA (nm) aIA (nm) PDI

168 ± 4 175 ± 8 0.07

175 ± 8 194 ± 26 0.1

195 ± 8 257 ± 35 0.2

sity average particle size of 257 nm. A measurable population of large particles ranging from 300 to 500 nm is clearly evident in the shear thickened sample, suggesting the formation of shear-induced agglomerates. The noticeable population of “large” particles is not observed in the size distributions of the non-thickened samples and these sizeable particles appear to be directly associated with the measured thickening response. However, there was no visible evidence of particles in the micron size range (all of the measured particles had sub-micron sized diameters). Typically dynamic light scattering instruments have an upper size threshold of 1–2 ␮m. Particles larger than the upper instrument limit have small diffusion constants that get “lost” in the signal noise and commonly settle out of solution too fast to be quantified using DLS. Therefore, DLS is not the most appropriate tool to size particles larger than ∼2 ␮m. Table 2. Microscopy of the samples was attempted, but with little success. The majority of particles are too small for white light optical microscopy. Transmission electron microscopy (TEM) is the most suitable technique for the size range of interest (100–10,000 nm; 0.1–10 ␮m). However, distinguishing a “true” shear-induced agglomerate from a sample prep-induced agglomerate (generated through the drying of sample on a TEM grid) is difficult if not impossible. Even if it was possible, the number of particles larger than 1 ␮m are so few that it would impractical to determine their statistical relevance. Also, DLS sizes on the order of 105 particles (or more) in a single test while microscopy sizes only 102 particles over multiple images. Unfortunately, there is no reasonable technique that can simultaneously probe both the superand sub-micron size scales when the populations differ by many orders of magnitude. In general, we feel that DLS gives a more representative depiction of the “true” particle distribution and as a result, slurry particle microscopy was not included as part of this study. It is possible that large, micron sized particles are present within the slurry, however, there is no feasible method that can directly measure their abundance. Yet, if the Gaussian fits from Fig. 7 are extrapolated beyond the fitted size ranges (0–350 nm for the NT samples and 0–600 nm for the ST sample) one could get a rough estimate for the amount of particles that may exist in the larger size bins (≥500 nm). In no way is data extrapolation a direct or accurate measurement technique, it is used solely as an approximation. Therefore, extrapolated results will be examined with some reservation. Extrapolations of the DLS data in Fig. 7 were completed assuming the particle size distributions are continuous and follow a Gaussian distribution beyond the examined size range. In order to establish a reference point, the number of particles in 2 mL of 25 wt% slurry (the sample size and concentration used for the rheo-polishing experiments) was estimated. Assuming the slurry particles are spheres with an average diameter of 175 nm, ∼1014 particles are present in 2 mL of slurry. For the shear thickened sample, around 2% of the total particle population resides in the 300–500 nm size range (∼1012 total particles; Table 3). The shear thickened slurry contains ∼7x and ∼13,000x more particles in the 300–500 nm size bin than the nonthickened slurries, with and without added salt, respectively. It

