Tribo-Electrochemical Considerations for Assessing Galvanic Corrosion Characteristics of Metals in Chemical Mechanical Planarization

Abstract

The manufacturing of integrated circuits involves multiple steps of chemical mechanical planarization (CMP) involving different materials. Mitigating CMP-induced defects is a main requirement of all CMP schemes. In this context, controlling galvanic corrosion is a particularly challenging task for planarizing device structures involving contact regions of different metals with dissimilar levels of corrosivity. Since galvanic corrosion occurs in the reactive environment of CMP slurries, an essential aspect of slurry engineering for metal CMP is to control the surface chemistries responsible for these bimetallic effects. Using a CMP system based on copper and cobalt (used in interconnects for wiring and blocking copper diffusion, respectively), the present work explores certain theoretical and experimental aspects of evaluating and controlling galvanic corrosion in barrier CMP. The limitations of conventional electrochemical tests for studying CMP-related galvanic corrosion are examined, and a tribo-electrochemical method for investigating these systems is demonstrated. Alkaline CMP slurries based on sodium percarbonate are used to planarize both Co and Cu samples. Galvanic corrosion of Co is controlled by using the metal-selective complex forming functions of malonic acid at the Co and Cu sample surfaces. A commonly used corrosion inhibitor, benzotriazole, is employed to further reduce the galvanic effects.

1. Introduction

Chemical mechanical planarization (CMP) is a critical part of material processing in the fabrication of modern integrated circuits (ICs) [1]. Regardless of its utility and efficiency, CMP is prone to generating various types of surface defects that are detrimental to device functions [2]. For this reason, and due to the rapidly growing numbers and complexity of the CMP steps, keeping CMP-induced defects to a minimum level has been an ongoing challenge for several years. Galvanic corrosion is a CMP defect frequently encountered in IC fabrication while planarizing the contact regions of metals/alloys from different positions in the galvanic series [3,4,5,6]. If galvanic corrosion operates in the interconnects in CMP, microscopic grooves or divots are formed at the Cu/barrier interfaces. Galvanically corroded CMP defects, usually characterized as edge-over-erosion (EOE) or “fang defects” [7], can lead to severe degradation and, eventually, failure of device performance.

Since the effects of galvanic corrosion are system-specific, it is difficult to set up a general recipe of slurry chemistries to prevent these effects for a general class of metal/alloy CMP systems. Thus, it becomes necessary to assess the criteria for suppressing galvanic corrosion on a case-by-case basis. At the same time, the complex tribo-electrochemical features of CMP slurries tend to affect the conventional electrochemical tests (usually performed in the absence of surface abrasion) that are designed to probe galvanic corrosion. The results obtained through such measurements often fall short of establishing a steadfast foundation to adequately address the defectivity issues that are directly linked to CMP-related galvanic corrosion.

While in general, galvanic corrosion occurs on alloys and metals, the specific systems considered in the present work strictly involve bimetallic combinations with metals from different parts of the galvanic series. A specific goal of this study is to identify and address the limitations of traditional electrochemical tests for studying bimetallic corrosion in metal CMP. We focus on the electrochemical techniques of potentiodynamic polarization (PDP) and open circuit potential (OCP) measurements that are most used to evaluate the galvanic corrosion characteristics of metal/alloy CMP systems. We examine how the different control variables of these techniques affect the measurements of the galvanic corrosion indicator variables. Phenomenological considerations for selecting optimized conditions for these measurements are discussed. In the same context, we demonstrate a tribo-electrochemical approach for handling some of the practical constraints of the conventional electrochemical probes of CMP-related galvanic systems. We demonstrate how these experiments can guide slurry designs to suppress CMP-induced galvanic corrosion. Additionally, we illustrate how the data obtained from these measurements can be used to determine the mechanisms of material removal in CMP.

The general framework of this investigation begins with a discussion of the phenomenological background of CMP-induced galvanic corrosion. The indicator variables of galvanic corrosion are described using mixed potential theory. Experimentally, a model CMP system of current interest based on Co (barrier metal) and Cu (wiring metal) is examined, where Co serves as the corrosion-prone anode in its galvanic couple with a Cu cathode. A moderately alkaline polishing slurry is employed with sodium percarbonate (SPC) as an oxidizer and malonic acid (MA) as a complexing agent for both metals. A commonly used corrosion inhibitor, benzotriazole (BTA), is examined to determine its function in suppressing galvanic corrosion of Co. The speciation characteristics of MA and BTA are shown in Figure S1 in the Supplementary Materials.

Tribo-electrochemical measurements of PDP and OCP with or without simultaneous implementation of surface abrasion are performed. The results provide mechanistic information about the process of material removal in CMP. Experiments are designed to demonstrate how the control variables of PDP and OCP measurements affect the recorded data. It also shows how these latter effects can potentially lead to misinterpreted and/or incomplete analyses of galvanic corrosion data. Certain strategies are discussed for optimizing tribology-controlled PDP and OCP experiments to examine the galvanic corrosion characteristics of metal CMP systems.

2. Phenomenological Background, Theoretical Considerations and Experimental Strategies

2.1. Activation and Effects of CMP-Induced Galvanic Corrosion

Since various aspects of galvanic corrosion in the context of metal/alloy CMP have been extensively discussed in the literature [5,8,9,10,11,12,13,14,15,16], only certain main elements of this phenomenon are briefly outlined here. Figure 1 schematically shows the main steps of a galvanic corrosion process activated at the contact region between a cathode metal (M_C) and an anode metal (M_A) immersed in an ion conducting slurry solution. For illustration, the solution is considered alkaline, and the reduction of dissolved O₂ is taken as a cathodic reaction supported at the surface of M_C. Due to its relatively higher nobility in the galvanic series, the cathode metal draws electrons for this reaction from the less noble anode metal. Consequently, the anode surface electro-dissolves in the form of M_Aⁿ⁺ with n denoting the charge valency of the anode’s cation.

The arrows in Figure 1A trace the steps of electron transfer, O₂ adsorption, dissolution of M_Aⁿ⁺ in the solution and production as well as desorption of OH⁻ following the oxygen reduction reaction (ORR). The “circuit” for the galvanic current arising from this electron transfer in the metals is closed by ion transport in the solution with anions and cations flowing toward the cathode and the anode, respectively. The anode region nearest to the bimetallic boundary is most affected by galvanic corrosion, where the path length for electron transfer from M_A to M_C is minimal. The galvanic current (I_g) flows from M_C to M_A in the metals. The current in the solution phase is supported by cation flow from M_A to M_C and anion flow in the opposite direction.

Figure 1B shows schematic Evans plots illustrating the characteristic parameters of galvanic corrosion that are generally measured using PDP experiments. The individual Evans plots for M_C and M_A in the given slurry solution are shown in blue and red lines, respectively. The potential (E) dependent currents in the cathodic (upper) and anodic (lower) Tafel branches for M_C are denoted as I_c(C) and I_a(C), respectively. The corresponding current components for M_A are I_c(A) and I_a(A). The PDP plot for M_C is characterized by the corrosion potential and corrosion current, E_corr(C) and I_corr(C), respectively. The corresponding variables for M_A are denoted as E_corr(A) and I_corr(A). E_corr(C) appears at a higher potential than E_corr(A) owing to the comparatively higher nobility of M_C. The driving potential for galvanic corrosion (galvanic polarization) of M_A is ΔE_corr, defined as

ΔE_corr = E_corr(C) − E_corr(A)

(1)

which represents a key variable for assessing the galvanic corrosion character of a bimetallic system.

In the schematic representation of Figure 1B, galvanic coupling of metals C and A occurs at the mixed potential coordinate (Log I_g, E_g), indicated by the yellow dot on the combined polarization plots. E_g is the galvanic potential and I_g is the galvanic current

$$[{I_{\mathrm{a}}\left(A\right)]}_{{E=E}_{\mathrm{g}}}=[{I_{\mathrm{c}}\left(C\right)]}_{{E=E}_{\mathrm{g}}}=I_{\mathrm{g}}$$

(2)

which is the electrode current established under equilibrium at E_g. Experimentally, a composite Evans diagram following the scheme of Figure 1B can be assembled using PDP plots separately collected for M_C and M_A in the same test solution. The corrosion variables ΔE_corr, I_corr(C), I_corr(A), E_g and I_g, necessary to characterize the [M_C,M_A] bimetallic system can be determined from the composite PDP plots as illustrated in Figure 1B.

The barrier planarization step of Cu CMP is a typical situation where favorable conditions for CMP-induced galvanic corrosion can be activated [5,8,13]. This CMP situation is schematically considered in Figure 2, where parts (a) and (b) show two essential steps of Cu CMP. After bulk Cu is removed in the first step (a), both Cu and the barrier material (M_b) are polished together in the second step (b). The cross-sectional dimensions of the Cu wiring line and the diffusion barrier are relevant for illustrating the electrical features of the Cu/barrier combination. These features of the Cu/barrier profiles are further discussed in Section S.2. of the Supplementary Materials accompanying this report. The dimensional restrictions on the thickness of the barrier line [17] are discussed in that context, referring to a Supplementary Schematic in Figure S2.

Most frequently, Cu and M_b occupy different positions in the galvanic series, and both materials are exposed to the ion conducting slurry (electrolyte) during barrier CMP. A galvanic current is established under this condition, with electrons flowing within the metals from the less noble side to the more noble side. The resulting effects of galvanic corrosion are schematically shown in Figure 2c,d following the drawing scheme used in Figure 1A. The barrier material (such as Co) is galvanically corroded if it is less noble than Cu (M_A ≡ M_b, and M_C ≡ Cu); the opposite case is found for barrier metals like Ru that are more noble than Cu (M_A ≡ Cu, and M_C ≡ M_b). Due to their characteristic spatial contours, these galvanic corrosion defects are often referred to as “fang” defects, and based on their location, they are also described as “edge over erosion” features [18,19].

The dimensional restrictions on the barrier liners are directly linked to the electrical performance requirements of Cu interconnects [20]. While the barrier layer thickness in modern devices is approaching the atomic dimension [21,22], the electrical resistivity of most materials increases at this dimension due to electron scatterings caused by Fuchs-Sondheimer and Mayadas-Shatzkes effects [23]. At the same time, the majority of barrier materials that are suitable to meet the electrical criterion (in addition to meeting the barrier requirements of good thermal stability, high melting point, and aptness for seedless copper deposition [24]) are rather separated from Cu in the galvanic series. For this reason, either the barrier material (such as Co, Mn and Ti, located below Cu in the galvanic series) or Cu (when used, for instance, with Ru-based barriers) tends to suffer galvanic corrosion during CMP. Current slurry formulations for CMP processing at the back end of the line require robust strategies to avert Cu-dishing and line-thinning, as well as losses of barrier materials [25,26].

