The effects of common slurry additives on the colloidal behavior of alumina suspensions used for copper chemical mechanical planarization ͑CMP͒ were investigated. The alumina suspensions were characterized by zeta potential and agglomerate size distribution measurements with various chemical additives. To simulate the slurry during copper CMP, the effect of the addition of ϳ100 nm diameter copper particles was studied. The presence of 0.12 mM copper caused a decrease in agglomeration for pH values less than 6.5 and an increase in agglomeration ranging from 200-1000 nm for pH values greater than 7 in aqueous solutions. Addition of glycine caused the formation of a soluble Cu-glycine complex that decreased agglomeration at pH values less than 4. The addition of 0.1 wt % H 2 O 2 did not affect the effective alumina agglomerate size without copper, but with copper in the solution the majority of the alumina agglomerated to ϳ2 m for all pH values. However, increasing H 2 O 2 concentration to 2.0 wt % decreased the agglomerate size by 100-400 nm. The pH of the slurry had the largest effect on the zeta potential and agglomerate size distributions.

The measured agglomerate size distributions of alumina abrasives in various slurry chemistries and at different pH values were used in a model to predict material removal rates ͑MRRs͒. The alumina agglomerate size and distribution were measured both with and without the presence of copper nanoparticles in the solution for each slurry chemistry studied. Although the agglomerate sizes were measured under quiescent conditions, it is determined that the agglomerated abrasive particles remain intact during chemical mechanical planarization ͑CMP͒, hence the measurements can be used in the CMP model. The model predictions using these measurements both with and without copper in solution were compared to experimental copper CMP data. The model was unable to predict the MRR when the slurry did not have any chemical additives because the dispersion was unstable and small fluctuations in the agglomerate size and distribution caused large changes in the predicted MRR. The model predictions were in excellent agreement with experimental MRR for a slurry with 0.1 M glycine in alkaline conditions. The model results from the size distribution measurements with copper in solution agreed slightly more with experiment than those without copper.

Nanohardness and etch rates of copper films sputter deposited onto a 30 nm tantalum coating on silicon wafers were measured after exposure to aqueous solutions containing various common chemical mechanical planarization slurry additives at different pH values. In most cases, the measured hardness values were consistent with the formation of surface films as indicated by the equilibrium potential-pH diagrams. In general, when the pH is low ͑Ͻ4͒, hardness values are that of Cu metal or slightly higher. As the pH increases to ϳ8, the hardness decreases to less than Cu metal as hydroxides form, or to higher values than Cu metal as oxides form. Exposure to solutions with glycine or ethylenediaminetetraacetic acid caused hardness values to be less than Cu metal, as areas of the surface became porous. Exposure to H 2 O 2 causes harder films in some areas and very porous soft films in other areas on the same surface, as passivation and dissolution of the surface occurs.Copper has become the interconnect material of choice for integrated circuits due to its low electrical resistivity and high thermal conductivity. 1 Copper metallization is mainly performed by electroplating of the single or dual damascene process with a suitable diffusion barrier layer ͑e.g., tantalum͒ followed by chemical mechanical planarization ͑CMP͒, which is needed to remove excess material and provide a globally planarized wafer surface. 2 The CMP process uses a slurry containing abrasive particles and chemical additives that account for both the chemical and mechanical action of material removal on the wafer surface. 3 The material removal rate ͑MRR͒ is significantly affected by the addition of chemical additives to the slurries. 2 These additives control the state of the copper ͑CuO, Cu 2+ , Cu 2 O, etc.͒ in the slurry and on the surface of the wafer, and need to be optimized so that both the interactions with the wafer surface and the effects on the abrasive particles in the slurry will provide an adequate MRR and planarized surface with minimal defects. Our previous experimental work investigated the effects of common slurry additives on the colloidal behavior of alumina suspensions used for copper CMP. 4,5 We used this data in the Luo and Dornfeld model of CMP 6,7 in order to understand and predict copper CMP. 8,9

The influence of common slurry additives on the colloidal behavior of alumina suspensions and the surface characteristics of copper were investigated. The effects of the addition of glycine and H2O2 to aqueous suspensions of alpha-alumina at various pH values were studied by zeta potential and agglomerate size distribution measurements. Depending on the state of copper in solution, alumina agglomerate size was found to either increase or decrease. Nanohardness of 1 µm copper films deposited onto 30 nm Ta on a silicon wafer was measured after exposure to the same slurry solutions and was found to be significantly affected by pH and addition of chemical additives. The agglomeration behavior and hardness measurements suggest that the states of the copper in solution or on the surface are consistent with potential-pH diagrams.

Measurements of copper nanohardness and etch rate were used with alumina agglomerate size distributions in a model to predict material removal rates ͑MRRs͒, which were then compared to experimental copper chemical mechanical planarization ͑CMP͒ data. Generally, model predictions improved using measured nanohardness compared to predictions using a constant nanohardness of Cu metal. When the slurry pH was acidic ͑Ͻ4͒ the model overpredicted the MRR. An increase in the slurry pH ͑Ͼ7͒ increased the nanohardness, and MRR predictions agreed with experimental results. For slurries with small etch rates ͑Ͻ8 nm/min͒, the nanohardness had little effect on the MRR predictions, and the model agreed with experiment. The model was very sensitive to the nanohardness for slurries with large etch rates ͑Ͼ8 nm/min͒, and was unable to predict the MRR.