Particle Technology in Chemical Mechanical Planarization †

In order to keep pace with Moore’s law, multilevel metallization has become the process of choice. Having a planar wafer surface before every successive step in multilevel metallization is important. The Chemical Mechanical Planarization (CMP) process is used in the semiconductor industry to achieve planar surfaces at every step of the multi-level metallization. In CMP, planarization is achieved primarily due to material removal by the use of abrasive particle slurries. In addition to the slurry chemistry, slurry performance is also dictated by the properties of the abrasive particles. A better understanding of the properties of abrasive particles and particle abrasion mechanisms will lead to better CMP. Some of the challenges in particle modeling, slurry stabilization and particle induced lubrication as well as recent developments in engineered particles for CMP are discussed in


Introduction
Since the invention of the integrated chip (IC) by Jack Kilby in 1959, one of the main objectives of the semiconductor industr y has been to increase the number of transistors on the IC in order to improve the processing performance.This fact is also corroborated by Gordon Moore's obser vation, now famously known as Moore's law: "The number of transistors occupying a given space doubles every two years" 1) .The 65nm node technology, which indicates the minimum DRAM ?pitch, is the current norm in the IC industr y and the 45nm node technology is slated for production in 2010 2) .For sub-0.5μm feature sizes, one of the barriers to improved performance is the interconnect delay.The RC time delay (a measure of interconnect delay due to resistance R and capacitance C of the metal line) is the total time taken by the voltage at one end of metal line to reach -63% of the total applied step input at the other end 3) .
The time delay increases with decrease in thickness of metal and dielectric and is independent of the line width.Thus, performance benefits expected from device scaling are offset by the interconnect delays.Hence one of the strategies employed to reduce the interconnect delay, apart from decreasing the metal resistivity and dielectric constant, is to reduce the line length.Multilevel Metallization (MLM) is employed for forming metal interconnections and its use also helps in reducing the line length, by dividing the interconnect length over multiple levels as opposed to a single plane.With the reduction in line dimensions, maintaining a planar surface topography after each subsequent processing step becomes important in MLM; an uneven surface topography created during each step in MLM presents challenges such as overexposures during photolithography thus changing the crucial dimensions and line contact interruptions among others.Many techniques such as reactive-ion etch and etch-back, spin on deposition and etch-back, flow-anneal processing are available for planarization, but compared to them chemical mechanical planarization is a relatively simple process and the only technique able to achieve excellent local as well as global planarization 4)-6) .Chemical Mechanical Planarization (CMP) is commonly used in the semiconductor industry to achieve smooth and planar surfaces on wafers by chemical and mechanical means 4) .The main components of the CMP process are wafer, pad and slurr y.Semiconductor wafer is the surface which needs to be planarized.Pad helps in transferring the mechanical forces to the wafer surface as well as helps in transport of slurry and materials removed from the wafer.Slurr y is an aqueous suspension which contains chemicals providing chemical action like modifying the surface layer, and abrasive particles which provide the mechanical action for material removal.The wafer moves in a pre-set geometric pattern against the polymeric pad, which is generally porous with embedded groves, onto which the slurry is fed; material removal occurs from the wafer surface rendering it smooth 7) .In the case of tungsten, the hardness and Young's modulus values of surface layer of tungsten are-22.5GPaand-450GPa respectively; while those for tungsten oxide (WO3) are-7.5GPaand-125 GPa respectively 8) .It is easier to abrade softer tungsten oxide as compared to tungsten.In tungsten CMP the chemical aspect comes into play by converting the surface layer of tungsten to tungsten oxide with the help of oxidizing agents such as ferricyanidephosphate 9) while the abrasive particles provide the mechanical aspect of abrading the softer sur face layer.The chemical aspect can also come into play through what is known as the chemical-tooth action' 10) , where the dissolution products from the substrate are adsorbed onto the particle surface and there is a net positive removal away from the surface.In case of glass polishing by ceria abrasives 10) , ceria forms bonds with silica which are ruptured due to particle movement away from the bonding site; also dissolved silica is adsorbed onto the abrasive particles, eventu-ally leading to silica build-up on ceria particle surface and material removal from the substrate.
In 2005, the CMP consumables market for slurries, pads and pad-conditioning agents was worth US$1.1 billion and is expected to grow to US$1.9 billion by 2009 11) ; the CMP slurry market alone was estimated to be US$522 million in 2005 12) with anticipated growth over the next few years to US$1 billion wherein lies the opportunity and challenges for the whole particle community.In the following sections some of the particle technology related issues in CMP are discussed.

