Mitigating Galvanic Corrosion of Molybdenum Diffusion Barriers in Chemical Mechanical Planarization of Copper Interconnects: A Case Study Using Imidazole in a Citrate Slurry of Neutral pH

Abstract

Molybdenum (Mo) is currently considered as a potential diffusion barrier material for copper (Cu) interconnects, and these interconnect structures are generally processed using the technique of chemical mechanical planarization (CMP). While a limited number of publications on Mo CMP are presently available, the considerations for mitigating CMP-induced galvanic corrosion of Mo have remained largely underexplored. Using a model CMP system in pH-neutral slurries of citric acid with silica abrasives, the present work demonstrates how Mo barrier lines in contact with Cu wires in the CMP environment can develop CMP defects of galvanic corrosion. Including imidazole in the slurry considerably reduces the galvanic current of this corrosion process. The mechanisms of galvanic inhibition and material removal are examined by employing strategic tribo-electrochemical measurements. Open-circuit potential and potentiodynamic polarization measurements performed under surface abrasion aid the characterization of CMP-enabling surface reactions. The slurry’s surface chemistry initiates the primary modes of material wear for CMP, and corrosion-induced propagation of subsurface wear mostly governs the measured material removal rates for both Mo and Cu. Although the Cu:Mo selectivity of material removal is affected as the galvanic corrosion of Mo is suppressed, this effect can be controlled by varying the slurry content of imidazole.

1. Introduction

Ultrathin diffusion barriers based on molybdenum (Mo) have drawn considerable attention in recent years for potential implementation in the back end of line (BEOL) in copper interconnects [1,2]. The material properties that make Mo suitable for this application include the high melting point and strong thermal stability of Mo, coupled with its effective retention of electrical conductivity at nanoscale dimensions [3]. Due to these properties, Mo-based diffusion barriers can effectively block Cu diffusion, withstand high temperatures of thermal annealing, and support adequate electrical conduction to regulate RC delay in signal processing. Additionally, hybrid metallization using Mo is currently considered as a promising new alternative to the conventional dual-damascene scheme of Cu metallization [4,5,6].

Chemical mechanical planarization (CMP) currently continues to be a critical step of interconnect metallization [7,8] in general, and will be necessary to adequately integrate Mo in advanced interconnects. In this regard, further developments in the enabling technology of Mo planarization largely rely on the availability of fundamental knowledge about the relatively new Mo-based CMP systems. However, the mechanistic details of Mo CMP represent a rather underexplored area in the commonly available CMP literature [9,10,11,12,13]. Using a series of tribo-electrochemical measurements, our present investigation focuses on certain mechanistic aspects of Mo CMP, with a specific focus on the considerations for regulating CMP-induced galvanic corrosion defects.

In the CMP configuration of a patterned wafer, Mo and Cu regions of the wafer are simultaneously planarized as the excess Cu overlying the barrier line is cleared and the Mo barrier is planarized. Under this condition, the Mo-Cu contact region is in contact with a chemically active slurry, and forms a galvanic couple with Cu and Mo acting as the cathode and anode in the combination, respectively. As a result, the Mo regions in contact with Cu tend to galvanically corrode, and the resulting corrosion defects on the Mo liners can eventually lead to device failure. Our main focus in this work is on these galvanic effects associated with the Mo-Cu bimetallic CMP system.

An effective approach to controlling galvanic corrosions in CMP is to reduce galvanic activation potential by selectively modifying the cathodic and anodic corrosion characteristics of the coupled metals. This generally involves the inclusion of one or more metal-selective corrosion inhibitors in the CMP slurry to suppress the cathodic and anodic activities of the cathode metal and anode metals in the galvanic couple, respectively. Implementing this approach in a Cu-Mo bimetallic system can be challenging since most commonly known corrosion inhibitors of Mo, such as benzotriazole and other triazoles, tend to exhibit (nonselective) similar effects of corrosion suppression on both Mo and Cu surfaces [13,14,15]. The present study is centered on the detailed mechanistic aspects of slurry formulation to address this issue.

The experiments reported herein use a laboratory-scale model CMP system based on Mo and Cu disc samples. The electrochemical characteristics of these samples are examined with tribo-electrochemical measurements of the CMP-related corrosion variables. The corrosion parameters are determined using linear sweep voltammetry (LSV) and transient open-circuit potential (OCP) measurements, performed in the presence and in the absence of surface abrasion with a polishing pad and silica abrasives in CMP slurries of different trial compositions. To ensure that these slurries can also serve as ion-conducting electrolytes for electrochemical measurements, a background solution of potassium nitrate is included in all the test slurries.

The experimental CMP slurries employ sodium percarbonate (SPC, containing H₂O₂) and citric acid as an oxidizer and a complexing agent, respectively. The Na⁺ (and carbonate) ions released from the dissolved SPC also promote the slurries’ ionic conductivity without significantly interfering with the main CMP reactions [15]. The overall rates of general corrosion (monitored as corrosion current densities) for both Cu and Mo are controlled using pH-neutral slurries [16,17]. The theoretical considerations for galvanic corrosion are discussed, and the criteria for regulating Cu-activated galvanic corrosion of Mo in the test slurries are analyzed. Employing imidazole (HIm, C₃H₄N₂), a form of 1,3-diazole, as a selective cathodic inhibitor for Cu, we demonstrate how this inhibitor can mitigate Cu-induced galvanic corrosion of Mo under abrasion. The electrochemical mechanisms responsible for suppressing the galvanic corrosion of Mo in the Cu-Mo CMP system are investigated using a series of tribo-electrochemical measurements.

The experiments reported herein also examine the interplay between the mechanisms of corrosion inhibition and those of material removal. The different modes of material removal are checked by comparing the material etch rates (ERs) measured in the absence of surface abrasion, and by monitoring the material removal rates (MRRs) in the presence of surface abrasion using a commercial CMP pad and silica abrasives. These results are compared with the corrosion rates (CRs) and tribo-corrosion rates (TCRs) to investigate the relative roles of chemical corrosion, corrosion-induced wear (CW) and wear-induced corrosion (WC) of the CMP metals. The underlying chemical steps of these wear mechanisms are probed and the CMP-related practical implications of the findings are discussed.

The discussions of results in this report are organized as follows. The test samples, slurry consumables, instruments, experimental techniques and data analysis procedures are described in Section 2. In Section 3.1, we discuss the phenomenological criteria used to guide the chemical formulations of the test slurries. With a set of tribo-potentiodynamic data, we characterize in Section 3.2 the CMP reactions that are supported at the Mo and Cu sample surfaces in the absence of any corrosion suppressors. These data bring out the necessity for controlling Cu-induced galvanic corrosion of Mo in the citric acid-based slurries, and in Section 3.3, we outline the strategies necessary to regulate these galvanic effects.

The results presented in Section 3.4 demonstrate how the galvanic corrosion of Mo can be largely controlled by utilizing HIm’s cathodic suppression feature at Cu surfaces. The mechanism of HIm-supported galvanic corrosion is analyzed in detail in Section 3.5, and the parallel effects of HIm on the selectivity of material removal are examined in Section 3.6. The considerations for experimentally controlling the selectivity criteria for material removal are further discussed in Section 3.7. The observations presented in Section 3.6 regarding HIm’s role in galvanic suppression are quantified in Section 3.8 with tribology-coupled measurements of OCP transients. The mechanisms of material removal from Mo and Cu in the experimental CMP slurries are examined in Section 3.9. Collective conclusions of this investigation are presented in Section 4.

2. Materials and Methods

2.1. Metal Samples and Polishing Slurries for CMP

Material etch rates (ERs) were separately measured for the two test metals in the absence of surface abrasion using beaker cells containing the different slurry solutions and polycrystalline Cu (99.9%) or Mo (99.95%) coupon samples of square shape (20 mm × 20 mm) and 1.0 mm thickness (McMaster-Carr, Elmhurst, IL, USA). Presence or absence of silica abrasives in the solutions did not measurably affect these results. During these ER measurements, the slurries were magnetically stirred at 300 RPM and maintained at a temperature of 40 °C using a hot plate stirrer (Fisherbrand/Isotemp, Shanghai, China) to mimic a typical setting of CMP conditions for the pad temperature and fluid movement [18]. Chemical stability of each test slurry (usually dictated by the stability of H₂O₂ in the slurry) [19] was checked and confirmed by measuring each slurry’s pH value before and after subjecting the slurry to constant stirring for 60 min at 40 °C under the conditions of ER measurements. Detailed results of these pH measurements are presented in the Supplementary Section. A fresh batch of slurry was used for each experiment.

The tribo-electrochemical experiments employed Cu (99.99% pure) and Mo (99.95% pure) disc samples, each of 2.54 cm diameter (Kurt Lesker Co., Jefferson Hills, PA, USA) and embedded in Teflon holders. Before using in test cells, each metal sample was polished to a mirror finish using several sandpapers of progressively increasing grit sizes (1000, 1500, 2000, 2500, 3000 and 5000). Subsequently, each metal sample was polished on a locally assembled rotary polisher equipped with a Microcloth and using water based 1 μm diameter alumina powder (both from Buhler, Lake Bluff, IL, USA). Residues remaining on the sample from surface preparations were cleaned with 5 min of sonication with a BRANSON 2510 ultrasonic cleaner (Branson Ultrasonics Corporation, Danbury, CT, USA) in a bath of distilled water.

The main polishing slurry (Reference slurry, or Ref) contained 0.1 M KNO₃ (to support ionic conduction) from Fisher Scientific, Waltham, USA. An amount of 20 mM sodium percarbonate (SPC, Na₂CO₃·1.5H₂O₂, a solid carrier of H₂O₂) was used as an oxidizer. SPC was chosen due to its high chemical stability (shelf life), cost efficiency and its ability to promote slurry conductivity [20,21]. An amount of 0.1 M citric acid (CA) served as a complexing agent, and all slurries were pH-adjusted to a value of 7.0 using KOH. Due to their strong hydration, the Na⁺ ions released from SPC did not contact-adsorb onto the CMP metal surfaces [22], while the accompanying HCO₃⁻ ions in pH-neutral slurries remained largely noninteracting with the metals [23]. All chemicals were reagent grade; SPC, CA and KOH were obtained from Aldrich Chemical, Milwaukee, WI, USA.

The experimental slurries contained 3 wt% SiO₂ 125 K colloidal abrasives, as 85 nm average diameter particles at 40 wt% SiO₂ (from NYACOL Nano Technologies, Inc., Ashland, MA, USA). This colloidal solution contained very low levels of trace impurities with sodium (Na) (~23 ppm), magnesium (Mg, ~5 ppm) and calcium (Ca, ~5 ppm) being the only elements larger than 1 ppm, according to the manufacturer’s data sheet. The pH of the suspended solution ranged between 10.3 and 11.3. Owing to the isoelectric point (IEP) of silica (between pH 1.5 and pH 3.0), the surface charge of the abrasive particles remained negative, and facilitated electrostatic stabilization for all the slurries tested. Two of the test slurries employed HIm [(C₃H₄N₂), Sigma-Aldrich, Darmstadt, Germany] as a strategic (metal-selective) corrosion inhibitor for the Mo-Cu bimetallic system.