2.2. Theoretical Formulation of Galvanic Corrosion Parameters

We first examine the corrosion variables of the individual Tafel plots of the metals C and A, considered in Figure 1B. In the classical Butler Volmer formalism [27], the general expressions for the cathodic and anodic currents (I_c and I_a, respectively) in the Tafel branches can be written as follows:

$$I_{\mathrm{c}}=I_{0\mathrm{c}}{\mathrm{e}}^{-{\alpha }_{\mathrm{c}}f{\eta }_{\mathrm{c}}}$$

(3)

$$I_{\mathrm{a}}=I_{0\mathrm{a}}{\mathrm{e}}^{{(1-\alpha }_{\mathrm{a}})f{\eta }_{\mathrm{a}}}$$

(4)

where η_c and η_a are applied overpotentials for the mixed potential system’s cathodic and anodic components, respectively. η_c = E − E_rc and η_a = E − E_ra, with E_rc and E_ra denoting the equilibrium Nernst potentials of these cathodic and anodic reactions, respectively; I_0c and I_0a denote the corresponding exchange currents, while α_c and α_a are the transfer coefficients of the cathodic and anodic reactions, respectively; f = nF/RT, where n is the reaction valence, R is the gas constant; F and T denote the Faraday constant and the sample’s absolute temperature, respectively. The metal-specific labels, A and C, are not included in the above parameters since Equations (3) and (4) apply to both metals.

The exchange currents in Equations (3) and (4) depend on the fractional surface coverages θ_c and θ_a, of the cathodic and anodic sites [28]. On the other hand, the exchange current densities, i_0c and i_0a are characteristic kinetic parameters of the cathodic and anodic reactions, respectively, as determined by the intrinsic rate constants of their corresponding reactions. Accordingly, I_0c = A₀θ_ci_0c, and I_0a = A₀θ_ai_0a, where A₀ is the geometric surface area of the CMP sample (electrode). The consideration of site-specificity also applies to the net electrode currents:

I_c = A₀θ_ci_c

(5)

and similarly, I_a = A₀θ_ai_a, where i_c and i_a represent the cathodic and anodic site-specific electrode current densities, respectively.

The mixed potential parameters, E_corr and I_corr for the individual metals can be obtained by noting that at the mixed (corrosion) potential, the difference between E_rc and E_corr serves as a cathodic overpotential, which results in a current I_corr in the cathodic branch of the metal’s Tafel plot. In other words, η_c = E_corr − E_rc, I_c = I_corr at E = E_corr. Similarly, for the anodic reaction occurring at E = E_corr, it is possible to write η_a = E_corr − E_ra and I_a = I_corr. Incorporation of these identities in Equations (3) and (4)

$$E_{\mathrm{ra}}-E_{\mathrm{corr}}={\beta }_{\mathrm{a}}\ln \left({\displaystyle \frac{I_{0\mathrm{a}}}{I_{\mathrm{corr}}}}\right)$$

(6)

$$E_{\mathrm{rc}}-E_{\mathrm{corr}}=-{\beta }_{\mathrm{c}}\ln \left({\displaystyle \frac{I_{0\mathrm{c}}}{I_{\mathrm{corr}}}}\right)$$

(7)

where β_a = 1/[f(1 − α_a)], and β_c = 1/fα_c. Mixed potential expressions for $E_{\mathrm{corr}}$ and $I_{\mathrm{corr}}$ can be obtained in the following forms by solving Equations (6) and (7) [29]:

$$E_{\mathrm{corr}}={\displaystyle \frac{{\beta }_{\mathrm{a}}{\beta }_{\mathrm{c}}}{\beta }}\left({\displaystyle \frac{E_{\mathrm{ra}}}{{\beta }_{\mathrm{a}}}}+{\displaystyle \frac{E_{\mathrm{rc}}}{{\beta }_{\mathrm{c}}}}\right)+\ln \left({\displaystyle \frac{i_{0\mathrm{c}}{\theta }_{\mathrm{c}}}{i_{0\mathrm{a}}{\theta }_{\mathrm{a}}}}\right)$$

(8)

$$I_{\mathrm{corr}}=(i_{\mathrm{corr}}{A}_0)={\left({\theta }_{\mathrm{a}}{i}_{0\mathrm{a}}\right)}^{{\displaystyle \frac{{\beta }_{\mathrm{a}}}{\beta }}}{\left({\theta }_{\mathrm{c}}{i}_{0\mathrm{c}}\right)}^{{\displaystyle \frac{{\beta }_{\mathrm{c}}}{\beta }}}{e}^{{\displaystyle \frac{{∆E}_{\mathrm{r}}}{\beta }}}$$

(9)

where β = β_a + β_c, ΔE_r = E_rc − E_ra and $i_{\mathrm{corr}}$ is the corrosion current density. In terms of $E_{\mathrm{corr}}$ and $I_{\mathrm{corr}}$, the Tafel equations for the metal take the following forms:

$$E-E_{\mathrm{corr}}={\beta }_{\mathrm{a}}\ln (I_{\mathrm{a}}/I_{\mathrm{corr}})$$

(10)

$$-(E-E_{\mathrm{corr}})={\beta }_{\mathrm{c}}\ln (I_{\mathrm{c}}/I_{\mathrm{corr}})$$

(11)

where Equations (10) and (11) correspond to the anodic $\left({E>E}_{\mathrm{corr}}\right)$ and cathodic $(E<E_{\mathrm{corr}})$ Tafel branches of their associated Evans diagram, respectively.

Considering next the case of bimetallic coupling, we note that at the galvanic mixed potential (E = E_g), the anodic current, $I_{\mathrm{a}} (A; E_{\mathrm{g}}),$ of metal A and the cathodic current, $I_{\mathrm{c}} (C; E_{\mathrm{g}})$, of metal C are equal and represent the galvanic current, $I_{\mathrm{g}}$. Implementing these conditions in Equations (10) and (11), the galvanic current can be expressed as

$$I_{\mathrm{a}} (A; E_{\mathrm{g}})=I_{\mathrm{corr}}\left(A\right)e^{{\displaystyle \frac{E_{\mathrm{g}}-E_{\mathrm{corr}}\left(A\right)}{{\beta }_{\mathrm{a}}\left(A\right)}}}=I_{\mathrm{g}}$$

(12)

$$I_{\mathrm{c}} (C; E_{\mathrm{g}})=I_{\mathrm{corr}}\left(C\right)e^{{\displaystyle \frac{E_{\mathrm{corr}}\left(C\right)-E_{\mathrm{g}}}{{\beta }_{\mathrm{c}}\left(C\right)}}}=I_{\mathrm{g}}$$

(13)

and the following expression for the galvanic potential, $E_{\mathrm{g}}$, can be obtained by combining Equations (12) and (13) [30]:

$$E_{\mathrm{g}}={\displaystyle \frac{{\beta }_{\mathrm{c}}\left(C\right)E_{\mathrm{corr}}\left(A\right)+{\beta }_{\mathrm{a}}\left(A\right)E_{\mathrm{corr}}\left(C\right)}{{\beta }_{\mathrm{c}}\left(C\right)+{\beta }_{\mathrm{a}}\left(A\right)}}+{\displaystyle \frac{{\beta }_{\mathrm{a}}\left(A\right){\beta }_{\mathrm{c}}\left(C\right)}{{\beta }_{\mathrm{c}}\left(C\right)+{\beta }_{\mathrm{a}}\left(A\right)}}\ln \left[{\displaystyle \frac{i_{\mathrm{corr}}\left(C\right)S_{\mathrm{c}}}{i_{\mathrm{corr}}\left(A\right)S_{\mathrm{a}}}}\right]$$

(14)

where $S_{\mathrm{c}}$ and $S_{\mathrm{a}}$ denote the effective surface areas of the cathode and anode metals, respectively. Incorporating Equation (14) into Equation (12) leads to the formula below describing the galvanic current density at the anode metals’ surface [30]:

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

(15)

where β_ca = β_c(C) +β_a(A). The rate of the anode’s galvanic corrosion is proportional to the galvanic current density i_g (A).

2.3. Electrochemical Assessment of CMP-Induced Galvanic Corrosion

According to Equation (15), the value of $i_g(A$) depends on several system-specific factors including the following: (i) The difference in the corrosion potentials (ΔE_corr) between the cathode metal (C) and the anode metal (A); (ii) the surface areas of the cathode (S_c) and anode (S_a) exposed to the slurry’s ionic environment; (iii) the uncoupled corrosion current densities of the cathode [i_corr(C)] and the anode [i_corr(A)]; (iv) the cathodic and anodic Tafel coefficients, [β_c(C) and β_a(A)], respectively, of the cathode and anode metals. The term, ΔE_corr, represents the activation polarization for galvanic corrosion, and due to its appearance in the exponential term in Equation (15), usually serves as a predominant factor in determining the value of i_g. For this reason, experimental assessments of galvanic corrosion involving CMP-related bimetallic systems have traditionally centered on ex situ measurements of ΔE_corr using PDP plots for uncoupled samples of the individual metals.

Since solution compositions play a strong role in dictating the value of ΔE_corr [31], adjustments of slurry formulations are coupled with these measurements to minimize the galvanic polarization. Metal CMP slurries developed according to this strategy generally require one or more cathodic inhibitors for the more noble cathode metal and/or anodic inhibitors for the less noble anode metal [32,33]. A cathodic inhibitor deposited at a cathode metal’s surface decreases the value of θ_c, and, according to Equation (8), shifts E_corr to a lower potential. Likewise, an anodic inhibitor adsorbed at an anode metal’s surface decreases the value of θ_a, which, according to Equation (8), shifts E_corr to a higher value. Thus, the values of E_corr(C) and E_corr(A) move closer to each other, decreasing the value of ΔE_corr. However, utilizing the full functions of metal selective anodic and cathodic inhibitors often becomes a challenging task in the CMP environment where both metals of a galvanic couple are simultaneously abraded [34]. A further consideration for determining ΔE_corr of galvanic couples using PDP is that the morphological and chemical makeups of the test surfaces often change when external potentials are applied to these surfaces [35,36].

Since the corrosion potential of an electrochemical interface is theoretically equivalent to the corresponding OCP, the OCP difference, ΔE_OC, between the cathode and anode metals of a galvanic couple can also be measured and used as an indicator of galvanically affected systems [37,38]. This approach helps to avoid potential induced surface modifications, but it does not explore how the individual metals’ corrosion currents and Tafel slopes affect galvanic corrosion. While the approach based on the analyses of ΔE_corr or ΔE_OC provides an overall phenomenological basis to assess galvanic corrosion of static test systems, several practical factors, specifically those of polishing-tribology, make it difficult to standardize the method of these assessments in the CMP context. The present work addresses certain essential aspects of this topic.