is possible that the increased number of larger particles found in the thickened sample are the culprit for the observed elevation in surface damage. However, they are too small (by themselves) to create the measured scratches during rheo-polishing and are also too small to bridge the 30 ␮m rheometer gap height. Even though these “medium” sized particles are plentiful (∼1012 ), it would be challenging for them to travel along the shear gradient and contact the wafer surface. Perhaps it is more likely that a small number of micron-sized particles are the culprit? According to the extrapolated Gaussian fits, very few particles can be found above 500 nm for the non-thickened slurries. Only fractions of particles are present in the size bins ≥500 nm for the no salt (NT) sample. For the non-thickened, salt added case, around 105 particles were estimated to exist in the 500–750 nm size range and only fractions of particles are predicted to exist in the larger size bins (Table 3). The shear thickened sample is predicted to have 20,000× more particles from 500 to 750 nm than the non-thickened, salt added slurry. Again, suggesting the presence of shear-induced agglomerates (unique from salt-induced agglomerates) that correlate with the witnessed shear thickening response. The data extrapolation for the shear thickening case indicated the presence of micron sized particles. According to the fit, approximately 2× 103 particles may exist in the micrometer size range for the thickened sample. Although the relative number of particles > 1000 nm (1 ␮m) is small, they are abundant enough to cause the amount of examined surface scratching. On the order of 102 scratches were counted for the shear thickening case, so only an equivalent number of micron sized particles would be needed to create the observed number of scratches. In reality, it only takes one “large” (> 1.0 ␮m) particle to create a scratch; locating that large “problem particle” however, is where the challenge remains. Overall, the sizing of the fumed silica CMP slurries (post rheo-polishing) suggest the existence of “large,” potentially scratch-forming particles that are exclusive to the shear thickened sample. An elevated number of relatively large particles from 300 to 500 nm were found in the shear thickened sample. These “medium” size particles are small in number (making up only 2% of the total particle population) and are much smaller in magnitude than the theoretical micron sized surface-damaging particles. However, work by Remsen et al. [6] found the best correlation between large particle counts and defectivity for particles greater than ∼680 nm. The measured “medium” sized particles observed in the shear thickened sample are in the proper size range with respect to the results reported by Remsen and coworkers. Therefore, it is reasonable that the extensive scratching observed in the shear thickening case is caused by multiple “medium” sized agglomerates (< 500 nm) contacting the wafer surface and over time, these numerous surface impacts create large trenches on the wafer surface. However, we feel that it is more likely that the 300–500 nm particles would serve as “defect-instigating” particles. When these particles contact the wafer they cause material fracture at the wafer-slurry interface. The fracture debris goes on to create larger and more frequent surface damage, which perpetuates larger and more frequent polishing debris (a continual cycle). Large polish debris can be in the tens to thousands of microns in size (visible by the naked eye); large enough to cause the reported surface damage, yet too big to be sized using DLS or even encapsulated using standard pipettes (the instrument used to collect slurry from the wafer surface after a polishing event). Conversely, it is also possible that the increased amount of surface damage in correlation with the observed shear thickening behavior is a result of a small fraction of micron-sized particles. The existence of these micron scale particles, unfortunately, could not be directly measured. Yet, when the Gaussian fit from the measured DLS data was extrapolated beyond the fitted range (0–600 nm), a small population of super-micron particles was predicted to exist

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

95

Table 3 Extrapolated Gaussian fits from the DLS data in Fig. 7. Sample

Size range (nm)

Fraction of particle population (%)

Particle count (# per 2 mL)

0.15 M KCl (ST)

300–500 500–750 750–1000 > 1000

2.2 0.008 3×10−6 2×10−9

2.2×1012 8×109 3×106 2×103

0.15 M KCl (NT)

300–500 500–750 750–1000 > 1000

0.3 3×10−7 4×10−15 2×10−20

3×1011 3×105 < 1.0 < 1.0

No salt (NT)