2.2. Measurements of Material Removal Rates and Tribo-Electrochemical Variables

Both ERs and MRRs were determined in the standard gravimetric approach as described elsewhere [24]. Reproducibility of these data was checked with repeated measurements using two separate samples in each case, and the standard deviations in the data were used as error bars in the data plots. Tribo-electrochemical experiments were carried out using a custom-built cell, integrated with a benchtop polisher. This cell has been described in considerable detail in our earlier publications [25,26], and certain essential features of the setup are briefly noted below.

The main components of the test cell included a Struers Labopol Benchtop polisher (Struers LC., Cleveland, OH, USA), combined with a Solartron 1287 potentiostat and a Frequency Response Analyzer 1252A (both from Ametek Scientific Instruments, Berwyn, PA, USA). The platen and the sample head were rotated or held stationary for different measurements. The polisher was equipped with a polishing head and a platen, both rotated at a common angular speed of 90 RPM (in polish stages, denoted as “P”) or zero RPM (in stationary hold stages, denoted as “H”). The main chamber of the cell served as a container of the slurry, and the slurry also served as the electrolyte to support electrochemical measurements. The metal test sample was pressed against the polishing pad at an applied down-pressure of 0.014 MPa in both the H and P configurations. Intermittent OCP polish–hold transients were recorded for 4 cycles with 4 min of polishing and 4 min of holding in each cycle.

An IC-1000 polishing pad (Rohm and Haas, Newark, NJ, USA) was attached to the top surface of the revolving platen. This pad is an industry-standard, rigid, microporous polyurethane pad designed for CMP, and is characterized by a hard, foam-like structure. Various detailed characteristics of this specific pad, including the considerations for pad–sample contact area [27], have been extensively discussed in the literature [28,29,30]. The CMP experiments were carried out at a down-pressure of 0.014 MPa (2 psi). A stainless-steel band placed on the inner wall of the rotating platen reservoir was employed as the counter electrode (CE). A saturated calomel electrode (SCE) was used as the reference electrode (RE) by incorporating a salt bridge to prevent seepage of KCl from the RE into the experimental slurries. The metal sample subjected to CMP was connected to the potentiostat as a (WE). All LSV experiments were performed with an optimized voltage scan rate of 5 mV s⁻¹, in the direction of increasing potentials as discussed previously [31].

Slurry (solution) resistances were measured using electrochemical impedance spectroscopy (EIS) following the procedures described elsewhere in detail [25]. In brief, EIS was carried out under stationary hold conditions with the sample surface lifted above the polishing pad by ~1 mm under stationary conditions. EIS measurements of the sample under a down-pressure were avoided for these experimental systems as the OCP (especially for Cu) then became strongly time-dependent and often the rate of this OCP variation became comparable to the lowest perturbation frequency of EIS. EIS data were collected using alternating current voltages of 10 mV average amplitude, distributed over 190 frequencies in logarithmically spaced intervals between 1 Hz and 10 kHz. These data were validated by using the Kramers–Kronig Transformation (KKT) method. All EIS data collected in this work are presented in the Supplementary Materials.

2.3. Data Processing Procedures

The recorded LSV data were corrected by subtracting the ohmic drops (IR_S) from the measured values of E and analyzed in the form of potentiodynamic polarization plots. Corrosion potentials (E_corr) and corrosion current densities (i_corr) of the CMP samples were determined by using the Tafel Extrapolation utility in Origin software (2024 version). Standard guidelines for Tafel extrapolation were followed, which included [32]: (1) Ensuring that at least one of the two current branches on each polarization plot exhibited a linear Tafel section over at least one decade of log (i); and (2) starting all Tafel extrapolations (tangents to polarization current branches) at least 50–100 mV above (for anodic branches) or below (for cathodic branches) E_corr. Both current branches on most of the polarization plots recorded in this work exhibited linear Tafel features (criterion 1). All Tafel extrapolations started at ≥70 mV away from E_corr on the voltage axis (criterion 2). The voltage ranges defining the Tafel segments of current branches were determined by implementing the above criteria for Tafel extrapolations using Origin software. These lower and upper potential bounds of the Tafel regions are listed in the Supplementary Materials.

All figures for this report were generated using Origin software. Electric equivalent circuit (EEC) model fits to the EIS data were obtained through complex nonlinear least square (CNLS) calculations using ZSimpWin software (version 3.50, AMETEK, Inc., Berwyn, PA, USA). The final circuit models obtained from CNLS analyses typically had <10% uncertainties in each of the circuit elements. KKT analyses were performed using ZView Software (version 3.5G from Scribner, LLC., Southern Pines, NC, USA, 2025). The solution resistances (R_s values) were obtained from the EEC models established by CNLS fitting the EIS data. The values of the CNLS-analyzed parameters and their associated uncertainties are reported in the Supplementary Materials. Detailed results of the KKT analyses are also presented in the Supplementary Materials.

3. Results and Discussion

3.1. Considerations for Slurry Selection and Measurement of Slurry Resistances

A general strategy for material removal in metal CMP with reduced roles of mechanical force is to form a structurally weak film of oxides/complexes at the CMP metal’s surface. Using Preston’s and Archard’s laws, the material removal rate (MRR) of CMP can be described as follows [33]:

$$\mathit{MRR} = k_{\mathrm{P}}Pv = {(k}_{\mathrm{w}}/H_{\mathrm{s}}) Pv$$

(1)

where k_p is the Preston coefficient; k_p = k_w/H_s, with k_w and H_s denoting the Archard wear coefficient, and the mechanical hardness of the chemically modified surface layer, respectively; $v$ is the relative velocity between the pad and the polished surface. The chemically lowered value of H_s allows the mechanical settings of P and V to remain at moderate values, which in turn help to minimize mechanically induced surface defects while maintaining adequate MRRs.

The conventional processing schemes used for Cu/barrier CMP include the following steps: (i) bulk Cu removal at relatively high rates, which typically stops with ~150 nm Cu left over the barrier lines; (ii) residual Cu removal (or, “Cu clearing”) at a slow rate, with MRRs about 1/5th or 1/6th of the MRRs used in Step (I); (iii) barrier (along with Cu and dielectric) planarization, which also is operated at low MRRs. The Cu-to-barrier material selectivity, S (Cu:Barrier) = [MRR(Cu)/MRR(Barrier)], in Step (i) is generally maintained at moderately high values, varying over a broad range, between 10:1 and 100:1 [34,35]. The MRRs for Cu and the barrier material often considered for Steps (ii) and (iii) remain within mutually comparable magnitudes [36,37]. While different polishing slurries (and different pads) have often been used for the three aforesaid CMP steps [38,39], slurries specifically designed to carry out both Steps (ii) and (iii) with combined slurry formulations have also been reported [15,24,40,41]. The CMP slurries used in the present work are in the latter category.

The main reasons for choosing neutral pH slurries in this study are as follows: (1) Acidic slurries are avoided since Mo tends to dissolve as [MoO(OH)₃(H₂O)₂]⁺ and Mo(OH)₄(H₂O)₂]²⁺ in low-pH media [42]. For Cu, neutral slurries help to reduce the dishing effects of CMP that are caused by excessive dissolution in acidic slurries [43]. (2) Alkaline slurries are avoided since they also promote dissolution of Mo due to enhanced production of MoO₄²⁻ [44], and introduces additional complexities in slurry formulations [15]. (3) Controlling the Cu:oxide CMP selectivity can often present certain challenges, in alkaline slurries, and the efficiencies of many commonly used corrosion inhibitors decrease in such slurries [45]. Moreover, H₂O₂, a standard oxidizer for metal CMP, becomes prone to chemical decomposition in alkaline solutions [46].

To suppress the effects of the galvanic corrosion of Mo in the Cu-Mo CMP system, we have used imidazole (HIm), which is a broadly known corrosion inhibitor for metals (and alloys) [47]. HIm is an amphiprotic species that can serve both as an acid (H⁺ donor) and a base (H⁺ acceptor), with a pKa of 14.5. The pK_BH⁺ value (dissociation constant for the conjugate acid of the weak base) of imidazole is 7.0 [48]. Thus, the neutral pH CMP slurry contains equal amounts of HIm and its protonated form, imidazolium (HImH⁺) (Figure 1). As a corrosion suppressor for Cu, HIm often works as a mixed inhibitor (to suppress both anodic and cathodic corrosions) [47,49], but in some cases, HIm has been found to preferentially act as a cathodic inhibitor of Cu [50]. As explained later in Section 3.5, the selection of HIm as a galvanic corrosion inhibitor in the present work is in part based on its latter function. The pH-dependent protonation and deprotonation characteristics of CA and HIm have been discussed previously, and for ready reference, calculated distribution plots of these species are shown in Figure 1. The slurry resistances obtained using EIS are listed in

Table 1. Chemical compositions and solution resistances of experimental slurries.

Slurry	Composition	Rs (Ω cm2) (Mo, Cu)
I	0.1 M KNO3 + 20 mM SPC + 0.1 M CA + 3 wt% SiO2 (Ref)	26.0, 19.0
II	Ref + 10 mM HIm	32.4, 24.8
III	Ref + 20 mM HIm	63.7, 42.3

, along with the chemical compositions of the three slurries, I, II and III, used in this work.

The R_s values in Table 1 increase for both metals with increasing HIm concentrations in the slurries. This suggests that the inhibitor molecules adsorbed at each metal surface form an interfacial “ohmic” layer, the resistance of which adds in series with the resistance of the bulk slurry. The mechanistic aspects of this process have been discussed in detail in our previous work [15]. Aside from R_s, the other model parameters of CNLS calculations (all obtained in the Sample-Up configuration under stationary hold conditions) do not have major direct implications in the central context of the present CMP-specific results. Therefore, these additional EIS parameters are not discussed in the main article; instead, a brief discussion of these variables, focusing on their overall association with the CMP-related results, is presented in the Supplementary Materials.

3.2. CMP-Enabling Surface Reactions for Cu and Mo Using Citric Acid at Neutral pH Without Corrosion Inhibitors

In slurry I, the MRRs measured at 0.014 MPa down-pressure for the Mo and Cu samples were 30.5 and 88.0 nm min⁻¹, respectively. As expected on the basis of grain size effects, these MRRs were relatively lower than those generally found for thin-film wafer samples. The corresponding ERs were measured as 9.8 and 49.9 nm min⁻¹, for Mo and Cu, respectively. The ERs were supported by the formation of soluble oxide/complex species of the metals, while the MRRs included contributions of both ERs and the polish rates of insoluble surface oxides/complexes supported on the metal surfaces under abrasion.