3. Materials and Methods

3.1. Sources and Treatments of Metal Samples and Polishing Slurries for CMP

Cu (99.9%) and Co (99.95%) coupon samples of 1.0 mm thickness (McMaster-Carr, Elmhurst, IL, USA), each cut into a 20 mm × 20 mm square piece, were used to measure material etch rates (ERs). The metal samples used for CMP and tribo-electrochemical tests were 99.99% pure Cu and 99.95% pure Co polycrystalline discs, each of 2.54 cm diameter (Kurt J. Lesker Co., Jefferson Hills, PA, USA). The samples were placed into separate Teflon holders, specifically designed to fit in the polisher head of a CMP machine, as described in detail elsewhere [29]. Before testing, each sample was polished to a mirror finish by sequentially using sandpapers of different grits (1000, 1500, 2000, 3000 and 5000), and finally using aqueous pastes of alumina powders of 1 μm diameter on a Microcloth (the last two items obtained from Buhler, Lake Bluff, IL, USA) attached to a rotary polisher built in our laboratory. The metal samples prepared in this way were subsequently cleaned for 5 min by sonication using a BRANSON 2510 ultrasonic cleaner (Branson Ultrasonics Corporation, Danbury, CT, USA).

All ERs were measured in beaker cells placed on a hot plate stirrer (Fisherbrand^TM, Shanghai, China) with magnetically stirred (at 300 RPM) test slurries maintained at 40 °C to approximately simulate a typical combination of pad temperature and slurry flow used in CMP [4]. The samples for both ER and MRR measurements were prepared using the same method described above. Both measurements were based on the gravimetric approach with a resolution of 0.01 mg. Standard errors in the measurements of ERs and MRRs have been included as error bars in the plots of these parameters. The metal samples were tested in four different CMP slurries, all prepared at pH = 8.5 using reagent grade chemicals and triple distilled water. The slurry chemicals included KNO₃ (Fisher Scientific, Waltham, MA, USA), Malonic acid [MA, CH₂(COOH)₂, from Alfa Aesar, Ward Mill, NY, USA], Benzotriazole (BTA, C₆H₅N₃), potassium hydroxide (KOH, a pH adjuster) and sodium percarbonate (SPC, Na₂CO₃·1.5H₂O₂) (last three items obtained from Aldrich Chemical, Milwaukee, WI, USA).

3.2. Tribo-Electrochemical Instruments and Measurements

The design and operation of the tribo-electrochemical test cell used in this work have been previously discussed in extensive detail [29]. In brief, the cell employed a Struers Labopol Benchtop polisher (Struers LC, Cleveland, OH, USA), electrochemically coupled with a Solartron 1287 potentiostat and a Frequency Response Analyzer 1252A (both from Ametek Scientific Instruments, Berwyn, IL, USA). The polisher contained two dynamic parts, a polishing head and a rotating platen, both rotated at a common angular speed of zero RPM (under stationary hold) or 90 RPM (under dynamic polish). The CMP sample also served as the working electrode (WE) for electrochemical measurements using the polishing slurry as an electrolyte.

The rotating platen of the polisher was covered with an IC-1000 polishing pad (Rohm and Haas, Philadelphia, PA, USA). The rotation speeds of the polisher’s platen and head, as well as the down pressure of polishing [0.014 MPa (2 psi) used for this work] could be independently varied. A stainless steel ring attached to the inner edge of the rotating platen was used as the counter electrode (CE), and a saturated calomel electrode (SCE) was used as the reference electrode (RE) by incorporating a salt bridge to avoid leakage of KCl from the SCE into the tested slurries. The slurry volume was 150 mL, and another 100 mL of slurry was included in the salt bridge. Abrasive particles were excluded from the slurry contained in the salt bridge to avoid clogging the latter’s glass frits.

The slurry pH was taken as an indicator of the chemical stability of each test slurry, and the pH values of the individual slurries were measured before and after reacting every slurry under dynamic conditions with Cu or Co. For these measurements, the CMP metal in the form of a rectangular thin coupon of 1 mm thickness was dipped (reacted) for 60 min in 100 mL of the slurry, stirred with a magnetic stirrer (to simulate the slurry motion during CMP) at 300 RPM, and maintained at 40 °C (to mimic the typical slurry temperature during CMP) in a beaker with 2.0 cm × 2.0 cm area on each side of the coupon exposed to the slurry. ERs were also determined using this arrangement in the gravimetric approach by measuring the metal coupon’s mass before and after reacting with each metal coupon. MRRs were determined in a similar way using gravimetric measurements.

Tribo-electrochemical tests included intermittent OCP transient measurements, electrochemical impedance spectroscopy (EIS), and liner sweep voltammetry (LSV). Measuring intermittent OCP transients required 16 min to complete 2 polish sequences, alternated with 2 hold sequences, extending each of these sequences for 4 min. The EIS experiments used a perturbation spectrum of 10 mV average amplitude, spreading over a frequency range from 1 Hz to 10 kHz in logarithmically spaced intervals. Recording data for each EIS cycle took about 3 min. LSV was conducted using voltage sweeps at 5 mV s⁻¹, optimized for tribo-potentiodynamic measurements as described previously [39], in the forward (increasing potential) or reverse (decreasing potential) directions, typically covering a potential range of ±0.5 V around the OCP. Completing each LSV scan required 3.3 min (200 s) to cover 1 V at a scan rate of 5 mV s⁻¹.

3.3. Data Analysis Protocols

Potentiodynamic polarization (PDP, or Tafel) plots were generated from the LSV data after correcting the raw data for the ohmic drop caused by finite slurry resistances, R_s. Thus, the voltage axis of a corrected PDP plot used the actual electrode potential E_e, calculated as E_e = E − IR_s, with E and I denoting the applied electrode potential and the measured electrode current, respectively. The values of R_s were determined by using electrochemical impedance spectroscopy and analyzing the EIS data in the complex nonlinear least square (CNLS) approach using ZSimpWin software (version 3.50). Tafel extrapolations of corrected PDP data and plotting of the figures were carried out using Origin software (2024 version).

4. Results and Discussion

4.1. Results of Material Removal Measurements

The four experimental slurries used in this work are listed in

Table 1. Nomenclatures, chemical compositions and Ohmic resistances of test slurries *.

Slurry	Composition	Rs (Ω) (Cu, Co)
I	0.1 M KNO3 + 1 wt% H2O2 + 3 wt% SiO2 (Ref)	72.5, 89.6
II	Ref + 0.1 M MA	23.7, 38.3
III	Ref + 1 mM BTA	138.1, 111.8
IV	Ref + 0.1 M MA + 1 mM BTA	57.2, 44.6

* The first and second entries (separated by a comma) in each row of the column for R_s represent slurry resistances measured using Cu and Co working electrodes (CMP metal samples), respectively.

with their compositions and solution resistances (R_s). Speciation diagrams of MA and BTAH as functions of slurry pH are shown in the Supplementary Material (Figure S1). At the experimental pH of 8.5 used here, Mal²⁻ (Mal ≡ C₃H₂O₄) is the predominant dissolved species of MA. When present in the slurry, the Mal²⁻ anions contribute to the slurry’s ionic conductivity, and for this reason, the R_s values in the MA-containing slurries are lower than their counterparts in the MA-free slurries.

Since the R_s values were measured under stationary hold conditions, the interfacial slurry fluids residing between the polishing pad and the CMP sample added their resistance in series with the solution resistance [40]. The amount and spatial distribution of this interfacial fluid generally depend on the surface film’s structure, and the latter feature varies between Cu and Co. This effect is responsible for the different values of R_s observed between the Cu and Co samples. This sample dependent behavior of R_s is further augmented in the BTA-containing slurries, especially when MA is not present to boost the solution conductivity.

The values of (A) ERs and (B) MRRs for the Cu and Co CMP samples in the different slurries are presented in Figure 3, along with (C) the selectivity S (Cu:Co) of Cu:Co material removal. For the Cu sample, the slurry dependent trends of ERs are largely maintained in the corresponding values of MRRs, while this is not the case for Co. These different features of the two metals can be linked to the observation that MA acts as a complexing agent for Cu (increasing both ERs and MRRs), while the role of MA in the case of Co is manifested as that of both a corrosion inhibitor (decreasing MRRs) and a complexing agent (increasing ERs). These different outcomes of MA-supported surface modifications for Cu and Co can be linked to different complex forming characteristics of MA at the oxidized surfaces of Cu and Co. The metal-specific reaction mechanisms of MA for complex generations on the Cu- and Co-CMP surfaces are examined in the next section. Slurries I–III are in the category that supports high Barrier:Cu selectivity [41,42,43,44], while slurry IV corresponds to the frequently utilized selectivity of ~1:1 [9,45,46].

4.2. Surface Reactions for Material Removal Examined Using Tribo-PDP Data

Surface modifications for metal CMP generally involve a decrease in the material hardness of the CMP sample, activated by mixed potential reactions of metal-oxide and surface complex formations [18,47,48]. In the following, we discuss these reactions for the present experimental systems. To check the presence and the relative roles of these reactions in the CMP test slurries, we employ a selected set of PDP measurements where the cathodic and anodic components of the CMP-enabling mixed potential reactions were voltage activated. In this way, the presence of the expected reactions could be verified by probing their signature features in the data. These data are presented in Figure 4 (for Cu) and Figure 5 (for Co), where certain signature features of MA and BTA in the slurries can be examined. These plots are taken from the data recorded during the first LSV scan applied in the forward (increasing voltage) direction using fresh metal samples.

In the hydrogen peroxide-based alkaline slurries used here, the cathodic currents are mostly generated by the electro-reduction of H₂O₂:

$${\mathrm{H}}_2{\mathrm{O}}_2+2{\mathrm{e}}^{-}=2{\mathrm{O}\mathrm{H}}^{-}$$

(16)

while the oxygen reduction reaction (ORR),

$${\mathrm{O}}_2+2{\mathrm{H}}_2\mathrm{O}+4{\mathrm{e}}^{-}=4{\mathrm{OH}}^{-}$$

(17)

likely has a relatively weaker contribution to these currents. The cathodic current branches shown in Figure 4 are supported chiefly by reaction (16), with a secondary contribution of reaction (17). Anodic reactions of Cu are driven by the OH⁻ product of reaction (16) [as well as reaction (17)] to form Cu₂O and CuO:

2Cu + 2OH⁻ = Cu₂O + H₂O + 2e⁻

(18)

Cu₂O + 2OH⁻ = 2CuO +H₂O + 2e⁻

(19)

and thus, the oxidized Cu surface can contain a mixture of Cu₂O and CuO.

In Figure 3A, measurable values of ERs are found in the four slurries used. Since complexing agent-free slurry I supports this ER of Cu, the associated soluble species of Cu in this slurry should be a dissolved oxide or hydroxide species of Cu, assisted by OH⁻ in the alkaline environment. According to published Pourbaix diagrams of Cu, the most probable candidate for this soluble species is Cu(OH)₂⁻ [49]. In the CMP situation considered here, this species can be generated by the following reaction of Cu₂O and OH⁻ resulting from the products in Equations (18) and (16), respectively [50]:

Cu₂O + H₂O + 2OH⁻ = 2Cu(OH)₂⁻

(20)

and this dissolution reaction can be attributed to the ER(Cu) values measured in the MA-free slurries. Finite contributions of this reaction to the ERs of Cu are also expected to remain operative in the MA-containing slurries. Reactions (18) and (19) are the predominant faradaic steps that give rise to the anodic current branches seen in Figure 4. The chemical reaction (20) changes the surface coverage of Cu₂O participating in reaction (19) and, thus, affects the net currents supported in these branches.