300–500 500–750 750–1000 > 1000

0.0002 1×10−14 3×10−25 3×10−75

2×108 < 1.0 < 1.0 < 1.0

solely in the shear thickened sample (∼2×103 out of 1014 total particles). Even though the number of particles > 1.0 ␮m is relatively small, they are frequent enough to generate the amount of examined surface scratches found in the shear thickening case. 4. Conclusions We have developed a unique methodology that allows for the in situ measurement of rheological behavior while “polishing” a wafer of semiconducting material. The newly developed rheopolishing technique permits shear thickening of the CMP slurries to be directly linked to surface scratching during a polishing event. The data presented here clearly show that damage causing particles can be generated during the polishing process itself, and are not necessarily present in the slurry as introduced to the polishing tool. In this study, the frequency of surface scratches significantly increased when thickening of the fumed silica slurry was observed. The examined scratches are believed to be the result of shear-induced agglomerates, which are different than the quiescent salt-induced agglomerates present in the non-thickened samples. These shearinduced agglomerates exist in small numbers, accounting for only ∼2% (or less) of the slurry’s total particle population (making them challenging to detect). However, these undesirable agglomerates are frequent enough to lead to statistically significant and severe surface damage. Recent efforts to protect the smaller, more delicate feature sizes of today’s advanced microelectronic devices, have lead to polishing with smaller applied down forces and with slurries containing lower solids loadings (< 10 wt%). The work presented here, is an indication that decreasing the platen speed could also help combat the formation of CMP-induced defects, which can destroy these delicate features. While reducing platen speed decreases polishing rates and device throughput, the increase in successful device yield might counterbalance the loss in production from decreased polishing speeds. Thus, reducing platen speeds (i.e. polishing shear rates) in combination with lower solids fraction slurries and smaller down forces, should be the path forward for the CMP community. Reducing all three of these parameters will help keep smaller features intact after polishing and help prevent shear-induced agglomeration (as witnessed here). Acknowledgments The authors thank the National Science Foundation (CBET0968042) and Cabot Microelectronics Corporation (CMC) summer internship program for providing the funding for this work. In addition, we acknowledge CMC for supplying the slurries and for allowing us to share our findings. We also thank Mr. Stan Lesiak for his assistance with the profilometry for this study.