The CMP-related soluble and insoluble metal oxides and complexes are generated via chemical and electrochemical corrosion-like reactions. The faradaic features of these CMP reactions are commonly examined using LSV-generated polarization (Tafel) plots. The corrosion variables obtained from such Tafel data, combined with the trends of these variables with respect to changes in slurry compositions, provide an adequate basis for assessing the CMP-specific (electro)chemical characteristics of trial slurry formulations. These considerations for the Ref. slurry I are explored using the potentiodynamic plots shown in Figure 2. The values of E_corr for Mo and Cu, measured with and without surface abrasion, are indicated in the Figure.

The cathodic (lower) branches of the Tafel plots in Figure 2 are dominated by the reduction currents of H₂O₂. At pH =7, this reduction reaction on both Mo and Cu occurs as follows [51,52]:

H₂O₂ + 2H⁺ + 2e⁻ = 2H₂O

(2)

where the H₂O produced at the CMP interface plays an active role in supporting the oxidizing function of H₂O₂. While reaction (2) dominates the cathodic activities of CMP metals in H₂O₂-based slurries, a parallel component of oxygen reduction reaction (ORR) generally operates due to O₂ dissolved in the slurries. At neutral (as well as alkaline) pH, the ORR occurs as follows [53]:

O₂ + 2H₂O + 4e⁻ = 4OH⁻

(3)

which provides a relatively weaker contribution to the metal sample’s cathodic current, superimposed over the stronger current of H₂O₂ reduction.

The oxidation reactions supporting the anodic (upper) current branches of the Tafel plots in the present study include

Mo + 2H₂O = MoO₂ + 4H⁺ + 4e⁻

(4)

MoO₂ + H₂O = MoO₃ + 2H⁺ + 2e⁻

(5)

while some of the MoO₂ generated in reaction (4) is chemically converted to dihydroxy (oxo) molybdenum, MoO(OH)₂ [54]. The latter reaction occurs as follows:

MoO₂ + H₂O = MoO(OH)₂

(6)

and this MoO(OH)₂ can serve as a removable material under polishing in CMP. While reactions (4) and (5) directly contribute to the anodic Tafel branches of the Mo sample, the chemical step in Equation (6) indirectly affects both Tafel branches of Mo by altering the surface coverages of MoO₂.

The anodic charge transfer steps in Equations (4) and (5) couple with cathodic reaction (2), and occur in the following mixed potential forms of Equations (7) and (8), respectively:

Mo + 2H₂O₂ = MoO₂ + 2H₂O

(7)

MoO₂ + H₂O₂ = MoO₃ + H₂O

(8)

while some MoO₃ surface sites weakly dissolve as HMoO₄⁻ [15,44,54]:

MoO₃ + H₂O = HMoO₄⁻ + H⁺

(9)

MoO₃ + OH⁻ = HMoO₄⁻

(10)

and contribute to the measured ERs.

The OH⁻ generated through the ORR in Equation (3) creates a favorable condition for additional dissolution of MoO₃ in the form of MoO₄²⁻:

MoO₃ + 2OH⁻ = MoO₄²⁻ + H₂O

(11)

which also contributes to the measured ERs for Mo.

In Mo CMP, a complexing agent is generally used to form a structurally weak insoluble film at the MoO₃-dominated Mo surface for low-pressure polishing, and also to regulate the dissolution reactions (10) and (11) [15]. With citric acid employed as a complexing agent, MoH₋₁Cit(OH)₂ is the expected predominant form of the Mo–citrate complex at pH = 7 [55], which can form according to:

MoO(OH)₂ + Cit³⁻ + H⁺ = MoH₋₁Cit(OH)₂ + H₂O + 2e⁻

(12)

MoO₃ + Cit³⁻ + 3H⁺ = MoH₋₁Cit(OH)₂ + H₂O

(13)

and in addition to MoO(OH)₂, this citrate complex becomes another component of the abradable film of Mo in CMP. Reaction (12) contributes to the currents in the anodic branches of the Tafel plots in Figure 2.

The surface oxidation reactions of Cu in H₂O₂-containing neutral pH slurries occur as follows [56,57]:

2Cu + H₂O = Cu₂O + 2H⁺ + 2e⁻

(14)

Cu₂O + H₂O = 2CuO + 2H⁺ + 2e⁻

(15)

CuO + H₂O = Cu(OH)₂

(16)

where the surface-bound H₂O reactant mostly comes as a product of reaction (2) in H₂O₂-containing slurries. The mixed versions of reactions (14) and (15) in combination with reaction (2) are 2Cu + H₂O₂ = Cu₂O + H₂O and Cu₂O + H₂O₂ = 2CuO + H₂O, respectively.

In the CA-based slurries, Cu₂O deposited at the Cu CMP surface reacts with Cit³⁻ to form a soluble Cu–citrate complex, (Cu₂Cit₂H₋₂)⁴⁻,

Cu₂O + 2Cit³⁻ = (Cu₂Cit₂H₋₂)⁴⁻ + H₂O + 2e⁻

(17)

which is stable at neutral pH [58], and contributes to the ER (Cu). As this species forms within porous films of Cu₂O, the oxide surface films of Cu become mechanically fragile and easy to remove by polishing in CMP. Reactions (14), (15) and (17) contribute to the anodic Tafel currents of the Cu sample, while the chemical reaction (16) indirectly affects both Tafel branches.

Excessive ERs of a Mo surface can adversely affect the planarization efficiency of CMP and lead to barrier losses [13,59]. The published values of ERs in Mo CMP are frequently found between 10 and 40 nm min⁻¹ [9,10,11]; the ER of 9.8 nm min⁻¹ measured here in slurry I is at the lower end of this range. The CMP-specific potentiodynamic polarization features of Mo and Cu observed in slurry I are comparable to the corresponding features of these metals previously reported [9,10,11,60,61,62,63].

Based on the above observations, the general electrochemical features of the CMP system considered in Figure 2 appear to be compatible with those commonly found in the current literature of Mo CMP. However, E_corr (Cu) measured in slurry I is notably more anodic than E_corr (Mo) in both the hold (H) and polish (P) configurations. This implies that Cu would act as a cathode (C) metal to induce galvanic corrosion in the anode (A) metal Mo at the Mo-Cu contact regions. The phenomenological considerations for controlling these galvanic effects are discussed in the next section.

3.3. Considerations for Regulating Galvanic Corrosion of Mo in Cu-Mo CMP

The rate of galvanic corrosion of an anode metal in a bimetallic combination is generally estimated in terms of the galvanic current density, $i_g\left(\mathrm{A}\right),$ at the anode’s surface [64,65]:

$$i_g\left(\mathrm{A}\right) = \exp \left({\displaystyle \frac{{∆E}_{\mathrm{corr}}}{{\beta }_{\mathrm{ca}}}}\right){\left[i_{\mathrm{corr}}\left(\mathrm{C}\right){\displaystyle \frac{S_{\mathrm{c}}}{S_a}}\right] }^{{\displaystyle \frac{{\beta }_{\mathrm{c}}}{{\beta }_{\mathrm{ca}}}}}{\left[i_{\mathrm{corr}}\left(\mathrm{A}\right)\right] }^{{\displaystyle \frac{{\beta }_{\mathrm{a}}}{{\beta }_{\mathrm{ca}}}}}$$

(18)

where ${∆E}_{\mathrm{corr}} = E_{\mathrm{corr}}\left(\mathrm{C}\right) - E_{\mathrm{corr}} \left(\mathrm{A}\right); {∆E}_{\mathrm{corr}}$ is the driving potential for galvanic corrosion; $E_{\mathrm{corr}}\left(\mathrm{C}\right)$ and $E_{\mathrm{corr}}\left(\mathrm{A}\right)$ are the corrosion potentials of the cathode and anode metals, respectively. β_ca = β_c(C) +β_a(A) with β_c(C) and β_a(C) denoting the anodic and cathodic Tafel parameters (Tafel slopes, divided by 2.303) for the cathode metal, respectively. β_a (C) $= 1/[f\left\{1-{\alpha }_a\left(\mathrm{C}\right)\right\}$], β_c (C) = 1/fα_c(C), ${\alpha }_a\left(\mathrm{C}\right)$ and ${\alpha }_c\left(\mathrm{C}\right)$ are the anodic and cathodic transfer coefficients for the cathode metal, respectively; $f$= nF/RT, n is the reaction valance, F and T denote the Faraday constant and the sample temperature, respectively; R is the gas constant. $S_{\mathrm{c}}$ and $S_{\mathrm{a}}$ represent the effective surface areas of the cathode and anode metals, respectively. β_a (A) and β_c (A), for the anode metal, are defined using the above formulas for β_a(C) and β_c(C) with ${\alpha }_a$(C) and ${\alpha }_c$(C) replaced by the corresponding transfer coefficients, ${\alpha }_a$(A) and ${\alpha }_c$(A), for the anode metal. $i_{\mathrm{corr}}\left(C\right)$ and $i_{\mathrm{corr}}\left(A\right)$ are the corrosion current densities at the cathode and anode regions of the galvanic contact, respectively.

Among the variables included on the right-hand side of Equation (18), the term ${∆E}_{\mathrm{corr}}$ is most relevant for controlling galvanic corrosion since the value of $i_g\left(\mathrm{A}\right)$ exponentially varies with changes in the value of ${∆E}_{\mathrm{corr}}$. In general, acceptable values of ${∆E}_{\mathrm{corr}}$ for regulating CMP-induced galvanic corrosion should remain below the thermal voltage, which typically is around 30 mV depending on the pad temperature during CMP [65]. However, the values of ${∆E}_{\mathrm{corr}}$ for the Mo-Cu CMP system considered in Figure 2 are considerably higher than this threshold potential, and without additional adjustments of slurry compositions, they would tend to induce galvanic corrosion of the Mo liners in contact with Cu wirings. Figure 3 schematically depicts (not to scale) this situation.

Figure 3A shows a relevant section of the CMP interface at the stage of barrier planarization in an ideal case where galvanic corrosion of the barrier material (Mo in this case) is absent. The Cu wiring and the barrier line (Mo) are labeled. The polishing pad (not included in the figure) is located below the CMP interface and simultaneously abrades surface sectors of both Cu and Mo exposed to the slurry [66]. Figure 3B corresponds to a case where galvanic corrosion of the anode metal Mo is activated at the Mo-Cu contact regions exposed to the slurry. The oxidizer, H₂O₂, from the polishing slurry adsorbs onto the Cu surface and reacts with H⁺ to be cathodically reduced as H₂O. The electrons needed for this reduction are supplied by a simultaneously activated anodic galvanic corrosion reaction of Mo.