When MA is present in the slurry, the dissolved Mal²⁻ chemically supports the partial dissolution of the CuO surface species [49,51]:

CuO + 2Mal²⁻ + H₂O = CuMal₂²⁻ + 2OH⁻

(21)

and this reaction is responsible for the somewhat increased ER of Cu in going from slurries I to II. Adsorption of BTA at the Cu surface in slurries III and IV suppresses these dissolutions and hence lowers the ERs.

The CuO formed on Cu via reactions (18) and (19) is a dual inhibitor with a relatively stronger effect of anodic suppression in H₂O₂-based solutions [52,53,54]. Thus, the partial dissolution of CuO from the Cu surface increases the anodic currents (accompanied by a moderate increase in the cathodic currents) in plot (b) in slurry II compared to those seen in plot (a) for slurry I. This same effect is observed by comparing plots (a) and (b) in the BTA-containing slurries III and IV.

In addition to the aforesaid surface reactions, the MA-based slurries allow the formation of a nearly insoluble Cu-malonate surface complex, CuMal [51]:

CuO + Mal²⁻ + H₂O = CuMal + 2OH⁻

(22)

and this CuMal plays a predominant role in material removal from abraded Cu in the MA-based slurries, although it does not contribute to the ERs for Cu [54]. This role of reaction (22) is evident in Figure 4, where the currents observed on Tafel plot (d) collected under surface abrasion in MA-based slurry II are larger than their corresponding currents associated with plot (c) recorded under polishing in MA-free slurry I. This shows that the CuMal surface complex is effectively abraded off the Cu surface.

As shown in Figure S1B in the Supplementary Materials, the BTA added to the CMP slurry exists as a mixture of BTAH (~33%) and BTA⁻ (~67%). The function of MA in augmenting the MRRs in Cu CMP is also observed by comparing plots (c) and (d) in the BTA-based slurries. Nevertheless, some portions of the Cu surface sites in these slurries are occupied by the chemisorbed BTA, BTAH and Cu-BTA complexes [55,56]. These latter sites are no longer available to support the material removal reactions of OH⁻ (MRR of Cu drops from slurry II to slurry III). Mal²⁻, however, cooperatively adsorbs with the BTA species, and both Cu-MA and Cu-BTA complexes contribute to material removal under polishing (Figure 3).

According to the literature, the main surface products of Co generated in alkaline media without additives in the 25–50 °C temperature range are solid Co(OH)₂ and soluble Co(OH)₃⁻ [57]. The surface reaction leading to the Co(OH)₂ species is as follows [14,57]

Co + 2OH⁻ = Co(OH)₂ + 2e⁻

(23)

where the reactant OH⁻ in the present case comes from reactions (16) and (17). It is now broadly accepted that Co(OH)₂ serves as a removable material in the CMP of Co in alkaline slurries [39,58,59,60]. The rather sizable ERs detected in all four experimental slurries indicate that a soluble Co species is generated in these slurries. Specifically, since this ER is supported in slurry I (which contains no complexing agents), the most likely soluble species of Co stabilized at pH = 8.5, according to the Pourbaix diagram of Co, is Co(OH)₃⁻ [57].

At the Co-CMP interface, Co(OH)₃ species can be generated as follows by partial dissolution of Co(OH)₂ [produced in reaction (2)] by the OH⁻ [generated in reaction (16)]:

Co(OH)₂ + OH⁻ = Co(OH)₃⁻

(24)

and this soluble species of Co is expected to control the ER(Co) values plotted for all four slurries in Figure 3A. Reaction (24) likely contributes to the ERs plotted for Co in Figure 3A. According to the data in Figure 4B, Co(OH)₂ acts as a dual inhibitor for Co by suppressing both the anodic and cathodic activities of the metal. As Co(OH)₂ is removed from the Co surface by polishing, both the anodic and cathodic current branches of Co move to higher currents in going from plots (a) to (c).

The reported values of the isoelectric point of Co(OH)₂ vary between 11.0 and 11.4 [61]; accordingly, the majority of the Co(OH)₂ surface sites formed on Co are positively charged at the slurry-pH of 8.5 used here. Considering the electrostatic mechanism of surface adsorption, the existing Co(OH)₂ surface sites of Co in the MA-based test slurries should facilitate chemisorption of Mal²⁻ at the Co(OH)₂ surface sites of Co. The adsorbed Mal²⁻ would react with Co(OH)₂, and a probable makeup of this reaction can be proposed by reviewing the literature on Co-dicarboxylic acid complexes as well as the published reaction schemes for Co CMP in slurries containing these acids.

Previously published results for Co CMP in oxalic acid-based alkaline slurries indicated the formation of a soluble Co-oxalate complex as the dissolved Ox²⁻ anions reacted with Co(OH)₂ at a Co surface [39,62]: Co(OH)₂ + 2Ox²⁻ = Co(Ox)₂²⁻ + 2OH⁻. Here, Ox ≡ C₂O₄, and Co(Ox)₂²⁻ is the predominant Co-oxalate complex at the CMP interface in the alkaline environment [63,64]. We also note that the aqueous speciation of Co-malonate complexes is similar to that of Co-oxalate [65,66]; specifically, like Co(Ox)²⁻, its malonate counterpart, Co(Mal)₂²⁻ is a dominant species in alkaline solutions [66,67]. Based on these observations, we propose the following reaction of Co-malonate complex formation to explain the ER (Co) data in Figure 3A:

Co(OH)₂ + 2Mal²⁻ = CoMal₂²⁻ + 2OH⁻

(25)

which, in addition to Equation (24), serves as a further mechanism of material dissolution in the present experiments involving Co. As reaction (25) is activated in the MA-based slurries, the values of ER(Co) increase in the transitions from slurries I and III to slurries II and IV, respectively.

As the (dual inhibitor) passivating layer of Co(OH)₂ is partly removed from the Co surface by reaction (25) in a stationary hold situation, both the anodic and cathodic current branches of Co move to higher currents in the transition from plots (a) to (b) in Figure 5A. Essentially, these same effects of MA are found to operate under surface abrasion of Co when we compare the Tafel plots (c) and (d) in Figure 5B. The relative positions of these plots along the potential axis are determined according to Equation (8) by the different surface coverages of Co(OH)₂ supported under the hold and polishing conditions and by the presence/absence of MA in the slurry.

In the BTA-based slurries, the solution species of BTAH and BTA⁻ adsorb onto the Co surface both in their complexed and un-complexed forms with Co [68,69]. The surface sites necessary to support reactions (23)–(25) are largely blocked by these BTA-related species in slurries III and IV. Consequently, both the ER and MRR values measured in these slurries are lower than their corresponding values measured in slurries I and II.

4.3. Mechanistic Aspects of the Chemical Mechanisms of CMP

Equations (16)–(25) together constitute the chemical mechanism of the Cu- and Co-CMP processes studied here. In chemically promoted CMP of metals, the reactions generating material wear under static conditions operate with the same reactants and products under the conditions of dynamic polishing [56,70], and the reaction rates often change in the latter case due to the activation of tribological effects [29]. Thus, signature features of reactions (16)–(25) have been detected under both the hold and polish conditions in the data presented in Figure 4 and Figure 5. The reactions (21), (22) and (25) of MA illustrate how the surface complexing mechanisms of MA differ between Cu and Co and, consequently, lead to the different MA-supported yields of ERs and MRRs for the two metals (Figure 3).

While PDP measurements serve to reveal the faradaic components of the CMP reactions under voltage activation, the CMP process, in practice, occurs under open circuit conditions in the absence of externally applied overpotentials. Since the faradic steps of CMP reactions in the latter case are supported in the mixed potential mode, it is useful to briefly note the mixed forms of these reactions. The oxidant-reduction steps in Equations (16) and (17) operate with the anodic steps, leading to material wear of both Cu and Co. For instance, the mixed potential form of reactions (16) + (18) for Cu is Cu + H₂O₂ = CuO + H₂O. Likewise, the mixed version of reactions (16) + (19) yields Cu₂O + H₂O₂ = 2CuO + H₂O, which is frequently associated with Cu-CMP in alkaline slurries [71,72]. For Co, reactions (16) and (23) are coupled as Co + H₂O₂ = Co(OH)₂. The mixed potential steps for Cu and Co supported by reaction (17) can be considered in a similar way.

As indicated in the above mixed reactions, the OH⁻ ions released by oxidant reduction steps are consumed by anodic steps of surface oxidation and complex formation, leading to material wear in the CMP metal. This process acts to maintain the effective slurry pH at the CMP surface. On the other hand, if the mixed potential mechanism is inefficient, the local pH at the CMP surface can sufficiently change and be detected in the bulk slurry, especially if the slurry-volume-to-sample-area ratio is small. As indicated in the results of pH measurements (presented in the Supporting Materials) using stirred slurries at 40 °C, no major changes were introduced in the slurry-pH after reacting each slurry with the CMP metal for an hour. This observation, in view of the foregoing discussion, is consistent with the mixed potential mechanism of material removal for metal CMP.

A close examination of the mechanistic factors associated with reactions (21), (22) and (25) can further clarify the effects of MA on the measured ERs and MRRs. These effects are dictated by the comparative adsorption properties of OH⁻ and Mal²⁻ at the oxidized and unoxidized sites of Cu and Co, which, in turn, determine the makeups of the complexes stabilized on the two metal surfaces. Anion adsorptions at metal surfaces are dictated by several factors, including electrostatic effects, metal-ion dispersion forces, metal-water and ion-water interactions, as well as geometric configurations of adsorbate packing [73]. The relative strengths of these effects vary in metal- and ion-specific manners. Accordingly, in alkaline polishing slurries, Cu exhibits a higher affinity toward adsorbing OH⁻ compared to the ions of organic acid complexing agents [14,15]. The reaction sites of complexing ions in these cases are dominated by Cu-oxides and/or hydroxides rather than by bare Cu. Reactions (21) and (22) are based on this prevalence Cu-OH⁻ interactions, and active roles of these reactions in MA-supported Cu CMP have been established previously [54].