Particle Based Modeling Approach in CMP
The major pad properties affecting the polishing process are its composition, reactivity, roughness and hardness.Slurry aids in modification, removal and transport of the wafer surface material.The substrate being polished will also have its unique inherent properties.CMP modeling hence is a challenging task due to the sheer number of variables involved.Most of the modeling efforts in CMP are inspired by Preston's equation 13) .It relates the material removal rate (MRR) of a glass substrate to the pressure applied and to the relative velocity between the pad and the substrate being polished, as given by the following equation MRR = KpPV Where, Kp = Preston coefficient P = Pressure V = Velocity Preston's equation is empirical in nature and the exact effect of process variables, e.g.abrasive hardness and size, on Kp is unclear.Brown developed a simple particle based, two body abrasion model for metal polishing 14) .It considers single abrasive particle penetration into the metal surface using Hertz elastic contact theory and relates the MRR as 1 PV 2E Where, E= Young's Modulus of the wafer surface For glass polishing, it was observed by Cook that Brown's model overestimated the polishing rates by over one order of magnitude 10) .This discrepancy was attributed to probable changes in the mechanical properties of the surface due to formation of a layer on it.These preliminar y models are too simplistic with very limited applicability.In actuality the abrasion in CMP is a three-body problem, where pad and wafer slide against each other with particles inbetween; and material removal occurs due to the cutting action of abrasives on the soft chemicallymodified wafer surface 15) .This section takes a look at some of the recent models developed taking into account the above mentioned factors.

Abrasive rolling/sliding model
During polishing, the abrasive particle comes in contact with the pad as well as the wafer.This model 16) proposes that at low pressures the friction at the substrate-abrasive interface is higher, causing the abrasive particles to roll against the wafer substrate; this leads to negligible material removal.As the pressure increases, the friction at the pad-abrasive interface increases and beyond a threshold pressure the abrasives start being dragged by the pad across the substrate causing a sliding motion of the particles leading to appreciable material removal.Choi et al 17) have corroborated this model on the basis of insitu friction force measurements during polishing.For a given applied load, the higher the interaction at the pad-abrasive-substrate three-point interface, the higher the material removal due to sliding friction resulting from motion of the abrasive particles.On the other hand, as the interaction at the padabrasive-substrate three-point interface decreases due to lower effective contact area, material removal decreases as a consequence of reduced friction due to rolling motion of the abrasive particles.For a given load, sliding occurs at low solids loading or smaller abrasive sizes and, rolling occurs at high solids loading or large abrasive sizes.This model needs further quantification for determining the threshold pressure where rolling motion changes to sliding motion.

Abrasive contact-area and indentation-volume model
This model proposed by Mahajan et al 18) is based on the obser vations of the oxide CMP using silica particles.Depending on the solids loading and particle size, the material removal occurs via either contact-area mode or the indentation volume mode according to this model.The higher the concentration of abrasive particles and the smaller their size, the more the abrasive surface area comes into contact with the substrate being polished.In the contact-area mode, smaller particles and higher solids loading would lead to a high MRR.The larger the particles and smaller the solids loading, the more the pressure exerted per particle.In the indentation-volume mode, large particles at lower solids loading would create a bigger indentation volume in the substrate leading to higher MRR.Indentation volume refers to the volume of the surface recesses caused due to the particles.The mathematical expressions for this model follow from the relation developed by Brown et al 14) A ∝ C0 1/3 φ −1/3 V ∝ C0 −1/3 φ 4/3 Where, A = total contact area V = total indentation volume C0 = solids loading φ = particle size Mahajan et al 18) do not give any conditions when the material removal process would cross the threshold between the two modes. 19)ost of the CMP models assume a uniform particle size, however Luo and Dornfeld 19) considered the effect of abrasive size distribution in their model.This model is an extension of their previous work 20) based on the concept of active abrasives capable of material removal.It is assumed that the polishing pad has uniform asperities and the active abrasives must be located on the contact area between pad and substrate; also the active abrasives must be larger than the gap formed between pad and wafer substrate due to intermediate presence of a large abrasive particle.This model has been shown to correlate the experimental results obtained by Bielmann et al 21) for tungsten CMP using alumina particles.The MRR has been shown to be proportional to abrasive size by the The average size of active abrasives increases with standard deviation while the number of active abrasives required for maintaining the same contact area with the substrate decreases.The standard deviation range of abrasive size has been divided into a number dominant region and a size dominant region based on this model.In the size dominant region the active abrasive size as well as MRR increases with standard deviation, and the increase in size of active abrasives is faster than decrease in their number.In the number dominant region the number of active abrasives as well as MRR decreases with standard deviation, and the decrease in number of active abrasives is faster than increase in their size.The use of monodisperse abrasives in a slurr y might lead to a low MRR if the standard deviation falls in the size dominant region.This model needs further expansion to obtain optimum value of standard deviation required to achieve effective material removal as well as to reduce scratches.