References [1] R.K. Singh, S.M. Lee, K.S. Choi, G.B. Basim, Z. Chen, B.M. Moudgil, Fundamentals of slurry design for cmp of metal and dielectric, MRS Bull. 27 (11) (2002) 833. [2] Z. Lu, S.H. Lee, S.V. Babu, E. Matijevic, The use of monodispersed colloids in the polishing of copper and tantalum, J. Colloid Interface Sci. 261 (1) (2003) 55–64, doi:10.1016/S0021-97979(02)00166-2. [3] W. Choi, U. Mahajan, S.M. Lee, J. Abiade, R.K. Singh, Effect of slurry ionic salts at dielectric silica cmp, J. Electrochem. Soc. 151 (3) (2004) G185–G189, doi:10.1149/1.1644609. [4] P.B. Zantye, A. Kumar, A.K. Sikder, Chemical mechanical planarization for microelectronics applications, Mater. Sci. Eng. Rep. 45 (3-6) (2004) 89–220, doi:10.1016/j.mser.2004.06.002. [5] E. Matijevic, S.V. Babu, Colloid aspects of chemical-mechanical planarization, J. Colloid Interface Sci. 320 (2008) 219–237. [6] E.E. Remsen, S. Anjur, D. Boldridge, M. Kamiti, S. Li, T. Johns, C. Dowell, J. Kasthurirangan, P. Feeney, Analysis of large particle count in fumed silica slurries and its correlation with scratch defects generated by CMP, J. Electrochem. Soc. 153 (5) (2006) G453–G461. [7] T.A. Ring, P. Feeney, D. Boldridge, J. Kasthurirangan, S.T. Li, J.A. Dirksen, Brittle and ductile fracture mechanics analysis of surface damage caused during cmp, J. Electrochem. Soc. 154 (3) (2007) H239–H248. [8] J.M. Steigerwald, S.P. Murarka, R.J. Gutmann, Chemical Mechanical Planarization of Microelectronic Materials, John Wiley & Sons, New York, 1997. [9] G.B. Basim, J.J. Adler, U. Mahajan, R.K. Singh, B.M. Moudgil, Effect of particle size of chemical mechanical polishing slurries for enhanced polishing with minimal defects, J. Electrochem. Soc. 147 (9) (2000) 3523–3528, doi:10.1149/1.1393931. [10] M. Stintz, H. Barthel, M. Moinpour, Particle metrology in CMP slurries—potential and limitations of relevant measuring methods, Mater. Res. Soc. Symp. Proc. 991 (2007) 107–118. [11] W. Lortz, F. Menzel, R. Brandes, F. Klaessig, T. Knothe, T. Shibasaki, News from the M in CMP—viscosity of CMP slurries, a constant? Mater. Res. Soc. Symp. Proc. 767 (1) (2003) F1.7.1–F1.7.10. [12] A. Amiri, G. Oye, J. Sjoblom, Influence of pH, high salinity and particle concentration on stability and rheological properties of aqueous suspensions of fumed silica, Colloids Surf. A: Physicochem. Eng. Aspects 349 (1-3) (2009) 43–54. [13] A. Amiri, G. Oye, J. Sjoblom, Stability and flow-induced flocculation of fumed silica suspensions in mixture of water–glycerol, J. Dispers. Sci. Technol. 33 (8) (2012) 1247–1256, doi:10.1080/01932691.2010.527811. [14] N.C. Crawford, S.K.R. Williams, D. Boldridge, M.W. Liberatore, Shear thickening of chemical mechanical polishing slurries under high shear, Rheol. Acta 51 (7) (2012) 637–647. [15] N.C. Crawford, B. Yohe, S.K.R. Williams, D. Boldridge, M.W. Liberatore, Shear thickening and shear-induced agglomeration of chemical mechanical polishing slurries using electrolytes, Rheol. Acta (2013), doi:10.1007/s00397-013-07119. [16] N.J. Wagner, J.F. Brady, Shear thickening in colloidal dispersions, Phys. Today 62 (10) (2009) 27–32. [17] J.M. Brader, Nonlinear rheology of colloidal dispersions, J. Phys. Condens. Matter 22. (36) (2010), Article number = 363101. [18] D. Boldridge, Morphological characterization of fumed silica aggregates, Aerosol Sci. Technol. 44 (2010) 182–186. [19] J. Kestin, M. Sokolov, W.A. Wakeham, Viscosity of liquid water in the range −8 ◦ C to 150 ◦ C, J. Phys. Chem. Ref. Data 7 (3) (1878) 941–948. [20] R.L. Hoffman, Discontinuous and dilatant viscosity behavior in concentrated suspensions. II. Theory and experimental tests, J. Colloid Interface Sci. 46 (3) (1974) 491–506. [21] J.J. Stickel, R.L. Powell, Fluid mechanics and rheology of dense suspensions, Ann. Rev. Fluid Mech. 37 (2005) 129–149, doi:10.1146/annurev.fluid. 36.050802.122132. [22] J.F. Brady, G. Bossis, Stokesian dynamics, Ann. Rev. Fluid Mech. 20 (1988) 111–157. [23] J.W. Bender, N.J. Wagner, Optical measurement of the contributions of colloidal forces to the rheology of concentrated suspensions, J. Colloid Interface Sci. 172 (1) (1995) 171–184.