Based on previously published reaction schemes [54,67], galvanically corroded Mo in test slurry I is expected to anodically dissolve in the form of HMoO₄⁻ as follows: Mo + 4H₂O = 7H⁺ + 6e⁻ + HMoO₄⁻. In combination with the cathodic reduction of H₂O₂ at the Cu surface, this galvanic corrosion of Mo occurs in the mixed potential form:

Mo + 3H₂O₂ = HMoO₄⁻ + H⁺ + 2H₂O

(19)

which results in galvanic defects on the barrier line, as shown in Figure 3B. Since H₂O₂ is the main solution-phase reactant in Equation (19), this galvanic corrosion is likely to spread with increasing concentrations of [H₂O₂] in the slurry.

A commonly practiced experimental approach to controlling the galvanic corrosion of metals in chemical solutions is to suppress the corresponding activation potential, ΔE_corr, while simultaneously regulating the values of $i_{\mathrm{corr}}\left(C\right)$s and $i_{\mathrm{corr}}\left(A\right)$. These effects can be implemented by using metal-selective corrosion inhibitors, namely a cathodic corrosion inhibitor for metal C and/or an anodic inhibitor for metal A in the solution. The phenomenological framework of this approach is based on the description in Equation (18), along with the following mixed potential formulation of $E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$ [68]:

$$E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}={\displaystyle \frac{RT}{nF}} ln \left({\displaystyle \frac{{\sum }_{k=1}^{c\ast }{[\theta }_{ck}{i}_{0k}^{\left(c\right)}\exp \left(E_{rk}^{\left(c\right)}/{\beta }_c\right)]}{{\sum }_{l=1}^{a\ast }[{\theta }_{al}{i}_{0l}^{\left(a\right)}\exp [-E_{rl}^{\left(a\right)}/{\beta }_a]}}\right)$$

(20)

where the mixed system is composed of c* cathodic reactions and a* anodic reactions. θ_ck and θ_al are fractional surface coverages, respectively, of the surface sites where the k^th cathodic reaction and the l^th anodic reaction occur; the corresponding exchange current densities of these reactions are denoted as i_0k and i_0l. $E_{rk}$ and $E_{rl}$ are the Nernst potentials of the k^th cathodic step and l^th anodic step, respectively. For simple illustration, we have assumed a common charge transfer valence (n) and a common transfer coefficient (α) for the individual reactions. The above definition of $E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$ applies to both the anode (A) and the cathode (C) metals of a galvanic couple.

If a cathodic corrosion inhibitor of the k^th reaction on metal C selectively suppresses the value of ${\theta }_{c\mathrm{k}}$, then the value of $E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}} \left(\mathrm{C}\right)$ according to Equation (20) would decrease. Likewise, an anodic inhibitor of the l^th reaction on metal A can selectively suppress the value of ${\theta }_{\mathrm{a}\mathrm{l}}$ thereby causing an upward shift in $E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$(A). Either one or both these effects can reduce the galvanic activation potential, ΔE_corr, for the bimetallic system [26,69]. Ideally, the slurry formulations based on these considerations can be simplified if a single corrosion inhibitor can be employed as a cathodic suppressor for metal C and an anodic suppressor for metal A. The resulting shifts in the values of $E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$(C) and $E_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$(A) should be accompanied by measurable reductions in the corrosion current densities, $i_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$(C) and $i_{\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}}$(A), of the cathode and anode metals, respectively.

Although various corrosion inhibitors for Cu have been extensively discussed in the literature [70,71,72], corrosion inhibition for Mo is still a relatively underexplored topic, especially in the CMP literature [6,11,13,73]. Certain amino acids have been employed as corrosion inhibitors in Mo-CMP, but many of these chemicals tend to act as complexing agents for Cu and promote (instead of suppressing) wear-generating surface reactions in Cu-CMP [11,74,75]. Any utility of such chemicals as selective corrosion suppressors for the Mo-Cu galvanic couple remains to be tested. The considerations of metal-selective corrosion inhibition for CMP applications are additionally nontrivial since most available results of corrosion inhibition for bimetallic systems are based on traditional studies of stationary surfaces in the absence of abrasion and hence do not generally reflect the tribological effects of CMP.

A further reason why controlling galvanic corrosion is significantly more challenging than controlling general corrosion in CMP is that the former involves complex and interfering tribo-chemistries of two separate metal surfaces under abrasion in the slurry, while the second case deals with a single metal–slurry interface. The difference between galvanic and general corrosions can be underscored also by noting that the number of variables that dictate the values of i_g [in Equation (18)] is considerably greater than those found in the mixed potential formulation of i_corr [Equation (S4) in the Supplementary Materials]. Due to these reasons, electrochemical manifestation of CMP-related galvanic corrosion is highly sensitive to the details of slurry compositions, including the slurry pH, complexing agent, oxidizer, and corrosion suppressor [76,77,78,79].

For the reasons noted above, most of the commonly reported slurry formulations that aim at controlling galvanic corrosion in CMP tend to be rather system-specific for a given bimetallic combination and operate within a relatively limited window of variability for the slurry components. To demonstrate this highly system-specific nature of galvanic corrosion in CMP, we have presented illustrative results of a series of trial experiments conducted in our laboratory involving two different bimetallic CMP systems (Mo-Cu and Co-Cu) using a range of slurry compositions based on different complexing agents, without and with the inclusion of two different corrosion inhibitors [benzotriazole (BTA) and HIm].

Metal-selective corrosion inhibitors for regulating galvanic corrosion in CMP can be identified through trial experiments performed under polishing conditions, and our present work is based on this method. A frequently used approach to including variability in the selection of CMP slurries for such studies is to vary the concentration of a specific corrosion inhibitor (identified through trial measurements to be appropriate for a given galvanic couple), while maintaining the remaining slurry compositions mostly unchanged [69,80,81]. This approach allows one to ensure that the corrosion inhibitor’s concentration can be used as a primary control variable to maintain an adequate balance between galvanic corrosion suppression and the (rate as well as) selectivity of material removal. Our present work also utilizes this tactic and uses three different concentrations (one being the zero concentration) of HIm in the test slurries.

3.4. Utility of Imidazole for Regulating Galvanic Corrosion of Mo in Cu-Mo CMP

To check the effects of HIm on the Mo-Cu CMP system, the polarization experiments performed in slurry I (Figure 2) were repeated in slurries II and III with two different HIm concentrations. The results of the latter measurements using both the hold and polish sample configurations are shown in Figure 4. The corrosion variables determined from the potentiodynamic data in Figure 2 and Figure 4 are organized in Figure 5 (showing E_corr, E_g and i_g) and Figure 6 (showing i_corr). In Figure 4, the corrosion potentials for the Mo and Cu samples under polishing move closer to each other’s values in the HIm-containing slurries. Thus, the inclusion of HIm in the CMP slurry exhibits its intended function to reduce the value of ${∆E}_{\mathrm{corr}}$ for the purpose of suppressing the galvanic corrosion of Mo.

While polishing in slurry III, E_corr (Cu) drops slightly below E_corr (Mo), indicating a relatively weak reversal of galvanic polarities in this case. Such reversals of galvanic polarity are frequently observed in potentiodynamic studies of bimetallic couples [69,82,83], and this effect is generally dictated by specific details of the experimental conditions. In the present case, the galvanic reversal of E_corr in slurry III can be attributed to potential-induced adsorptions of slurry species, an effect that is also responsible for shifting LSV-measured E_corr values away from the corresponding open-circuit potentials (OCPs, E_OC) [65]. Nevertheless, even after polarity inversion in slurry III, |ΔE_corr| (27 mV) remains within its commonly accepted cut-off value, which, for metal CMP, is defined as the thermal voltage established at the sample surface during polishing at an average pad temperature around 40 °C [25].

On the Mo-Cu combined polarization plots in Figure 4, the galvanic current density (i_g) and galvanic potential (E_g) correspond to the current and voltage coordinates of the intersection points, respectively, where the anodic current of the Mo anode crosses the cathodic current of the Cu cathode. The vertical dashed lines mark the locations of these points (a similar approach was used to determine E_g and i_g for slurry I using Figure 2). The Mo-Cu galvanic potentials (E_g) measured in the individual slurries are compared with the corrosion potentials plotted in Figure 5A,B. The galvanic current densities (i_g) at the Mo surface, as obtained from the data in Figure 4 (and Figure 2) by assuming S_a = S_c, are plotted in Figure 5C. As expected in terms of the mixed potential theory of galvanic corrosion, E_corr (A) ≤ E_g ≤ E_corr (C) for the cases without galvanic polarity inversion in Figure 4 [64]. A preferential proximity of E_g to its associated E_corr (A) or E_corr (C) defines the anodic vs. cathodic control characteristics of a given galvanic couple [84]. According to this characterization scheme, the galvanic corrosion of Mo in most of the cases in Figure 5A,B is under mixed control, with E_g nearly centered between E_corr (Cu) and E_corr (Mo). In the polishing case involving slurry III, the galvanic corrosion is in the anode-controlled mode.

The corrosion currents measured for both the Mo and Cu samples under stationary hold (Figure 6A) follow essentially the same slurry-dependent trends observed under polishing in Figure 6B. However, these currents are noticeably weaker in the absence of surface abrasion, which indicates that both sample surfaces in the absence of polishing are partly passivated by oxide/complex species of the individual metals. These passivating species are mechanically removed during polishing, resulting in observable increases in the metals’ surface activities. The instantaneous surface coverages of these species are determined by the mutually competing rates of their formation and removal. Since dissolution is the only mechanism of material removal in the stationary hold case, a CMP sample’s electrochemical signal recorded in this case is dominated by low ER values of the passivating surface species accumulated at the unabraded surfaces. The surface coverages of the oxide/complex species generally are significantly lower (ideally zero) under polishing conditions when the surface layers are mechanically removed.

The data for the polish case in Figure 5C show how the galvanic currents induced by Cu at a Mo anode can be suppressed in the polishing situation of CMP using HIm. A major reason why this is possible is that HIm acts as a cathodic inhibitor for Cu and shifts E_corr (Cu) in the cathodic direction. The latter effect can be seen in Figure 5B where the values of E_corr (Cu; polish) decrease with increasing [HIm] in the CMP slurry. The values of i_corr (Cu; polish) in Figure 6B also follow this same decreasing trend of E_corr (Cu; polish) and thereby confirm that the predominant effect of surface modifications for Cu by HIm in the polishing case is that of cathodic inhibition, and not of anodic promotion (since anodic promotion would have increased i_corr while decreasing E_corr). The results for i_corr (Cu; hold) depicted in Figure 6A follow the same trend of i_corr (Cu; polish). This implies that the Cu sample maintains the surface conditions of cathodic inhibition by HIm in both polish and hold situations.