CMP Slurr y Stabilization
Golini and Jacobs 22) suggest that in polishing the abrasive particle size is generally less than 0.5μm as opposed to microgrinding.Also, mono-dispersed abrasive particle size distribution (PSD) leads to a very smooth surface, while poor abrasive PSD control leads to excessive surface scratching 10) .Basim et al. 23), 24) has showed that the formation of soft-agglomerates or presence of a small percentage of large size abrasive particles in the CMP slurry leads to increase in surface roughness as well as number of surface defects, which can be directly correlated to the size as well as concentration of the soft-agglomerates and large-particles.The initial particle size in CMP slurries is ver y less and also filtration systems are used before the slurry reaches the polisher, hence the observed defects on the wafer can be attributed to formation of agglomerates during CMP process 25)   .Thus, the stability of CMP slurries with respect to particle agglomeration is an important issue in order to achieve effective planarization.

Challenges in CMP slurr y stabilization
The CMP slurries present a unique challenge for stabilization due to their chemistries and extreme pH 4), 26), 27) .The slurry should have adequate viscosity in order to maintain an adequate flow rate during the polishing 4) .The slurry is also subjected to high shear and normal forces 4), 27) .Electrostatic Stabilization is not sufficient in the high ionic strength environment, because the charge screening will render electrostatic repulsion ineffective, which can lead to formation of agglomerates and increased defectivity.Steric Stabilization and Electrosteric Stabilization using Poly (-mer or -electrolyte) may increase the viscosity of the slurry, thereby affecting its flow rate 28) ; this may also lead to localized slurry concentration gradient on the wafer, thereby leading to uneven etch and removal rates, and defectivity.There have been some recent attempts to stabilize the CMP slurr y using polymer / polyelectrolyte 29), 30) and, though material removal selectivity was shown to increase, it was found that the viscosity of the solution increased as well as the abrasive PSD became uneven; defec-tivity studies 30) suggest an increase in within-wafer non-uniformity, with an increase in the molecular weight of the polymer.Thus, CMP slurries warrant either new stabilization mechanisms or a synergistic combination of different stabilization mechanisms.Amphiphilic surfactant molecules have been shown to fit the criterion to disperse particles in conditions similar to CMP slurry 28) , hence researchers have focused on surfactants for achieving dispersion in CMP slurries 25)-27) .

Lubrication in stabilized CMP slurries
Adler et al 28) showed that resistance to elastic deformation of surfactants is the primary stabilization mechanism occurring in high ionic strength silica slurries (pH-4) similar to CMP slurries, which were stabilized using surfactant molecules.It was observed that the particle agglomeration was found to decrease as sufficient amount of surfactant became available to cover the particle fully, which was also shown to decrease the defectivity as measured by the surface roughness 25)-27) .In CMP slurries stabilized using CnTAB surfactant molecules, Basim et al 27) obser ved that the slurr y stability increased as the surfactant chain length increased due to interparticle repulsion; but it led to a lowering of material removal rates (Fig.

3.2a).
It was seen that when the surfactant was added to the baseline slurry in the absence of salt, the lateral (frictional) force between the particle and substrate remained constant even though the loading (normal) force was increased; this explains the possible lower material removal rates obser ved.When salt was added, it was obser ved that the lateral force increased beyond a cer tain loading force, which  can be explained due to charge screening between the negatively charged silica and positive surfactant headgroup leading to surfactant desorption from the particle at higher loading force (Fig. 3.2b).Thus it was shown that along with particle-particle interactions, particle-substrate interactions should also be considered in formulation of stable CMP slurries.

Engineered Abrasives for CMP Slurries
Apart from the chemistry of slurry involved, slurry performance is also governed by the properties of the abrasive particles.Just to name a few, hardness, chemical composition, shape and density of the particles control the slurry behavior as well as affect the CMP performance.Though ceria is better in material removal of silica due to chemical-tooth action 10) it can pose problems with respect to slurr y stability due to its relatively high density of -7g/cc leading to clogged slurry flow lines 31) .Also, it is difficult to control the size and shape of synthesized ceria abrasives.
Similarly silica abrasives are ineffective in copper CMP; and alumina particles due to their relatively high hardness of 9-8 on mho's scale 4) are more susceptible to cause scratches during polishing.In order to overcome these problems and take advantages of some of the inherent properties of the abrasive material, researchers have engineered specific abrasives for CMP slurries.