In the case of Co, however, the data in Figure 3A,B suggest that Mal²⁻ anions compete with OH⁻ for the unoccupied adsorption sites of Co and Co(OH)₂. The potential of zero charge (PZC) of Co is −0.64 V [68], and (as shown in experimental data presented later in this report) the lowest OCP values recorded here for the Co-based test systems were around −0.26 V, considerably above the metal’s PZC. Thus, for all the experimental systems of Co tested in this work, the Co surface contained excess positive charges. The relatively higher negative charge content of Mal²⁻ should make this anion a stronger candidate than OH⁻ for adsorption at the positively charged Co surface sites. Such an effect would reduce the population of chemisorbed OH⁻ supporting rection (23), and hence, the coverage of Co(OH)₂ at the Co surface would correspondingly decrease. Since Co(OH)₂ serves as a removable material in Co-CMP, the MRRs for Co drop when the formation of Co(OH)₂ is restricted in the MA-containing slurries. This expected effect of MA is manifested in the data presented in Figure 3B.

According to Equations (21) and (22), Mal²⁻ adsorbed at the oxidized Cu surface generates both soluble CuMal₂²⁻ and insoluble CuMal complexes, which contribute to increasing both the ERs and MRRs of Cu in the MA-based slurries (Figure 3). On the other hand, the Mal²⁻ adsorbed at the Co(OH)₂ sites of Co only forms the soluble species CoMal₂²⁻. Thus, in addition to restricting the production of mechanically removable Co(OH)₂, the Mal²⁻ adsorbed at the Co CMP surface does not generate any insoluble Co-malonate complexes to serve as a removable material under abrasion. Consequently, the MRRs of Co drop in the presence of MA, and the net effect of MA in the slurry is then manifested as that of a corrosion inhibitor.

4.4. Modes of Material Removal

The surface reactions (16)–(25) describe the underlying (electro)chemical processes of material removal for the Cu/Co test systems used, but the reaction steps alone do not indicate the interplay of chemical and mechanical components of CMP. To examine the latter aspect of CMP mechanisms, the chemical, mechanical and combined chemical-mechanical modes of material removal can be described in terms of the following frequently used expression [74,75]:

$$\mathit{MRR}=\mathit{ER}\left(P\right)+x {[r}_{\mathrm{f}}\left(P\right)]+R_{\mathrm{cw}}+R_{\mathrm{w}}$$

(26)

where $\mathit{ER}\left(P\right)$ and $r_{\mathrm{f}}\left(P\right)$ denote, respectively, the material-etch (dissolution) rate and the rate of insoluble surface (oxide/complex) film formation under polishing conditions of CMP; x is a proportionality factor between the rates of formation [$r_{\mathrm{f}}\left(P\right)$] and removal (R_f) of surface films. If the last two rates are mutually balanced, then x = 1. $R_{\mathrm{wc}}$ and $R_{\mathrm{cw}}$ are material removal rates supported by wear-induced corrosion and corrosion-induced wear, respectively.

The first two terms together on the right-hand side of Equation (26) represent the contributions of chemical wear (R_c) and wear-induced corrosion (R_wc). These attributions can be noted by phenomenologically interpreting the terms R_c and R_wc as follows: R_c = ER(H) + ${x [r}_{\mathrm{f}}\left(H\right)]$, and R_wc = [ER(P) − ER(H)] + ${x[ r}_{\mathrm{f}}\left(P\right)-r_{\mathrm{f}}\left(H\right)]$. Experimentally, R_wc can be measured as the tribo-corrosion rate (TCR), which is the difference between corrosion rates (CRs) supported in the presence and in the absence of surface abrasion [54]. The contribution of $R_{\mathrm{w}}$ to the observed MRR in metal CMP is insignificant in most cases of metal CMP [76]. This was verified in the present work by measuring MRRs (and ERs) for both the Cu and Co samples in a “blank slurry” of pH-neutral distilled water mixed only with 3 wt% silica abrasives and no other chemical additives (data not included here). Both the ERs and MRRs measured using this blank slurry yielded negligible values, and in all cases, remained confined within the uncertainties of these measurements.

The material corrosion rate, CR(P), recorded under polishing conditions of CMP, provides an estimate for the strength of chemically dictated material removal. This follows from the definition CR(P) = ER(P) + $r_{\mathrm{f}}\left(P\right)$, where the right-hand side becomes equal to R_c + R_wc if x ≈ 1. The foregoing formulation of CR represents a generalized version of the conventional CRs that focus mostly on electro-dissolution steps. The generalized CR defined above is often used in the CMP context and includes all anodic processes that contribute to material wear via dissolution as well as via the formation of hardness-reduced surface layers of oxides/complexes.

The terms CR(P) (under polishing) and CR(H) (under stationary hold) can be determined by measuring the corrosion current densities and using the definition $\mathit{CR}=(M{i}_{\mathrm{corr}})/\rho \mathit{nF}$, where M and ρ denote the molecular weight and mass density of the CMP material, respectively. The values of i_corr for the Cu and Co samples under hold (H) and polishing (P) conditions were determined through Tafel extrapolations of the PDP plots in Figure 4 and Figure 5. These corrosion current densities and their corresponding CRs are presented in Figure 6A and Figure 6B, respectively.

As seen in Figure 6B, the CRs for both metals generally increase with the inclusion of MA in the slurry and decrease in the BTA-containing slurries. In view of the chemical description of CMP [26], these are the expected effects of MA as a complexing agent and BTA as a corrosion inhibitor. While the MRRs for Cu in Figure 3 follow essentially the same trends of CR(Cu), the MA-affected MRRs of Co exhibit a different trend of CR(Co). This shows once again how MA, an established CMP promoter for Cu, does not play the same role in the present case of Co CMP and essentially acts as an adjuster of Cu:Co selectivity (Figure 3C) in material removal. This observation also points to the fact that chemical mechanisms alone do not dictate the MRRs for the present CMP systems. This is also evident in the observation that all the MRRs examined in Figure 3 for Cu and Co are significantly higher than their corresponding values of CRs in Figure 6B. These system-dependent differences between MRR and CR(P) values are plotted in Figure 6C. Such large differences between the rates of material removal and corrosion are also found in previous studies of metal CMP [14,77,78,79].

CR values measured from corrosion currents (can be affected by, but) do not register the components of ERs that originate from chemical reactions without involving interfacial charge transfers. While the dissolution reactions (20), (21), (24) and (25) in the present study represent such chemical steps, the reactants of these steps, CuO and Co(OH)₂, form via faradaic processes and hence contribute to the measured corrosion currents. Based on these observations, the rate differences plotted in Figure 6C can be wholly attributed to the term R_cw in Equation (26), while no measurable contributions of R_w to measured MRRs are detected. The underlying mechanisms responsible for activating this R_cw mode have been discussed previously [79,80]. The factors contributing to the R_cw term in the present study can include mechanisms such as accelerated propagation of dislocations defects and corrosion fatigues [81], elastic disparities between unmodified and modified surface materials [39,82,83] and/or corrosion-related lowering of the local threshold for plastic deformation [84].

4.5. Experimental Constraints and Considerations for Assessing CMP-Related Galvanic Corrosion

4.5.1. Effects of Mechanical Abrasion on PDP Results

Evaluation of the activation potential ΔE_corr for checking galvanic characteristics of CMP systems is commonly carried out through two separate exsitu measurements, where the E_corr values of individual metal samples of a bimetallic system are separately determined. With a few exceptions [11,85], most previously reported potentiodynamic evaluations of galvanic corrosion involving CMP systems have customarily used stationary metal samples in bulk solutions without applying down pressures or mechanical abrasion to the CMP surface [10,86,87,88,89,90,91]. Measurements of OCP values have also been reported using similar (pressure- and abrasion-free) sample configurations [37,38,47,48,92]. The data recorded in bulk solutions using such arrangements are useful for identifying the main (electro)chemical reactions. However, both OCP and E_corr values of metal–electrolyte interface are known to change under surface abrasion, and this phenomenon provides the working foundation of the intermittent OCP transient technique broadly used in the general field of tribo-electrochemistry [93,94,95,96,97,98] as well as in electrochemical studies of CMP systems [11,99,100,101,102,103].

Some measurements of (ΔE_OC and/or) ΔE_corr have been reported where only the sample has been rotated while the polishing pad has been held stationary under pressure. In this arrangement, the velocity (and hence the MRR) decreases from the CMP sample’s edge toward its center, and the center misses dynamic abrasion [104]. The sample configuration used in our present and recent studies closely mimics that of a CMP tool where both the sample and the platen are rotated at a common angular velocity [29].

Table 2. Effect of surface abrasion on PDP-measured corrosion potentials of Cu and Co in CMP Test Slurries *.

Slurry	Ecorr(H;Cu)	Ecorr(H;Co)	Ecorr(P;Cu)	Ecorr(P;Co)	ΔEcorr(H)	ΔEcorr(P)
I	0.139	−0.368	0.175	0.033	0.507	0.142
II	0.035	−0.366	0.161	0.114	0.401	0.047
III	0.120	−0.229	0.195	0.090	0.349	0.105
IV	0.089	0.045	0.186	0.144	0.044	0.042

* All the corrosion potentials are quoted in the unit of V vs. SCE, and the values of ΔE_corr are in V. E_corr(H;Cu) and E_corr(P;Cu) denote corrosion potentials for the Cu CMP sample measured under the conditions of stationary hold (H) and dynamic polish (P), respectively. Similar notations are used for the Co CMP sample by replacing “Cu” with “Co” in the labels.

presents illustrative results of abrasion-affected E_corr values, assembled by Tafel extrapolations of the PDP plots for Cu and Co from Figure 4 and Figure 5.

In all the cases examined in Table 2, E_corr (Cu) > E_corr(Co), indicating the roles of Cu and Co as the cathode and anode metals, respectively, in the Cu-Co bimetallic couple. Furthermore, the corrosion potentials for both Cu and Co exhibit notable changes in the transition from the sample conditions of no abrasion under down pressure (H) to active abrasion under down pressure (P). Thus, the galvanic polarizations, ΔE_corr, calculated from these E_corr data, would also contain abrasion-induced variations in their values. If they are not accounted for, these variations can be problematic for implementing such PDP results in the defect-mitigation strategies of CMP.

Aside from the general corrosion variables of the individual metals, the galvanic corrosion parameters, i_g and E_g, of the bimetallic couple also change between the cases of no surface abrasion and active abrasion. To explore these latter effects, the values of E_g and for the Cu-Co bimetallic test system have been determined from Figure 4 and Figure 5 and plotted in Figure 7A and Figure 7B, respectively. The values of i_g are particularly illustrative for evaluating the slurry-dependent galvanic corrosion characteristics of the Cu-Co CMP system studied here. Although the galvanic corrosion affinities of the Cu-Co CMP systems are generally described in terms of ΔE_corr values, some results for i_g (Co) are available in the literature. For instance, in their investigation of Cu-Co CMP in glycine-based slurries, He et al. have used a corrosion inhibitor, 1, 2, 4-Triazole, to minimize galvanic corrosion of Co and reported galvanic current densities that varied between 22 and 163 μA cm⁻², depending on the inhibitor concentration [85]. Johnson and Roy have used varied amounts of BTA in an acetate-based CMP slurry to regulate galvanic corrosion of Co in a bimetallic combination with Cu. They reported galvanic current densities under abrasion with values ranging from 0.9 and 9.8 μA cm⁻² [14]. The overall results of these earlier studies are comparable to the range of i_g observed here in Figure 7B.