Since the polishing experiments in this work are designed to mimic the basic mechanical arrangement of CMP, the i_g (polish) data are more relevant than those of i_g (hold) to evaluate the performance of HIm as a suppressor of CMP-related galvanic corrosion. Nevertheless, comparing these i_g (polish) results with published galvanic current densities for other CMP systems is difficult since many of the reported i_g values are based on measurements performed without implementing mechanical surface abrasion. Among the few presently available reports of tribo-electrochemically measured galvanic currents, He et al.’s experiments involving a Co-Cu CMP system yielded i_g (polish) for Co corrosion in a range between 22 and 163 μA cm⁻² [80]. Tamilmani et al. have measured i_g (polish) in the 15–100 μA cm⁻² range for the galvanic corrosion of Ta in a Ta-Cu CMP system [76]. The value of 21 μA cm⁻² for i_g (polish) measured in this work using slurry III is in an appropriate range of the above results.

It is useful to note in the present context that galvanic corrosion currents electrochemically measured in closed-volume electrolytes (slurries) often tend to decay over extended intervals. This temporal decay has been observed for CMP-related bimetallic combinations both in the absence [69,85] and in the presence [80,86] of surface abrasion. While steady growths of reactive surface films play a role in governing this effect in the stationary hold cases, depletion of chemicals (especially H₂O₂) responsible for activating galvanic corrosion reactions can be a predominant natural mechanism for decaying galvanic currents over time in the presence of surface abrasion. The galvanic currents reported in this study were measured using freshly prepared samples, and therefore, were mostly free of prolonged time-dependency.

3.5. Mechanisms of Imidazole-Suppressed General Corrosion and Galvanic Corrosion

To discuss detailed mechanisms of corrosion inhibition by imidazole for the Mo-Cu CMP system, it is necessary to examine if the two species, HIm and HIm⁺, of the inhibitor are associated with any preferential adsorption criteria at the two metal surfaces. The excess charges located at the surfaces play an important role in determining these criteria. A metal CMP surface generally contains both bare metal sites and oxide/complex sites of the metal. The net excess charge of such an “amphi-functional”’ surface is determined by the charges appearing at the oxide-free metal regions (dictated by the metal’s potential of zero charge, PZC) and those at the oxide-covered regions (dictated by the isoelectric point, IEP, of the predominant oxide) [87,88].

The reported PZCs of Mo and Cu are −0.54 V [89] and −0.95 V [90], respectively, and both these potentials are more negative than the E_corr values of the corresponding metals in the test slurries used (Figure 5A,B). Thus, the bare metal sites of both Mo and Cu contain excess positive charges in the CMP situation, and due to electrostatic repulsion, adsorption of HImH⁺ should be relatively ineffective at these metal sites. The IEPs of both CuO and Cu₂O are between pH values of 8.0 and 10.8 [91,92]. As a result, these oxide surfaces are also positively charged at the test slurry’s neutral pH, and do not provide electrostatically favorable adsorption sites for HImH⁺. Thus, the effects of corrosion inhibition observed for the Cu surface in this work can be largely associated with those of HIm, and not of HImH⁺.

The reported IEPs of MoO₂ are found between pH 4 and pH 5 [93], and those of MoO₃ are between pH 1 and pH 1.5 [94,95]. These surface oxides of Mo contain excess negative charges in the pH-neutral slurries, and serve as electrostatically favorable adsorption sites for HImH⁺. Therefore, unlike the case of Cu, both the charged and uncharged species of the corrosion inhibitor are expected to interact with the partially oxide-covered Mo CMP surface.

The observed function of HIm as a cathodic inhibitor for Cu can be explained in terms of imidazole’s adsorption characteristics on Cu surfaces. HIm adsorbs via its pyridine-type nitrogen atom (which has a lone pair of electrons in an sp² orbital) from position 3 to form a strong N-Cu bond [71,96]. This results in a vertical adsorption configuration of HIm, which, in turn, promotes interactions of the adsorbed molecule with water in the double layer. Through these interactions, HIm competes with water molecules for the adsorption sites of Cu and thereby limits the surface coverage of water on Cu [97]. The adsorbed HIm on Cu disrupts the continuous hydrogen bond network of the interfacial water, and since this network is essential for transporting solution-phase H⁺ ions to the Cu surface, the availability of H⁺ to support the cathodic step of reaction (2) becomes limited in the presence of HIm adsorption on Cu [97]. The HIm adsorbates on Cu also directly block the adsorption sites of H⁺ [50]. With the surface coverages of H⁺ and H₂O [reactant and product of reaction (2), respectively] being restricted in this way, the cathodic activity of Cu for H₂O₂ reduction is suppressed by adsorbed HIm species.

The effects of HIm on the corrosion variables of Mo in Figure 5 and Figure 6 are restrained compared to the corresponding effects observed for Cu. The value of E_corr (Mo; polish) in Figure 5B increases with the addition of 10 mM HIm to slurry I, and shows a slight decrease as the [HIm] is doubled in slurry III. For E_corr (Mo; hold) in Figure 5A, no specific trends can be established with variations in [HIm] in the slurries. The effects of HIm on the corrosion current densities, (A) i_corr (Mo; hold) and (B) i_corr (Mo; polish), in Figure 6 are rather minimal, while a slight increasing trend in the corrosion currents of Mo can be detected with increasing [HIm]. This behavior of i_corr (Mo), in combination with that of E_corr (Mo) in Figure 5B, is indicative of a slight apparent increase in the Mo sample’s anodic activity in the increased presence of HIm/HImH⁺. However, as we demonstrate later in the detailed discussions of the MRR and OCP data, the main effect of HIm/HImH⁺ adsorptions at the Mo CMP surface under open-circuit conditions is manifested as that of an anodic corrosion inhibitor.

Based on the above observations, it is evident that voltage-induced effects of LSV measurements complicate the potentiodynamic features of the Mo CMP surface. In the context of corrosion inhibition, this type of effects can be activated if a corrosion inhibitor adsorbs onto its substrate metal via a faradaic charge transfer step and if this step is enhanced by the external voltage scan of LSV used to collect potentiodynamic data. While a majority of corrosion inhibitors exhibit non-faradaic adsorptions at metal surfaces, in certain cases, inhibitors participate in faradic reactions with their substrates to form relatively passive layers of surface complexes. A typical example of such faradaically supported inhibitor adsorption is that of benzotriazole (BTAH) at a Cu surface to form a Cu-BTA complex through an anodic charge transfer reaction [98]. Another known example of inhibitor adsorption via charge transfer is the formation of a Co-BTA complex, [Co(BTA)₂.H₂O]_n, due to anodic adsorption of BTAH at Co surfaces [99].

A signature feature of faradaically adsorbed corrosion inhibitors at metal surfaces is that the correspondingly measured values of i_corr do not follow their usual decreasing trend with increasing inhibitor concentrations, and often change in an opposite direction of the expected trend. For instance, an experimental system matching the description of the latter case can be noted in a recent report by AbdElRhiem et al. [100], where corrosion currents were measured at a Cu–alloy surface in a corrosive solution with increasing concentrations of BTA, used as an inhibitor. The measured values of i_corr in these authors’ study increased with increasing [BTA], and a similar pattern is exhibited here in Figure 6A by the i_corr (Mo; hold) data. As we discuss below, the present observations made from Figure 6A can be explained in terms of anodically activated adsorption of the HImH⁺ inhibitor species onto Mo–oxide sites.

HImH⁺ cations from the pH-neutral CMP slurry can adsorb at the negatively charged MoO₂ and MoO₃ sites of the Mo CMP surface via coulombic interactions. While this adsorption process likely remains non-faradaic under the natural OCP conditions of CMP, the adsorbates can anodically form Mo–imidazole peroxo complexes due to the voltage activation of positive LSV scans used to collect potentiodynamic plots. For instance, the anodic reaction proposed below [101]

MoO₃ + HImH⁺ + 3H₂O = MoO₄(HIm).2H₂O + 3H⁺ + 2e⁻

(21)

can generate a MoO₄(HIm) complex at the surface that would eventually be mechanically removed during polishing. The anodic current of reaction (21) would contribute to the net corrosion current monitored from the Mo–slurry interface. At the same time, the passive sites of MoO₄(HIm) would block other faradic reactions [such as reaction (12)] in addition to also blocking chemical reactions [like reaction (13)] that would otherwise occur at the MoO₃ sites. These processes would interfere with and complicate the normal potentiodynamic response of the CMP surface, especially in the stationary hold setting where the unabraded Mo surface becomes heavily covered by oxides and complexes. The i_corr (Mo; polish) data in Figure 6B also contain some features of this effect.

The mechanism of anodically supported corrosion inhibition proposed in Equation (21) explains the observations in Figure 5 and Figure 6 where the slurry-dependent data trends for Mo tend to deviate from those observed in the polishing case. While this mechanism is mostly tentative in the absence of additional experimental evidence, the effects of such reactions in the context of corrosion inhibition can be readily accounted for within the framework of mixed potential theory. A brief discussion based on simple considerations of mixed potential formulation is included in the Supplementary Materials to demonstrate how the variations in the corrosion variables with changes in inhibitor concentrations can be convoluted in the presence of faradaically active corrosion inhibitors. In situations like this, strategically designed OCP experiments (discussed later in the context of the OCP data) would be a more suitable method than potentiodynamic polarization measurements to analyze the electrochemical signatures of corrosion inhibition [102].

3.6. Effects of Imidazole on Material Removal and CMP Selectivity

Most corrosion inhibitors commonly used in metal CMP suppress the substrate metal’s MRRs [103,104]. Therefore, while selecting corrosion inhibitors to regulate the corrosion features of metal CMP, it is necessary to check how the rates and selectivity of material removal are affected by the inhibitors. The results of these tests for the present systems are presented in Figure 7. Panels (A) and (B) in Figure 7 show the slurry-dependent values of ERs and MRRs, respectively, for the two metals tested. The relatively short error bars (standard deviations) associated with the plotted data confirm adequate repeatability of these measurements. For both metals, the ERs and MRRs decrease with increasing [HIm] in the CMP slurries. Thus, although HIm exhibits a preferential function as a corrosion inhibitor for Cu, this surface modifier suppresses material removal from both Mo and Cu.

Figure 7C shows the inhibition efficiencies (IEs) of HIm in slurries II and III for controlling the ERs and MRRs of Mo and Cu. Here, IE [for ER (or MRR)] = ([ER (or MRR) − ER′ (or MRR′)] × 100%)/ER (or MRR), with the unprimed and primed variables denoting their values in the HIm-free and HIm-containing slurries, respectively. The MRR in each case represents a combination of the (dynamic) ERs and mechanical polish rates. For Mo, the IE (ER) values are larger than their IE (MRR) counterparts. This situation is preferred since the polish rates can be effectively controlled by adjusting the polisher’s mechanical controls, while the ERs are often dominated by chemical dissolution steps that are difficult to control without substantially changing the slurry compositions. This control of polish rates over dissolution rates is necessary especially in the CMP of ultrathin barrier lines where the dimensional margin for defect regulation is severely restricted.