Mixed abrasive slurries
Silica abrasives are known to have a minor effect on copper material removal, but are effective in polishing of tantalum and the protective oxide film formed on it due to chemical-tooth action 32), 33) .In order to achieve comparable MRR of copper and tan-talum during the second step CMP of wafer with copper interconnects having tantalum as diffusion barrier material, Jindal et al 34) used the mixed abrasive slurry of alumina and silica abrasives for CMP.The mean aggregate particle sizes for calcined alumina and fumed silica abrasives were 220 and 70-90nm respectively and, the slurry pH was maintained around 4.
When a slurry of 0.5% alumina and 2.5% silica maintained at pH of 4 was used, the smaller silica particles formed a sheath around the alumina particles due to electrostatic interactions.This led to decreased polishing action of alumina on copper.It was seen that MRR of copper and tantalum were of similar magnitude as compared to when abrasives were used independently.

Nanoporous abrasive slurries
Scratching and de-lamination of the wafer substrate materials, especially of low-k dielectric materials, are some of the major problems encountered in CMP.These problems can be overcome if the hardness of  the abrasive is reduced while maintaining its ability to chemically modify the wafer while polishing, this can be achieved by synthesis of porous abrasive particles 35) .
The porous silica nanoparticles were synthesized by hydrolysis and condensation reaction using glycerol as porogen; on calcination of the dried particles, the traces of organic glycerol were removed creating a porous network in the particles 35) .It was observed that the hardness decreased while the elasticity of the particles increased with porosity leading to lower scratching during CMP of black diamond 36) .
The slurr y formulated from these particles was also observed to be stable owing mainly due to a decrease in Hamaker constant of the porous particles 37) .

Summar y and outlook
Though many facets are now known, the exact science of CMP is still not completely understood.The role of abrasive particles in CMP is just one of the parameters controlling the planarization process.The modeling effort in CMP mainly takes into account the mechanical aspects of the particle and surface and rarely is the chemical aspect considered.There is a need for developing the models taking both the chemical and mechanical aspects into account which can satisfactorily explain the experimental observations.The models need to incorporate parameters which would yield the degree of global planarity in addition to acceptable material removal rate.As in many particle industries, slurr y stability is still an issue in CMP which is further magnified due to scratching problems created by particle aggregation.Particle technology needs to address the issue of designing abrasive particles that are able to provide stable slurries as well as achieve efficient planarization.Our understanding of the CMP process needs to be further enhanced in order to tailor the particles according to substrate and application requirements.Also new routes to utilize CMP for producing tailored surfaces using functionalized particles need to be explored.Lastly CMP is a means to an end and not an end in itself; so the particle community has an opportunity to apply the CMP knowledge to fields other than microelectronics.

Fig. 1 .
Fig. 1.Schematic of CMP process; ωs and ωp refer to angular velocity of carrier and platen respectively.(Fig. 1 from reference 7; Reproduced by permission of the MRS Bulletin.)

Fig. 3 :
Fig. 3: AFM images of the silica wafers polished with (A) baseline 0.2μm 12wt% monosize silica slurry and (B) slurry with dry aggregates (Reprinted from reference 24 with permission from Elsevier).Fig. 2.3: Material Removal as function of standard deviation of the abrasive size (Reprinted from reference 19 with permission from IEEE; c 2003 IEEE).

Fig. 3 .
Fig. 3.2a: C 12 TAB, C 10TAB, and C 8 TAB surfactants at 32, 68, and 140 mM concentrations in the presence of 0.6 M NaCl at pH 10.5: slurry particle size (stability) and material removal rate responses (Reprinted from reference 27 with permission from Elsevier).

Fig. 3 .
Fig. 3.2b: AFM friction force measurements on silica wafer with 7.5-μm-size particle attached to the tip for solutions containing C 12 TAB, C 10 TAB, and C 8 TAB surfactants at 32, 68, and 140 mM concentrations in the absence of salt at pH 10.5 (on left) and in the presence of 0.6 M salt in the solution (on right) (Reprinted from reference 27 with permission from Elsevier).

Fig. 4 .
Fig. 4.2a: Change in hardness (on left) and Hamaker constant (on right) of silica particles with porosity (Reprinted from reference 37).

Fig. 4. 1 :
Fig. 4.1:Copper and tantalum disk polish rates by slurries containing alumina/fumed silica particles dispersed in DI water at pH 4 with 3 wt % total particle loading (Reprinted from reference 34; Reproduced by permission of ECS -The Electrochemical Society).

Fig. 4 .
Fig. 4.2b: Change in number of scratches on black diamond, silica and silicon nitride substrates polished with 0.2μm silica slurry (5wt% loading, pH 3) as a function of surface porosity of silica particles (Reprinted from reference 36).