In agreement with other authors’ findings [11,85], the results presented here in Table 2 and Figure 7 demonstrate how the values of ΔE_corr, E_g and i_g change in going from static to polishing conditions of CMP [14,15,105,106]. The underlying factors responsible for these effects are briefly noted below, and further results demonstrating these effects are presented in the next section. Frictional heat generated during CMP is a major factor responsible for changing E_corr of a CMP interface between hold and polish conditions. The pad temperature can rise up to ~60 °C due to friction-generated heat under typical conditions of CMP [107]. The rates of additive adsorption/desorption from the slurry, as well as those of surface reactions, generally increase due to this thermal effect of CMP [108,109]. These processes directly affect the values of (i_0c/i_0a) and (θ_c/θ_a) in Equation (8). Depending on the relative values of β_a and β_c, the thermal effect may also be manifested in the value of E_corr through temperature-dependent variations of Tafel slopes (although this latter effect appears to be insignificant in the present experiments, based on a visual comparison of the Tafel slopes of the PDP plots recorded with and without surface polishing).

In addition to the thermal effects, other tribological processes linked to the down pressure and hydrodynamics of CMP also play active roles in governing the values of E_corr (and E_OC) under polishing conditions. For instance, the contact pressure between a CMP sample and a polishing pad strongly affects the competitive and cooperative adsorptions of the interfacial slurry additives onto a CMP surface [110]. The adsorbate coverages established under the polishing pad’s pressure are different from those supported in a bulk slurry where the sample surface does not have a pad in its contact. During CMP, adsorption of reaction intermediates occurs from a moving slurry maintained under pressure. Depending on the pad-sample separation (thickness of interfacial slurry), the shear rate can vary typically between (50 and 100) × 10³ s⁻¹ [111]. This affects the viscosity of the slurry at the CMP surface [112] in a competitive manner with the (opposite) thermal effect of friction [107]; depending on the operative lubrication regime, the adsorption/desorption characteristics of the CMP-enabling reaction intermediates further change under active friction of surface abrasion [113]. All these effects can contribute to shifting the value of E_corr in a transition from the hold to polish conditions.

4.5.2. Effects of Area Factors

As shown in Equation (15), the galvanic current density at the anode metal’s surface in a bimetallic combination is proportional to the term, ${{(S}_{\mathrm{c}}/S_{\mathrm{a}})}^{{\beta }_{\mathrm{c}}/{\beta }_{\mathrm{ca}}}$. Accordingly, the smaller the anode’s surface area, the higher the rate of galvanic corrosion at the anode surface. This effect of the cathode/anode area ratio is a significant governing factor for the anode’s galvanic corrosion [114]. However, PDP-based ex situ measurements using separate anode and cathode samples, as those considered in Figure 4 and Figure 5, do not account for these area effects. The surface areas S_c and S_a are taken to be equal in the calculations of i_g shown in Figure 7, and this is a commonly used limiting assumption of such ex situ experiments.

The area ratio $(S_{\mathrm{c}}/S_{\mathrm{a}})$ for an actual CMP system is difficult to estimate in a laboratory setting since this ratio tends to be device-specific. Moreover, this area ratio is expected to change during the CMP process if galvanic corrosion of the anode is already initiated. In the presence of Cu dishing (schematically shown in Figure S2 in the Supplementary Material), the surface area of Cu in contact with the slurry changes as dishing progresses [17]. Depending on whether Cu serves as an anode or a cathode for a given CMP system, this changing surface area of Cu due to dishing will have different effects on the galvanic corrosion characteristics of the barrier/Cu combination. Thus, owing to the difficulty of mimicking the bimetallic geometry of an actual device in an experimental model, the cathode/anode area effects remain largely unexplored in such model-based experiments.

4.5.3. Effects of PDP Control Variables

Electrochemical variables of CMP systems, recorded with or without the sample’s surface abrasion, generally depend on the scan rate, scan direction and voltage range of PDP [115,116,117]. Minimizing these effects is critical for employing PDP to measure corrosion variables, including those (like i_g and E_g) commonly used to assess galvanic corrosion characteristics of bimetallic systems. Furthermore, electrochemical test cells involving CMP systems are often characterized by relatively sizable solution resistances (R_s) in the 50–100 Ω cm² range [118,119,120,121]. Unless corrected for this ohmic drop, the PDP-measured values of i_g and E_g can be misleading and difficult to compare with published values. This is particularly important for CMP-related systems where the margin of uncertainty in the results is quite restricted due to the scaling of IC features. Some of the aforesaid effects can be unavoidable due to system-specific restrictions on the measurement conditions. Nevertheless, while operating within these limits, adequately accounting for the limiting experimental factors is necessary to set up a general framework for quantitative analyses of CMP-related bimetallic corrosion.

The irreversible effects of the cycling sequence on PDP-measured corrosion parameters [117] are generally activated by the overpotentials used for PDP experiments. These overpotentials are applied to access the Tafel region, and hence need to considerably exceed the thermal voltage. At such voltages, the metal sample surface in a CMP slurry can undergo surface reconstructions, many of which are irreversible. Voltage-induced surface reconstruction is a well-known electrochemical phenomenon [122] and has been reported for both Cu- [35,36,123] and Co-based [124] systems. For instance, Cu surface complexes can be reconstructed under voltage activation to generate Cu nanoclusters with certain electrochemical properties that are not shared by the original surface [125]. Likewise, Co(OH)₂, a product of CMP-enabling surface reactions on Co in alkaline slurries, can undergo reconstruction in a chemically active medium [124], where the resulting surface’s electrochemical response would differ from that of the initial Co surface. PDP (LSV) responses of such surfaces recorded before and after reconstruction tend to vary in successive voltage scans [117].

Typical examples of some of the aforesaid experimental artifacts, which are recorded with PDP in this study for the Cu and Co CMP samples in slurries I and II, are shown in Figure 8. The corresponding results for slurries III and IV are presented in Figure 9. All these plots are corrected for the ohmic drop of solution resistance by replacing the applied potential E with the corrected electrode potential, E_e = (E − IR_s). In each panel of Figure 8 and Figure 9, the PDP data for Cu and Co are plotted together to bring out the slurry-dependent nature of the galvanic polarization, ΔE_corr, between the two metals. Each sample was subjected to two successive LSV scans in the forward direction (increasing potentials) and two successive scans in the reversed direction (decreasing potentials).

Figure 8 and Figure 9 show the results for the forward LSV scans, and the corresponding data for the reverse scans are presented in the Supplementary Materials. The values of E_corr obtained from the different PDP plots in Figure 8 and Figure 9 are stated within boxes in the panels. The effects of uncompensated solution resistance on the PDP-measured corrosion variables were explored by replotting the full set of the aforementioned data without correcting the values of E for the ohmic drop of R_s. These latter results are shown in the Supplementary Materials.

The collective results presented in Figure 7 and Figure 8, together with those shown in Figures S3–S6 in the Supplementary Materials, bring out the different effects of LSV scan settings and finite slurry resistances on the galvanic corrosion variables. The key observations assembled from these Figures are summarized in

Table 3. Effects of slurry resistance, LSV scan sequence and scan direction on PDP-based measurements of Cu-Co Galvanic parameters *.

Galvanic Variable	Slurry I	Slurry II	Slurry III	Slurry IV
Eg(H-F) (V vs. SCE)	−0.100, −0.204	−0.145, −0.144	−0.044, −0.034	0.071, 0.037
Eg(H-R) (V vs. SCE)	−0.152, −0.190	−0.069, −0.113	0.038, 0.006	0.123, 0.073
E/g(H-F) (V vs. SCE)	−0.108, −0.203	−0.147, −0.143	−0.045, −0.038	0.072, 0.036
E/g(H-R) (V vs. SCE)	−0.140, −0.179	−0.061, −0.111	0.026, −0.002	0.119, 0.066
Eg(P-F) (V vs. SCE)	0.071, 0.068	0.130, 0.134	0.115, 0.103	0.157, 0.155
Eg(P-R) (V vs. SCE)	0.053, 0.053	0.134, 0.136	0.095, 0.091	0.154, 0.153
E/g(P-F) (V vs. SCE)	0.076, 0.076	0.132, 0.136	0.122, 0.108	0.159, 0.156
E/g(P-R) (V vs. SCE)	0.060, 0.058	0.136, 0.139	0.101, 0.101	0.156, 0.155
ig(H-F) (μA cm−2)	55.1, 39.4	62.2, 58.4	21.1, 25.9	6.60, 31.8
ig(H-R) (μA cm−2)	44.1, 45.9	64.2, 48.4	33.8, 32.2	39.2, 43.2
i/g(H-F) (μA cm−2)	49.0, 36.3	59.6, 56.2	19.2, 23.8	5.70, 29.3
i/g(H-R) (μA cm−2)	41.9, 42.9	60.8, 47.3	31.1, 30.4	37.0, 41.7
ig(P-F) (μA cm−2)	22.0, 10.9	21.7, 20.6	17.7, 18.5	15.4, 17.1
ig(P-R) (μA cm−2)	22.6, 22.4	27.7, 28.0	23.9, 24.3	21.0, 21.3
i/g(P-F) (μA cm−2)	18.9, 16.8	18.2, 17.7	13.7, 15.0	12.0, 14.1
i/g(P-R) (μA cm−2)	19.7, 19.6	23.4, 23.2	19.0, 18.4	20.0, 17.0

* The unprimed (E_g, i_g) and primed (E^/_g, i^/_g) parameters were obtained from IR_s-corrected and -uncorrected PDP plots, respectively. Parameters labeled with H-F and H-R were determined from PDP plots recorded under the stationary hold condition using forward and reverse LSV scans, respectively. The variables determined from forward and reverse LSV scans under surface polishing are labeled with P-F and P-R, respectively. The first and second values, separated by a comma, in each row represent results obtained from the first and second scan sequences of LSV, respectively.

. As seen in Table 3, the values of both the relevant galvanic variables, E_g and i_g, are affected by the choices of the direction and sequence of LSV scans, and all these values are further affected if the PDP plots are not corrected for the CMP slurries’ ohmic effect. As shown by other authors, additional changes in the LSV control parameters (such as ranges and rates of potential scans) can further affect the values of corrosion variables measured using PDP [115,126,127,128]. Since all PDP experiments in the current study were performed using rate- and range-optimized LSV scans (according to [39]), these additional effects were not explored here.