Figure 7. (A) Displays the etch rates (ERs) and (B) shows the material removal rates (MRRs) of the Mo and Cu samples in slurries I, II and III. Panel (C) compares the inhibition efficiencies (IEs) of HIm for suppressing the ERs and MRRs of the Mo and Cu samples in the three test slurries. The data plotted in (D) compare the Cu:Mo material selectivity of CMP in the values of ERs $\left(\mathrm{S} \left(\mathrm{ER}\right) = \mathrm{ER} \left(\mathrm{Cu}\right)/\mathrm{ER} \left(\mathrm{Mo}\right)\right)$ and MRRs $\left(\mathrm{S} \left(\mathrm{MRR}\right) = \mathrm{MRR} \left(\mathrm{Cu}\right)/\mathrm{MRR} \left(\mathrm{Mo}\right)\right)$, measured in slurries I, II and III. The error bars shown in panels (A) and (B) indicate standard deviations in the associated data obtained through repeated measurements. The error bars in panels (C) and (D) were obtained from those in (A) and (B) by using error propagation calculations.

Figure 7. (A) Displays the etch rates (ERs) and (B) shows the material removal rates (MRRs) of the Mo and Cu samples in slurries I, II and III. Panel (C) compares the inhibition efficiencies (IEs) of HIm for suppressing the ERs and MRRs of the Mo and Cu samples in the three test slurries. The data plotted in (D) compare the Cu:Mo material selectivity of CMP in the values of ERs $\left(\mathrm{S} \left(\mathrm{ER}\right) = \mathrm{ER} \left(\mathrm{Cu}\right)/\mathrm{ER} \left(\mathrm{Mo}\right)\right)$ and MRRs $\left(\mathrm{S} \left(\mathrm{MRR}\right) = \mathrm{MRR} \left(\mathrm{Cu}\right)/\mathrm{MRR} \left(\mathrm{Mo}\right)\right)$, measured in slurries I, II and III. The error bars shown in panels (A) and (B) indicate standard deviations in the associated data obtained through repeated measurements. The error bars in panels (C) and (D) were obtained from those in (A) and (B) by using error propagation calculations.

For Cu, the IE (ER) and IE (MRR) values are nearly comparable. Due to its vertical adsorption orientation (combined with its relatively low adsorption energy) on Cu [105]), the surface coverage of HIm on a Cu CMP surface can generally be varied over a broad range before reaching saturation coverage. This can be effectively done by adjusting [HIm] to adequately balance between corrosion mitigation and material removal, as well as to regulate the material selectivity of CMP [103]. The mechanical variables, P and v, of polishing can be varied to additionally adjust the material selectivity of CMP. Usually, surface films of imidazole and/or its complexes formed on metal surfaces can be readily removed in the polishing step of surface planarization [79].

3.7. Considerations for Adjusting Selectivity of Material Removal

As the data in Figure 7 indicate, the use of material-selective corrosion inhibitors to control CMP-related galvanic corrosion affects the selectivity of material removal. In the following, we briefly examine certain practical aspects of optimizing MRR selectivity while regulating galvanic corrosion effects. Dishing of Cu during barrier planarization was an effect of major concern in many earlier studies of Cu CMP [66,106,107]. For this reason, low values of Cu:barrier removal selectivity have often been used as a means to control the dishing of Cu [108]. However, substantial reduction in the barrier thickness (especially since the emergence of ≤32 nm nodes) has placed new importance on the loss of barrier liners in the barrier polishing stage [109]. This has introduced complexity in the assessment of chemical selectivity using blanket wafers for Cu/barrier slurries. A broadly practiced approach to addressing this issue is to maintain the removal rates of all materials at mutually comparable values to avoid Cu dishing, (as well as dielectric erosion) and barrier loss while polishing coplanar configurations of Cu (with the dielectric regions) and the barrier material.

Based on the above considerations, the CMP experiments aiming at Cu/barrier slurry development often focus on achieving 1:1 selectivity of Cu:barrier removal [37,81,110,111]. To experimentally set this selectivity requirement of a polishing slurry, the material-specific MRRs are frequently measured using separate blanket wafers of Cu and the barrier material [112]. A specific challenge for this approach is that a polishing pad’s pressure is uniform across the entire surface of a blanket wafer, whereas on a patterned wafer, the down-pressure of polishing is not uniformly distributed at the Cu and barrier regions that have different hardness factors [113]. As a result, the actual material-specific polish rates for patterned structures can be substantially different from those supported by the blanket wafers used for designing CMP selectivity. At the same time, galvanic corrosion measurements using blanket wafers or disc samples generally focus on a specific situation where the anode and cathode metals at a galvanic contact have equal surface areas; however, depending on the CMP system, the actual ratio of these surface areas on a patterned wafer can be a strong determining factor for the anode metal’s galvanic corrosion [64].

The considerations necessary to account for area effects in galvanic current measurements have been studied by several authors [64,114,115], and practical implications of these effects in the context of studying CMP of patterned wafers have been noted in our earlier work [65]. The role of pattern density in nonuniform pressure distribution (due to nonuniform areas of different materials) in the CMP of patterned wafers has also been discussed in the literature [116,117,118]. For a simple illustration of this effect on CMP selectivity, we briefly discuss below how material-specific topographic selectivity can affect mechanically assisted material removal. The essential elements of this present discussion are based on the general theoretical framework presented by Lai et al. [33] and Noh et al. [66].

On a patterned wafer, the down-pressure of polishing is not uniform since the fractional areas of Cu and B are different. For a patterned CMP substrate containing Cu wiring and a general diffusion barrier material (B), the MRRs for Cu and B (at the stage of barrier planarization) can be expressed, using Preston’s law [Equation (1)], as follows:

MRR(Cu) = k_P(Cu)P_Cuv

(22)

MRR(B) = k_P(B)P_Bv

(23)

where k_P(Cu) and k_P(B) denote the Preston coefficients for Cu and the barrier material, respectively, and v is the platen sample relative velocity. P_Cu and P_B are down-pressures acting on the patterned regions containing Cu and B, respectively. The Preston coefficients can be expressed as k_p(Cu) = k_w/H_Cu and k_p(B) = k_w/H_B, where H_Cu and H_B are the hardness factors for the removable surface layers of Cu and the barrier, respectively. k_w is a wear coefficient, the value of which is determined by the mechanical polishing mechanism, and can be assumed to remain largely invariant between Cu and B when both materials are simultaneously planarized [33]. In a phenomenological approach, the chemical effects of CMP can be considered as implicitly included in the effective values of H_Cu and H_B.

The [Cu:B] CMP selectivity is defined in the usual form: S(Cu:B) = MRR(Cu)/MRR(B). The condition, S(Cu:B) = 1, in Equations (22) and (23) implies that

k_P(Cu)P_Cu = k_P(B)P_B

(24)

and the spatially averaged down-pressure $(\stackrel{\mathrm{-}}{P})$ applied to the patterned sample surface for CMP has the form:

$$\stackrel{\mathrm{-}}{P} = {\displaystyle \frac{{ F}_d}{A}}= P_{\mathrm{Cu}}{A}_{\mathrm{f}} + P_{\mathrm{B}}\left(1 - A_{\mathrm{f}}\right)$$

(25)

where $F_d$ is the down force of polishing, A is the wafer area, A_f is the area fraction of Cu and (1 − A_f) is the area fraction of the barrier material deposited. Denoting the width of each Cu line (assumed to be the same over the CMP area) and the pitch pattern as w and λ, respectively, and taking A_f ≈ w/λ = ρ [119], Equation (25) can be written as follows:

$$\stackrel{\mathrm{-}}{P} = P_{\mathrm{Cu}}\rho + P_{\mathrm{B}}\left(1 - \rho \right)$$

(26)

where ρ is commonly referred to as the “pattern density” of a wafer.

To meet the requirement of S (Cu:B) = 1, Equation (24) imposes the condition,

$$P_{\mathrm{Cu}}={\displaystyle \frac{H_{\mathrm{Cu}}{P}_{\mathrm{B}}}{H_{\mathrm{B}}}}$$

(27)

and accordingly, combining Equations (26) and (27), it is found that $P_{\mathrm{Cu}}$ and $P_{\mathrm{B}}$ are neither equal to each other, nor are they equal to the average pressure applied to the sample surface:

$$P_{\mathrm{Cu}} = {\displaystyle \frac{\stackrel{\mathrm{-}}{P}}{\rho + \left\{\left(1 - \rho \right)\left(H_{\mathrm{B}}/H_{\mathrm{Cu}}\right)\right\}}}$$

(28)

$$P_{\mathrm{B}}={\displaystyle \frac{\stackrel{\mathrm{-}}{P}}{\left(1 - \rho \right)+\left\{\rho \left(H_{\mathrm{Cu}}/H_{\mathrm{B}}\right)\right\}}}$$

(29)

Using Equations (28) and (29) in Equations (22) and (23), it is possible to write

$${\mathit{MRR}}_{\mathrm{Cu}}={\displaystyle \frac{k_{\mathrm{w}}\stackrel{\mathrm{-}}{P}v}{H_{\mathrm{Cu}}\left[\rho +\left\{\left(1 - \rho \right)\left(H_{\mathrm{B}}/H_{\mathrm{Cu}}\right)\right\}\right]}}$$

(30)

$${\mathit{MRR}}_{\mathrm{B} }={\displaystyle \frac{k_{\mathrm{w}}\stackrel{\mathrm{-}}{P}v}{H_{\mathrm{B}}\left[\left(1 - \rho \right)+\left\{\rho \left(H_{\mathrm{Cu}}/H_{\mathrm{B}}\right)\right\}\right]}}$$

(31)

where the effective hardness values of both Cu and B are affected (scaled) by the pattern density.

According to the definition of $\rho$ used here, $\rho$ = 1 for a Cu blanket wafer or disc, and $\rho$ = 0 for a barrier wafer or disc. In these special cases, Equations (30) and (31) simplify back to Equations (22) and (23), respectively. Thus, the scaling factors associated with the MRRs for patterned wafers are not accounted for in the MRRs measured using blanket wafers; as a result, setting the CMP selectivity to 1:1 using MRRs for blanket wafers of Cu and the barrier material would not necessarily correspond to the actual situation of a patterned wafer as described in Equations (30) and (31). A practical approach to linking slurry compositions with CMP selectivity using blanket wafers (or metal discs) could be to allow an adjustable range of S values rather than setting a strictly defined selectivity of 1:1. This in turn requires S-tunable slurries where moderate adjustments in the additive concentrations and/or the slurry pH can support an adequate range of tunability. For the slurries used in the present work, adjustments in the value of [HIm] can serve as a useful control variable to accomplish this goal (Figure 7D).