96

N.C. Crawford et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 87–96

[24] B.J. Maranzano, N.J. Wagner, Flow-small angle neutron scattering measurements of colloidal dispersion microstructure evolution through the shear thickening transition, J. Chem. Phys. 117 (22) (2002) 10291–10302. [25] E. Brown, H.M. Jaeger, The role of dilation and confining stresses in shear thickening of dense suspensions, J. Rheol. 56 (4) (2012) 875–923, doi:10.1122/1.4709423. [26] A. Fall, N. Huang, F. Bertrand, G. Ovarlez, D. Bonn, Shear thickening of cornstarch suspensions as a reentrant jamming transition, Phys. Rev. Lett. 100 (1) (2008), http://dx.doi.org/10.1103/PhysRevLett.100.018301, Article number = 018301. [27] E. Brown, H.M. Jaeger, Dynamic jamming point for shear thickening suspensions, Phys. Rev. Lett. 103 (8) (2009), http://dx.doi.org/10.1103/ PhysRevLett.103.086001, Article number = 086001. [28] E.B. White, M. Chellamuthu, J.P. Rothstein, Extensional rheology of a shearthickening cornstarch and water suspension, Rheol. Acta 49 (2) (2010) 119–129, doi:10.1007/s00397-009-0415-3. [29] J. Depasse, Coagulation of colloidal silica by alkaline cations: Surface dehydration or interparticle bridging? J. Colloid Interface Sci. 194 (1) (1997) 260–262, doi:10.1006/jcis.1997.5120. [30] S. Chen, G. Oye, J. Sjoblom, Effect of pH and salt on rheological properties of Aerosil suspensions, J. Dispers. Sci. Technol. 28 (6) (2007) 845–853. [31] R.M. Pashley, DLVO and Hydration forces between mica surfaces in Li+ , Na+ ,K+ , and Cs+ electrolyte solutions—a correlation of double-layer and hydration forces with surface cation-exchange properties, J. Colloid Interface Sci. 83 (2) (1981) 531–546, doi:10.1016/0021-9797(81)90348-9. [32] G. Peschel, P. Belouschek, M.M. Muler, M.R. Muller, R. Konig, The interaction of solid-surfaces in aqueous systems, Colloid. Polym. Sci. 260 (4) (1982) 444–451, doi:10.1007/BF01448150. [33] J.P. Chapel, Electrolyte species-dependent hydration forces between silica surfaces, Langmuir 10 (11) (1994) 4237–4243, doi:10.1021/la00023a053.

[34] B.P. Binks, S.O. Lumsdon, Stability of oil-in-water emulsions stabilized by silica particles, Phys. Chem. Chem. Phys. 1 (12) (1999) 3007–3016, doi:10.1039/a902209k. [35] Y. Otsubo, Size effects on the shear-thickening behavior of suspensions flocculated by polymer bridging, J. Rheol. 37 (5) (1993) 799–809, doi:10.1122/1.550464. [36] M. Kamibayashi, H. Ogura, Y. Otsubo, Shear-thickening flow of nanoparticle suspensions flocculated by polymer bridging, J. Colloid Interface Sci. 321 (2) (2008) 294–301. [37] J. Zebrowski, V. Prasad, W. Zhang, L.M. Walker, D.A. Weitz, Shake-gels: shearinduced gelation of laponite–PEO mixtures, Colloids Surf. A: Physiochem. Eng. Aspects 213 (2-3) (2003) 189–197, doi:10.1016/S0927-7757(02)00512-5. [38] F. Bossard, V. Sfika, C. Tsitsilianis, Rheological properties of physical gel formed by triblock polyampholyte in salt-free aqueous solutions, Macromolecules 37 (10) (2004) 3899–3904, doi:10.1021/ma0353890. [39] G.B. Basim, B.M. Moudgil, Effect of soft agglomerates on CMP slurry performance, J. Colloid Interface Sci. 256 (2002) 137–142. [40] W.B. Russel, D.A. Saville, W.R. Schowalter, Colloidal Dispersions, Cambridge University Press, Cambridge, 1989. [41] M. Bielmann, U. Mahajan, R.K. Singh, Effect of particle size during tungsten chemical mechanical polishing, Electrochem. Solid State Lett. 2 (8) (1999) 401–403, doi:10.1149/1.1390851. [42] F.C. Chang, P. Kumar, R. Singh, K. Balasundaram, J. Lee, J. Lee, R.K. Singh, Role of interparticle forces during stress-induced agglomeration of CMP slurries, Colloids Surf. A: Physiochem. Eng. Aspects 389 (1-3) (2011) 33–37, doi:10.1016/j.colsurfa.2011.09.001. [43] B. Tolla, D. Boldridge, Distortion of single-particle optical sensing (SPOS) particle count by sub-countable particles, Part. Part. Syst. Charact. 27 (2010) 21–31. [44] L. Anthony, J. Miner, M. Baker, W. Lai, J. Sowell, A. Maury, Y. Obeng, The how’s and why’s of characterizing particle size distributions in CMP slurries, Electrochem. Soc. Proc. 98 (7) (1998) 181–196.