While multiple experimental factors can contribute to the effects of PDP control variables on corrosion measurements [117], a relatively straightforward explanation for these effects follows from the considerations of voltage-induced adsorptions-desorptions of reaction intermediates [129] in PDP. To illustrate this mechanism, we refer to Figure 10A, which schematically shows an LSV (PDP) scan in the forward direction (toward increasing voltages), starting at an initial electrode potential E_i at the instant t_i and ending at time t_f to reach the final potential, E_f, of this scan. The time dependence of the scanned voltage in the direction of increasing voltages can be expressed as

$$E\left(t\right)=E_i+vt$$

(27)

where $v$ is the potential sweep rate; $v=\mathrm{d}E/\mathrm{d}t$. To set up the LSV parameters experimentally, the OCP of the system is measured first, and the voltage bounds of LSV are chosen such that |E_OC − E_i| ≈ |E_f − E_OC|, with the applied voltage E crossing the value of E_OC approximately at a time (t_f − t_i)/2 after starting the voltage sweep. Figure 10B shows a schematic Tafel plot generated by the LSV scan in Figure 10A, where I_i and I_f are the net electrode currents measured at the start and the end of LSV, respectively. The cathodic and anodic Tafel branches are formed, respectively, during the first and second halves of the LSV scan.

Using the variables of the cathodic Tafel region, it is possible to write the following:

$$E_{\mathrm{corr}}=E_{\mathrm{i}}+\left|{\left({\displaystyle \frac{\partial \ln I_{\mathrm{c}}}{\partial E}}\right)}^{-1}\right|\ln \left({\displaystyle \frac{I_{\mathrm{i}}}{I_{\mathrm{corr}}}}\right)$$

(28)

which connects E_corr with E_i in the first half of the voltage sweep. Equation (3) can be utilized to show that in the absence of voltage-dependent adsorptions of reaction intermediates,

$$\left({\displaystyle \frac{\partial \ln I_{\mathrm{c}}}{\partial E}}\right)={\displaystyle \frac{-1}{{\beta }_{\mathrm{c}}}}$$

(29)

which essentially corresponds to the definition of the cathodic Tafel slope. Under this condition, Equation (28) yields the usual form of E_corr expected [from Equation (11)] following the LSV scan depicted in Figure 10: $E_{\mathrm{corr}}=E_{\mathrm{i}}+{\beta }_{\mathrm{c}}\ln (I_{\mathrm{i}}/I_{\mathrm{corr}})$, which is the same as E_OC. Equation (29) corresponds to the case of an ideal Tafel plot, where the intersection of the cathodic and anodic current branches match with the value of E_OC. As illustrated below, this does not happen in the presence of voltage-induced adsorption/desorption of reaction intermediates occurring during LSV.

If cathodic intermediates adsorb during PDP, it is shown in the Supporting Material that, in this case,

$${\displaystyle \frac{\partial \ln I_{\mathrm{c}}}{\partial E}}=\left[{\displaystyle \frac{1}{{\theta }_{\mathrm{c}}}}{\displaystyle \frac{\partial {\theta }_{\mathrm{c}}}{\partial E}}\right]-{\displaystyle \frac{1}{{\beta }_{\mathrm{c}}}}$$

(30)

which indicates a change in the expected Tafel slope that occurs during the cathodic section of LSV due to a voltage-dependent ${\theta }_{\mathrm{c}}$. The sign of the derivative on the left-hand side of Equation (30) depends on the relative values of the two terms on the right-hand side, which depend on the sign and the magnitude of $(\partial {\theta }_{\mathrm{c}}/\partial E)$. Thus, if voltage-induced adsorption of reaction intermediates operates, the condition specified in Equation (29) for the equivalence between E_corr and E_OC changes to the form in Equation (30). Likewise, in the anodic section (E_OC < E < E_f) of the LSV scan, an altered form of the anodic Tafel slope can be obtained in a similar approach.

In a reversed LSV scan (going from E_f to E_i), the anodic Tafel slope will be affected first to shift the value of E_corr with respect to that of E_OC, followed by a change in the cathodic Tafel slope. These combined changes in the Tafel slopes are responsible for shifting the corrosion current density from its value for an ideal Tafel system. If the adsorptions occurring in the first half of an LSV scan initiate the main shift of E_corr, then the direction and the amount of this shift will be dictated by the direction of the LSV scan. Furthermore, these adsorption-mediated effects of voltage-activated reactions often contain some quasi-reversible and/or irreversible components, which prevent the sample surface from returning to its previous state in repeated LSV cycles. In the presence of these irreversible effects, a Tafel plot recorded in each LSV scan is unlikely to coincide with another sequentially recorded plot. These effects of scan direction and scan sequence are demonstrated in the LSV data presented in Figure 8 and Figure 9, as well as in Table 3.

A further effect of LSV scan variables on recorded polarization plots is that of the voltage scan speed, $v$. The effects of scan rate variations were not separately studied in this work since all LSV data were collected at an optimized value of $v$. The protocols for this optimization process have been described in our earlier work, where it was observed that optimization of the LSV scan rate would minimize voltage-induced adsorption effects [39]. For both the Cu and Co samples studied here, all the polish-activated plots exhibit rather insignificant variations with respect to repeated cycles and with respect to the reversal of scan direction, while the corresponding data for the stationary hold cases do not contain this feature (Figure 8 and Figure 9 as well as related Figures S3–S6 in the Supplementary Materials). An examination of the E_g and i_g values listed in Table 3 for the hold and polish conditions also shows how these values are less affected by the LSV control variables in the presence of surface polishing than in the absence of polishing. These are the expected results of optimizing the LSV scan speed under the polishing conditions.

While considering the effects of LSV scans on corrosion variables, it is useful to note that the local pH values at certain electrode surfaces have been found to change under strong polarization [130,131,132]. Most electrochemical systems reported in this category are quite different in their compositions and faradaic characteristics compared to those of the typical metal-CMP interfaces. To our knowledge, such effects of polarization-induced pH swing have not been published in the commonly available electrochemical literature on metal CMP systems. However, if the local pH at a CMP surface varies while probing the surface with LSV, the exchange current densities Equation (8) can be affected, which, in turn, can make the E_corr values obtained from these measurements differ from those of E_OC recorded in the absence of polarization. Checking for such an effect in a CMP situation would require in situ monitoring of pH at the CMP interface.

Measuring local pH at an electrode surface requires scanning probe techniques or surface-enhanced Raman or infrared spectroscopies [133]. The presence of strong hydrodynamic and tribological effects at a CMP interface can significantly complicate the application of these techniques for surface pH measurements. Thus, detecting and resolving any active effects of local pH variations on the LSV-measured values of corrosion variables can be a nontrivial task even while using model systems simulating the polishing conditions of CMP. Any pH-induced effects operating in such cases would remain mixed with those of potentially dependent adsorbate coverages in the LSV-measured values of corrosion variables. However, as we demonstrate later in this report, the artifacts in both these categories can be substantially minimized by measuring E_corr with LSV under active surface abrasion where tribology dictates or overrides certain features of CMP chemistry.

4.5.4. Effects of Open Circuit Potential Drifts

Since E_OC is theoretically equivalent to E_corr, the galvanic polarization term, ΔE_corr, in Equation (15) can be evaluated as ΔE_OC. The latter approach is often preferred to PDP-based measurements as a means to avoid surface modifications due to externally applied voltages. However, a major difficulty in recording stable OCPs for CMP samples in reactive slurries is the transient nature of these OCPs, which is commonly observed under stationary hold conditions. The temporal instability of E_OC in these cases is caused by cumulative effects of sustained surface processes, including deposition and/or dissolution of slurry species and products of surface reactions. The effects contributing to OCP instabilities in unstirred solutions under static conditions may include some changes in the local surface pH [134], as well as surface charge-induced ion adsorptions based on relative values of the CMP sample’s potential of zero charge and the OCP [135]. Furthermore, the surface of a CMP metal, in the absence of abrasion, develops metal-oxide and/or -hydroxide layers [Equations (18), (19) and (23)]. These oxidation products modify the double layer structure, which, as a result of the Frumkin effect, can destabilize the OCP of a static interface [136].

Due to the drifting nature of OCPs developed at stationary metal-slurry interfaces, it is difficult to quantify specific single values of ΔE_OC for studying such CMP-related bimetallic systems. Demonstrative results of intermittent OCP transients recorded under alternated hold (H) and polishing (P) conditions are presented in Figure 11. Here, panels A–D correspond to data collected in slurries I–IV, with two alternately repeated 4-min long sequences of H and P employed for each slurry. The potential transients for both Cu and Co are plotted together in each panel to show how the OCP difference between the two metals changes with the application and withdrawal of surface abrasion. The effects of polishing on E_OC(Cu), especially those seen in the MA-free slurries I and III, are small compared to corresponding changes observed for Co. This observation is consistent with the relatively low MRRs of Cu (Figure 3) and follows the correlations between MRRs and the OCP differences, |E_OC(P) − E_OC(H)|, that have been recently reported for certain CMP systems [40]. Accordingly, the finite values of |E_OC(P) − E_OC(H)| for Cu found in the MA-based slurries II and IV indicate once again the active role of MA in surface complex formation for material removal from Cu.

The OCPs recorded in the hold situation, especially those for Co, show considerable temporal drifts. According to Equation (8), these OCP drifts can be largely attributed to continued reductions in the value of θ_c due to cumulative accumulations of Co(OH)₂ at unabraded Co surfaces due to reaction (23). Evidently, the cathodic inhibition character of Co(OH)₂ dominates its surface passivating feature in the OCP situation. Under abrasion, the coverage of Co(OH)₂ at the Co surface decreases, and E_OC(P) increases. In the case of Cu, E_OC(H) > E_OC(P) indicates the effect of unabraded CuMal as an anodic inhibitor at the Cu sample surface. While polishing, E_OC for both Cu and Co reach nearly steady state values as the cumulative effects of surface reactions are mostly eliminated by a balance between chemical formation and mechanical removal of the reaction products described in Equations (22) and (23).

The value of a drifting OCP depends to some extent on the time when it is monitored within a transient interval. On the other hand, the full stabilization of OCPs in these measurements generally exceeds the typical polish times of CMP, which makes it difficult to emphasize the use of stabilized OCPs for checking CMP-related galvanic corrosion. To examine the general time dependence of the data in Figure 11, we use a strictly phenomenological approach by focusing on the values averaged from the first two minutes and last two minutes of the OCP profiles recorded under hold and polishing conditions. These slurry-dependent results are summarized in Figure 12B and Figure 12C (for the initial and final two minutes of data collection, respectively). For comparison with these OCP data, the corresponding values of E_corr taken from Table 2 for the first LSV scans are included in Figure 12A.

The values of E_OC measured for Co under stationary hold exhibit the most variations with respect to the initial and final stages of data recording. The disparities between the values and the slurry-dependent trends of E_OC and E_corr are also most notable for this case. The E_OC(P) data for Cu are relatively more consistent within the timescales of comparison as well as with their E_corr counterparts. Thus, the data-averaging scheme provides a reasonable approach to comparing OCP values with their corresponding E_corr values for abraded samples. These data underscore again the difficulty of quantifying the values of CMP-related corrosion variables measured without applying surface abrasion.