A main utility of the experiments based on disc electrodes and blanket wafers for studying electrochemical aspects of metal CMP is that these experiments help to identify the essential slurry chemistries dictating the CMP mechanisms, the galvanic corrosion features and the material removal characteristics of CMP systems. Once the main framework of these enabling surface chemistries is established, further refinements of slurry formulations can be implemented (for instance, by adjusting the relative concentrations of the main slurry additives) to optimize the material selectivity of CMP while maintaining adequate controls of both general and galvanic corrosion effects.

3.8. Quantification of Galvanic Corrosion Results Using Intermittent OCP Transients

In the general description of mixed potential theory, E_corr and E_OC of an electrochemical interface are mutually equivalent parameters, and both define the equilibrium potential of the interface. However, due to their different measurement conditions, these parameters for a given system can be different in their values, and these differences can be rather sizable depending on the adsorption characteristics of the system. Generally, it is the value of E_corr that is primarily affected by certain potentiodynamic variables like the rate, range and direction of the voltage sweep used to measure E_corr with LSV [65]. The voltage-induced effects that are responsible for deviating the value of E_corr from that of E_OC include adsorption/desorption of solution species, activation of faradic side reactions and the associated effects of surface reconstruction [120]. In an earlier report, we have discussed certain essential aspects of this topic, and demonstrated that, in the context of measuring galvanic activation potentials, OCP results (collected in the absence of external voltage perturbations) are likely to be more reliable than those obtained from potentiodynamic measurements of E_corr [65].

Figure 8 shows typical OCP transients measured under intermittent hold (H) and polish (P) using (a) Mo and (b) Cu CMP samples in the three CMP slurries tested in this work. Each of the alternated polish and hold sequences is 4 min long, while four H cycles and four P cycles are included within the data scan. The typical signatures of tribo-noise [25] are detected as low-amplitude fluctuations in E_OC in the OCP data recorded during surface polishing. The E_OC profiles of both Cu and Mo observed in each H or P cycle are consistently repeated as the cycles are repeated. Therefore, it is evident that the CMP process does not introduce any irreversible surface modifications and does not change the CMP-specific surface chemistries over an extended timescale. Moreover, the E_OC (P) values for both metals are notably different from the corresponding values of E_OC (H), which, according to previously reported results [15], suggest that the surface layers (electro)chemically modified during the H stages are effectively removed in the P stages.

For both the metals examined in Figure 8, the E_OC plots remain mostly time-independent during each polishing cycle. This shows a predominantly steady-state nature of the electrochemical and chemical processes occurring at the CMP surfaces during polishing. In metal CMP, such steady-state behaviors of E_OC are typically observed when surface layers are mechanically removed at a balanced rate with that of chemical surface modification [25,31]. In Figure 8, this is evidenced in the absence of any OCP drifts observed through the multiple P-H cycles repeated over a relatively long sampling time of 32 min. In the description of Equation (20), a balance established between the formation and removal of surface films corresponds to a situation where the surface coverages, θ_ck and θ_al, of the cathodic and anodic reaction sites (contributing to the film’s formation) achieve steady-state values due to ongoing chemical formation and mechanical removal of the reaction products. As a result, the logarithmic term, and hence the value of E_corr (≡E_OC) in Equation (20), becomes time-independent during surface polishing. On the other hand, if the formation vs. removal rates were not matched, the cumulative effects of repeated P cycles would be manifested in a long-term temporal drift in the E_OC.

Contrary to the observations made for the E_OC (P) cycles, the E_OC (H) transients (especially those for Cu) in Figure 8 exhibit considerable variations, indicating continued deposition of products from uninterrupted surface reactions that occur during the stationary hold cycles. The entire 32 min long OCP profiles of Cu shown in Figure 8 progressively shift toward lower potentials in going from slurries I to III. These cathodic shifts in E_OC for Cu are the expected manifestation of cathodic inhibition caused by HIm adsorption at the Cu surface. At the same time, the OCP plots for Mo tend to move toward higher potentials in the same transition of slurries, although these anodic shifts in E_OC (Mo) are comparatively less drastic than the cathodic shifts in E_OC (Cu). The anodic shifts in E_OC (Mo) activated by HIm are indicative of this additive’s anodic inhibition function on the Mo surface.

In the context of OCP measurements, the activation potential for galvanic corrosion is defined in terms of the gap, ΔE_OC, between the OCP values of the cathode and anode metals. Due to mutually opposing OCP shifts for Cu and Mo observed in the HIm-containing slurries, the gap between the OCP profiles of the two metals in Figure 8 decreases with increasing [HIm], and nearly disappears in slurry III under the polishing conditions. To further quantify the implications of this observation, we plot in Figure 9 the representative OCP values, ${\overline{E}}_{\mathrm{O}\mathrm{C}},$ determined by averaging the last 2 min of data from the first P and H cycles shown in Figure 8 for Cu and Mo.

In the stationary hold case, ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Mo) in Figure 9A shifts in the anodic direction as 10 mM HIm is added to the Ref slurry, but this potential shift does not change significantly as the slurry’s HIm content is increased to 20 mM. Essentially, the same slurry-dependent trend of ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Mo) is maintained under polishing in Figure 9B. These results further solidify the inference, that HIm acts as a moderate anodic inhibitor for Mo. Moreover, the adsorption of HIm at the Mo surface likely reaches a saturation coverage in slurry II, so that the additional increase in [HIm] in slurry III does not lead to a further anodic shift in the value of E_OC (Mo).

${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Cu) in Figure 9B gradually changes to lower values with increasing [HIm] in the slurries; as we have already noted in the context of Figure 5, this reflects the broadly known cathodic inhibitor character of HIm on Cu [47,50]. This function of HIm is mostly masked in the stationary hold cases in Figure 9A, where faradaic reactions of various deposited species at the unabraded Cu surface confound the latter’s OCP characteristics.

By comparing the ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Mo) data from Figure 9B with the E_corr (Mo) data in Figure 5B, we find that under surface polishing, the slurry-dependent trends of both ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Mo) and ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Cu) are largely maintained in the corresponding data for E_corr (Mo) and E_corr (Cu). However, in the stationary hold situation, the slurry dependencies of the average OCPs in Figure 9A deviate from those of their corresponding E_corr data shown in Figure 5A. These observations further confirm that for both metals, the features of E_corr activated in the H stage (by accumulated reaction products) mostly disappear during polishing. Thus, the equilibrium potentials of the CMP interfaces are probed in a more accurate way in the polishing stage than in the hold stage. These results are consistent with those reported in our previous study of a Co-Cu bimetallic CMP system [65].

The (slurry-dependent) activation potentials for the galvanic corrosion of Mo, measured as ΔE_OC and ΔE_corr, are compared in Figure 10 for the stationary hold and dynamic polish situations. The values of ΔE_OC have been calculated from Figure 9 [ΔE_OC = ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Cu) − ${\overline{E}}_{\mathrm{O}\mathrm{C}}$ (Mo)], and those of ΔE_corr have been determined from Figure 5A,B. In all the cases considered here, noticeable differences are found between the values of ΔE_OC and ΔE_corr, which indicate how potentiodynamic measurements of E_corr are affected by voltage-activated processes. The ΔE_OC values obtained in the absence of any external voltages are relatively more useful since the voltage-induced artifacts are absent in these data, and also since traditional CMP of a metal is performed under such OCP conditions without using any voltage perturbations.

As seen in Figure 8, the OCP values (and especially those of the Cu sample) continue to change throughout each hold cycle during data acquisition. On the other hand, steady-state OCPs are established during surface polishing. Likewise, the ΔE_corr data in Figure 5 exhibit their expected slurry-dependent trends under polishing conditions, which is not the case for the stationary hold situation. Therefore, to examine the activation characteristics of galvanic corrosion in the present study, we focus specifically on the results for ΔE_OC (polish) in Figure 10 that were obtained during surface polishing. Among the different systems tested here, the case for slurry III leads to the minimum values of |ΔE_OC (polish)| and |ΔE_corr (polish)|. The reversal of galvanic polarity observed in the ΔE_corr (polish) data for this case no longer appears in the ΔE_OC (polish) data; this shows that the polarity reversal effect seen in the ΔE_corr measurements is likely linked to voltage-induced artifacts like faradaic side reactions activated during the LSV scans used to measure E_corr.

Since the value of ΔE_OC (polish) for slurry III is effectively minimized, and since the associated measurement conditions are compatible with those of metal CMP, we take this OCP result as a main outcome of optimizing the citrate-based CMP slurry chemistry to regulate the galvanic corrosion of Mo. According to the data in Figure 5C, this CMP system also helps to effectively suppress the galvanic current density at the Mo surface. The 15 mV value of ΔE_OC (polish) for slurry III is well below the thermal voltage of ~30 mV established at the typical temperatures of a pad–sample interface in metal CMP [121]. Thus, within the limits of the current experiments, slurry III appears to meet the main requirements for regulating the CMP-related galvanic corrosion of Mo at the Cu-Mo contact regions.

3.9. Relative Roles of Surface Chemistry and Tribology in Material Removal

The results presented in Section 3.2, Section 3.3, Section 3.4, Section 3.5 and Section 3.6 indicate that slurry chemistries play direct roles in governing the (general as well as galvanic) corrosion characteristics of the Mo-Cu CMP system. According to the data in Figure 7, it is also evident that the material removal rates of CMP are strongly affected by these slurry chemistries. However, in many cases of metal CMP, these chemical effects play rather indirect roles in determining the net MRRs, as the rates of (electro)chemical surface modifications in such cases are found to be considerably lower than the corresponding MRRs. A customary approach to probing these comparative (and often synergistic) roles of slurry chemistry and tribology is to compare the MRRs in different test slurries with the corresponding rates of surface corrosion, CR (P), measured under surface polishing. Corrosion rates are used here as surface modification rates, because for most metal CMP systems (including those studied in this work), the CMP-enabling chemical changes in the abraded surface occur through corrosion-like reactions of material wear.

The phenomenological basis for assessing CMP mechanisms by comparing the values of MRRs and CR (P) is rooted in the following expression,

$$MRR= R_{\mathrm{c}} + R_{\mathrm{w}\mathrm{c}}+R_{\mathrm{c}\mathrm{w}}+R_{\mathrm{w}}$$

(32)

which is commonly used to examine the chemical and mechanical components of CMP [122,123]. $R_{\mathrm{c}}$, $R_{\mathrm{w}\mathrm{c}}$, $R_{\mathrm{c}\mathrm{w}},$ and $R_{\mathrm{w}}$, denote, respectively, the material removal rates activated by the wear mechanisms of (electro)chemical corrosion, wear-induced corrosion, corrosion-induced wear, and mechanical wear. If the structurally weak and mechanically removable surface films of a CMP metal are (electro)chemically formed at a rate, r_f (P), and if these films are mechanically removed at a rate, R_f (P), then $R_{\mathrm{c}}$ + $R_{\mathrm{w}\mathrm{c}}$ ≈ ER (P) + x r_f (P), where x = R_f (P)/r_f (P).

Under steady-state conditions of CMP, the value of x is expected to be close to 1. As we noted above in the discussion of Figure 8, the relatively flat OCP transients observed during surface polishing justify the assumption that x ≈ 1 in the present experiments. Moreover, balanced rates of surface film formation and removal generally lead to repeatable MRRs measured with different polishing intervals and different samples. In the discussion of Figure 7, we have already noted this repeatability of MRRs by referring to the rather small error bars associated with the MRR data. Based on these observations, Equation (32) in the present case can be expressed as follows:

$$MRR \approx \left[ER\right(P)+{ r}_{\mathrm{f}}\left(P\right)]+ R_{\mathrm{c}\mathrm{w}}+ R_{\mathrm{w}}{= CR\left(\mathrm{P}\right)+ R}_{\mathrm{c}\mathrm{w}}+ R_{\mathrm{w}}$$

(33)

where the surface corrosion rate, CR (P), activated under polishing is taken as a combination of the rates of dynamic etch, ER (P), and insoluble film formation.

While the last two terms in Equation (33) cannot be readily accessed using standard tribo-electrochemical methods, the net contribution of these terms to the measured MRRs can be evaluated in terms of the difference, [MRR − CR(P)]. This analysis is carried out in Figure 11. The corrosion rates (CRs) of the Mo and Cu CMP samples measured under (A) hold and (B) polish conditions are shown in the top two panels of Figure 11. These results have been obtained from the i_corr data using the formula $\mathit{CR} = (M{i}_{\mathrm{corr}})/\rho \mathit{nF}$, where M and ρ are the molecular weight and mass density of the CMP sample, respectively; n is the number of electrons transferred in the mixed potential reactions (discussed in Section 3.2) causing material wear. Active presence of tribo-corrosion is evident here since the CRs measured with surface abrasion are higher than those monitored under stationary conditions. The tribo-corrosion rates (TCRs), determined by subtracting the CR (H) values from their corresponding CR (P) values, are plotted in Figure 11B.

The CR (P) values for Cu decrease with increasing [HIm], following the slurry-dependent trends of their corresponding ERs and MRRs, and thus, indicate a clear correlation between material removal and corrosion-like surface modifications. For the CR (H) data in Figure 11A, this correlation is not clearly established for Mo because, (as already noted in the discussion of Figure 6A), these data are affected by electroactive reaction products accumulated at the unabraded sample surfaces. Nevertheless, since the CR (P) values dominate those of the TCRs, the latter data mostly retain the slurry-dependent variations in their associated MRRs and ERs. While the variations in MRR and CR(P) follow mutually similar patterns, the absolute values of CR(P) for all three test slurries are notably lower than those of their corresponding MRRs; this is shown in Figure 11D, where the differences, [MRR − CR(P)], calculated using the data from Figure 7B and Figure 11B, are plotted.

The data in Figure 11D are consistent with those previously reported for other cases of metal CMP [15,65,124,125], and in the description of Equation (33) indicate that a substantial component of material removal comes from the terms, $R_{\mathrm{c}\mathrm{w}}$ and $R_{\mathrm{w}}$. However, the MRRs for metal CMP are generally associated with negligibly small components of $R_{\mathrm{w}}$ [126]. To examine the contributions of R_w to the measured MRRs in this study, we note that wear rates monitored under cathodic protection are often used in certain applications to selectively measure mechanical wear without the interference of chemical corrosion [127,128]. However, this approach requires the application of a rather large cathodic overpotential where strong hydrogen evolution is activated, and this frequently leads to hydrogen embrittlement followed by hydrogen-induced stress cracking in the test material [129]. Since the intrinsic hardness of the test sample can be altered in this process, properly adapting the cathodic protection approach to measure R_w for CMP systems is not straightforward. Therefore, to check the strength of $R_{\mathrm{w}}$ in the present work, we measured both the ERs (at 40 °C) and the MRRs of Mo and Cu in a “blank” medium, only using distilled water and 3 wt% silica abrasives at pH = 7, without including any chemicals. These latter experiments showed that the intrinsic aqueous chemistries of Cu and Mo supported rather small rates of static etch, 0.5 and 0.96 nm min⁻¹, for Cu and Mo, respectively. The MRRs measured in chemical-free distilled water were also small, 1.45 and 5.36 nm min⁻¹, for Cu and Mo, respectively.

According to the Pourbaix diagrams of Mo, the soluble species HMO₄⁻ and MO₄²⁻ can be generated by water at pH = 7 without requiring any chemical additives [via Equations (9)–(11), followed by Equations (4)–(8)]. These soluble species can be linked to the ERs for Mo measured in neutral distilled water. Additionally, the sparingly soluble species of MoO(OH)₂ can form over the Mo sample surface at pH = 7 due to reactions (4)–(6), all of which can be supported by additive-free water. Similar situations for Cu are found from an inspection of the Pourbaix diagrams of Cu in pure water. At the sample temperature of Cu established during CMP, the soluble species Cu(OH)²⁻ may be released at a low level in neutral water to support a measurable value of the ER by the reaction: Cu + 2(OH)⁻ = Cu(OH)²⁻ [130]. At the same time, insoluble Cu(OH)₂ can be deposited onto the Cu sample surface in neutral water via reaction (16) [131]; mechanical removal of this species by polishing can be attributed to the blank medium MRR measured for Cu.

Thus, the low levels of ERs and MRRs measured for Mo and Cu in distilled water are likely dominated by the intrinsic aqueous chemistries of these metals, possibly in combination with even lower contributions of strictly mechanical wear. The components of MRR and ER supported by chemical-free water are inseparable from and hence remain embedded in the values of CR (P) measured using the full slurry composition with water serving as the slurry’s solvent. Due to this integrated function of water in the chemical component of material removal, it is difficult to separately measure an absolute value of R_w. Nevertheless, based on the considerations of Equation (1) and published experimental data, the contribution of strictly mechanical R_w to the overall MRR is expected to be quite small in metal CMP. Therefore, the measured difference, MRR − CR(P), can be attributed to the term R_cw. As discussed previously [65], the mechanisms of material wear responsible for the component R_cw can include elastic disparities between unmodified and modified surface materials [31,132,133], accelerated propagation of dislocations defects and corrosion fatigues in subsurface regions [134], and/or corrosion-related lowering of the local threshold for plastic deformation [124].

4. Conclusions

The present study explores the potential of using citric acid as a complexing agent in SiO₂-based pH-neutral slurries for simultaneously planarizing Cu and Mo. A specific focus of this work is on analyzing and controlling the galvanic corrosion features of this bimetallic CMP system, coupled with advancing the fundamental understanding about CMP-related galvanic corrosion in general. The results of the tribo-potentiodynamic and intermittent OCP measurements show that Mo, as an anode metal in the Cu-Mo CMP system, is prone to undergoing galvanic corrosion induced by the cathode metal, Cu. It is also shown that the electrochemical indicators (galvanic current and activation potential) of galvanic corrosion can be quite different between measurements performed without and with surface abrasion for CMP, and that only the electrochemical variables measured in the presence of polishing are reliable for CMP applications.

Our investigation further indicates that imidazole can serve as an effective corrosion inhibitor for both Cu and Mo in the polishing slurries tested. The galvanic suppression effects can be controlled by controlling the slurry concentration of HIm. With 20 mM HIm in the slurry, the galvanic current density flowing through the Mo anode under surface polishing is reduced to a value of 21 µA cm⁻², which is notably lower than the corrosion current density of Mo (120 µA cm⁻²) correspondingly activated by the metal’s CMP-enabling surface chemistry under polishing. The activation potential for the galvanic corrosion of Mo under this condition also drops to 15 mV, which is well below the average thermal voltage (~30 mV) supported at the CMP interface. It is shown that the mechanism of galvanic corrosion suppression by HIm is based on site-blocking adsorptions of imidazole onto the Cu and Mo surfaces. Additionally, evidence is presented for faradic (charge transfer-mediated) adsorption of imidazolium cations onto Mo–oxide surface sites.

The experiments reported herein also examine the mechanisms of material removal from Cu and Mo. It is demonstrated that the CMP-specific surface chemistries (corrosion and tribo-corrosion) are essential to supporting material removal from both metals. Removal rates measured using distilled water at pH = 7 without any chemicals show that the intrinsic aqueous chemistries of Cu and Mo only support rather small rates of static etch (0.5 and 0.96 nm min⁻¹ for Cu and Mo, respectively) and material removal under polishing (1.45 and 5.36 nm min⁻¹, for Cu and Mo, respectively). The surface chemistries supported by SPC (H₂O₂) and CA increase the MRRs and ERs for Cu by factors of 38 and 68, respectively. For Mo, these increases in MRR and ER occur by the factors of 2.4 and 3.3, respectively. However, these chemically supported removal rates only provide minor contributions to the overall MRRs measured here in CMP, as the MRR values appear to be dominated (~97% for Cu and ~83% for Mo) by the effects of corrosion-induced subsurface wear.

The analytical framework illustrated in this work for designing and evaluating CMP slurries can be readily extended to other systems of metal CMP. This approach can be useful to measure the CMP-specific electrochemical variables under tribology-coupled conditions closely mimicking those of fab-based polishing with reusable benchtop model systems in the laboratory setting. A future expansion of this method could involve the implementation of wafer (rather than disc) samples in the tribo-electrochemical test cell. To highlight the relevance of the present results in the overall context of BEOL processing, we note, in particular, that the neutral pH (and chemically dominant) approach to slurry formulation demonstrated here is aligned with the practical considerations for low-k integration [135]. Strong alkaline slurries often damage low-k dielectrics (like hydrogen silsesquioxane) by hydrolyzing the latter’s Si–OH bonds [136]. On the other hand, acidic slurries can damage the carbon-based groups in porous low-k materials like SiCOH, making the dielectric hydrophilic with increased k values [137]. At the same time, both acidic and alkaline CMP slurries are susceptible to developing localized corrosion on Cu surfaces [138].

Many of the pH-related issues of metal CMP mentioned above can be effectively addressed by employing pH-neutral slurries, and by adequately assessing the slurries’ functions in a comprehensive tribo-electrochemical approach, as shown here. Furthermore, the rates of CMP-specific surface modifications (CRs in Figure 11) in this work are substantially lower than those of the MRRs. Under these conditions, residual surface coverages of the metal–citrate and metal–HIm complexes should be relatively insignificant and readily removable during post-CMP cleaning of their co-adsorbed oxide residues [139]. Due to the foregoing reasons (in addition to those already noted in Section 3.1), the slurry selection strategies, in combination with the detailed tribo-electrochemical protocols of slurry evaluations illustrated in this work, should help to further develop the general guidelines for Cu and barrier CMP.