4.5.5. Cathodic Coupling of Anode Metal in a Galvanic System

Equation (15) assumes the case of strong galvanic polarization, where the cathodic currents of the cathode metal are balanced by the anodic currents of a coupled anode metal. The dissolution current density of the corroding anode can be readily equated to i_g in this case. Nevertheless, CMP-related bimetallic systems often exhibit a complex behavior where both anodic and cathodic currents of the anode are coupled to the cathodic current of the cathode under low galvanic polarization. The anode’s dissolution current density, $i_{\mathrm{d}}\left(A\right)$, in this case, is no longer equal to $i_{\mathrm{g}}\left(A\right)$, but takes the following form [30]:

$$i_{\mathrm{d}} (A)=i_{\mathrm{g}}(A)\left[1+{\displaystyle \frac{i_{0\mathrm{c}}\left(A\right){\theta }_{\mathrm{c}}\left(A\right)}{i_{0\mathrm{c}}\left(C\right){\theta }_{\mathrm{c}}\left(C\right)}}\right]$$

(31)

which reduces to Equation (15) only if [$i_{0\mathrm{c}}\left(A\right){\theta }_{\mathrm{c}}(A$)] << [$i_{0\mathrm{c}}\left(C\right){\theta }_{\mathrm{c}}(C$)].

Examples of bimetallic corrosion in the category of those characterized by Equation (31) can be found in Figure 2 and Figure 3, respectively, in our previous studies of Ru-Cu [16] and Cu-Mn [92] CMP systems. The simple determination of $i_{\mathrm{g}}$ from a single intersection point between the cathode’s cathodic current and the anode’s anodic current does not account for the actual rate of galvanically activated anode dissolution in such cases. A more detailed analysis of the PDP data implementing the considerations of Equation (31) becomes necessary to determine the anode dissolution currents under these conditions. The protocols necessary for this analysis have been discussed previously [137].

4.5.6. Effects of Microstructure, Morphology and Sample Composition

Corrosion characteristics of a material are strongly determined by the latter’s microstructures and compositional details [138,139]. Thus, the values of corrosion variables for metal samples used in potentiodynamic measurements have been shown to be sensitive to the samples’ constituent grain sizes [140]. Since different types of samples (thin films deposited by different techniques and polycrystalline discs/coupons) are used in the experiments reported by different groups, these results cannot be readily unified in an established set of guidelines to assess galvanic corrosion characteristics of barrier-metal bimetallic combinations that exist in actual interconnect structures.

Characterizing the microstructural factor of CMP-induced galvanic corrosion would require tribo-electrochemical measurements like those discussed in this report using different CMP samples of varied microstructures. Nevertheless, as we have discussed elsewhere, there are certain practical difficulties for using such test samples to measure tribo-electrochemical parameters in the presence of mechanical abrasion. Specifically, while using the experimental setup (used in the present study and) described in Reference [29], the nonconducting substrates commonly used to deposit blanket thin films do not allow for a simple method to electrically access the moving sample under abrasion. Due to this reason, polycrystalline samples with electrically conducting front and back surfaces are employed for the tribo-electrochemical measurements conducted in our current approach. While these polycrystalline metal samples can mostly mimic the electrochemical response of the metal’s thin-film version, they are limited for studying the morphological features that are governed by the microstructural details of thin films.

Aside from its microstructure, the roughness of a corroding surface is another morphological property of the latter that has been found to affect the spread/rate of static surface corrosion for certain systems [141,142,143]. However, the pre-CMP surface roughness features of metal films exist at relatively low levels in most cases (~5–8 nm) [144,145], and in efficient CMP processes, this roughness rapidly drops to atomic scale values within a short interval (<60 s) during which the surface is polished. As indicated in Figure 1, galvanic corrosion in Cu CMP is activated near the end of the CMP process when the polished Cu surface arrives at the same level as the barrier line. This Cu profile is considerably flatter than the pre-CMP Cu profile. To our knowledge, there is no experimental evidence in the CMP literature to indicate that this fast-evolving and essentially flat CMP surface can affect the galvanic corrosion characteristics of CMP. A detailed discussion of this topic is beyond the main scope (indicated in the title) of the present work. Nevertheless, since the effects of roughness-induced surface corrosion are often discussed in the general literature of corrosion engineering, the implications of such effects in the context of CMP are briefly addressed in the Supporting Material.

4.6. Estimating Cutoff Values for CMP-Related Galvanic Corrosion Variables

The results of this investigation illustrate certain practical difficulties of using PDP-based measurements in a comprehensive analytical framework for accurately assessing CMP-related galvanic effects. This limitation of PDP experiments using uncoupled metals has been specifically noted in the earlier literature on galvanic corrosion [146]. As illustrated here, measurements of OCP transients also may not be immune to misleading conclusions for similar applications. However, while designing slurry formulations for material selective barrier CMP, it is generally necessary to check how the slurry chemistries would affect the galvanic corrosion characteristics of the Cu/barrier contact regions in an interconnect structure. Therefore, while operating within the limits of the commonly used experimental methods, it is useful to set up a working framework to estimate the galvanic corrosion variables for CMP systems. Certain phenomenological considerations for adapting this approach are briefly discussed below, focusing on the measurements of the variables, i_g(A), ΔE_corr and ΔE_OC.

In view of the system-dependent and measurement-specific variabilities, it is difficult to standardize a specific threshold of “tolerable” galvanic currents in the CMP context. In our earlier investigations, we have used relative values of i_g (anode) and i_corr (anode) to set up a scale for screening slurry formulations for galvanic corrosion [15,16]. For a CMP metal, the value of i_corr indicates how the metal’s CR contributes to material removal. If this metal becomes an anode in a bimetallic couple, the metal’s i_g value provides a measure of the rate at which galvanic defect sites are generated at the metal surface. From this observation and using a strictly phenomenological approach, it is reasonable to avoid slurry formulations that would yield i_g(anode) ≥ i_corr(anode). The comparative values of i_corr(Co) in Figure 6A and i_g(Co) in Figure 7B follow the guideline of this strategy for reducing galvanic corrosion. Slurry designs based on this approach would generally aim at lowering the value of i_g(anode) to maximize the gap, [i_corr(anode) − i_g(anode)].

To discuss the galvanic polarization terms, ΔE_corr and ΔE_OC, we summarize in Figure 13 the values of these parameters measured here under different experimental conditions. Figure 13A compares the values of ΔE_OC obtained from the OCP data shown in Figure 12B,C for the stationary hold (H) sample configuration. Figure 13B presents the corresponding results taken from Figure 12B,C for the samples undergoing surface polishing (P). For ease of comparison, the values of ΔE_corr taken from Table 2 for the different test systems are also included in Figure 13. Although the parameters, ΔE_OC (initial), ΔE_OC (final) and ΔE_corr are theoretically equivalent, their values measured under the hold condition are notably different.

The variations observed between the values of ΔE_corr and ΔE_OC are considerably suppressed for the abraded metal samples in Figure 13B, where the values of ΔE_OC (initial), ΔE_OC (final) and ΔE_corr are mutually comparable and consistent within a narrow range for each of the slurries examined. Evidently, the equilibrium surface coverages of reaction intermediates under abrasion are largely determined by the mechanically controlled rates of adsorption and desorption of these species. If the voltage-dependent variations of θ_c and θ_a are much weaker than these tribological effects, the terms (∂θ_c/∂E) and (∂θ_a/∂E) should have negligible contributions to the values of E_corr measured with LSV. At the same time, if any effects of local pH variations operate under static conditions, they would be suppressed under surface abrasion through a combination of hydrodynamic effects and adequately sustained mixed potential reactions. Therefore, to examine slurry characteristics for suppressing galvanic corrosion of Co, we focus on the data in Figure 13B collected under these conditions of surface abrasion.

In the general field of corrosion engineering, ΔE_corr ≤ 0.1 V is often considered as a requirement for suppressing bimetallic corrosion [147], while for certain applications, this threshold voltage is also taken as 0.2 V. In the context of CMP for microelectronics applications, a more stringent limiting value of ΔE_corr is generally considered. Using the concept of the Arrhenius equation, this value can be taken as the thermal voltage, E_th = RT/F. If 40–55 °C is taken as a typical range of sample temperatures under polish [148], then E_th ~0.03 V. Previous studies of different CMP systems in our laboratory have reported ΔE_corr values recorded under polishing conditions that were below this estimate of E_th [14,15,105,106]. According to Figure 13B here, the Co-Cu system in the MA-based slurries II and IV essentially meets the requirement for minimizing galvanic corrosion as specified by the value of E_th.

5. Conclusions

The results presented in this work examine certain electrochemical protocols and their practical constraints for determining the primary indicators of CMP-related galvanic corrosion. Employing CMP samples of Co and Cu in malonic acid-based slurries, we focus this investigation on the ex situ measurements of PDP and OCP that are most used for studying such galvanic systems. By integrating the mechanical component of CMP in the framework of these electroanalytical techniques and by comparing the electrochemical data recorded with and without surface abrasion, we demonstrate that tribology plays a critical role in properly bringing out the implications of the experimental results for studying CMP-specific galvanic corrosion.

Aside from facilitating the analysis of galvanic corrosion effects, the results of this study provide relevant information to determine the tribo-electrochemical mechanisms of material removal in CMP. Hydrogen peroxide from SPC oxidizes the Co and Cu surfaces, and subsequent reactions of the complexing agent, malonic acid, constitute the chemical component of CMP. While acting as a surface complexing agent for Cu, malonic acid is found to suppress the polarization for galvanic corrosion of Co in the CMP slurry. The predominant mechanisms of material removal for both Co and Cu are identified as tribo-corrosion and corrosion-induced wear, while corrosion-like surface reactions have relatively smaller contributions to the overall rate of material removal. At the same time, these reactions are critical for activating all the operative modes of material wear and removal. The information collected in this way about the CMP mechanisms facilitates the task of designing slurry formulations, including the considerations for MRRs as well as those for minimizing (general and) galvanic corrosions.

For the Co-Cu bimetallic system tested in this work, the indicator variables of galvanic corrosion determined by both the PDP and the OCP techniques strongly depend on the presence or absence of surface abrasion for CMP. The results suggest that the data collected under abrasion should be more reliable in the practical context of CMP. Certain practical constraints of traditional (abrasion-free) electrochemical measurement may remain measurably operative in the tribo-electrochemical experiments. For instance, the direction and sequence of a PDP scan may affect the measured parameters of galvanic corrosion. Likewise, the duration of OCP stabilization on a comparative scale, with the temporal phase and length of OCP measurement, may also affect the values being monitored. These effects of experimental control variables on the measured values of corrosion parameters have been explained in terms of the voltage and time-dependent adsorption/desorption of reaction intermediates. Certain protocols have been suggested here to optimize the experimental control variables linked to the (tribo-)electrochemical assessments of CMP-specific galvanic corrosion effects.