- Email: [email protected]
Modeling the stability of electroless plating bath—diffusion of nickel colloidal particles from the plating frontier
Journal of Colloid and Interface Science 262 (2003) 89–96 www.elsevier.com/locate/jcis
Modeling the stability of electroless plating bath—diffusion of nickel colloidal particles from the plating frontier X. Yin, L. Hong,a,∗ B.-H. Chen,b and T.-M. Ko c a Department of Chemical and Environmental Engineering, National University of Singapore, 10 Kent Ridge Crescent, 119260 Singapore b Department of Chemical Engineering, National Cheng Kung University, 1 University Road, Tainan 70101, Taiwan c IBM Microelectronics, 2070 Route 52, Mail Stop E40, Hopewell Junction, NY 12533, USA
Received 16 May 2002; accepted 11 February 2003
Abstract Electroless nickel (EN) plating is a process in which Ni2+ ions are reduced by hydrogen atoms adsorbed at a fresh Ni surface. However, detaching of a handful of tiny Ni metal particles from a substrate causes the entrance of these particles into the plating solution. The metal particles offer very reactive surfaces for the reduction of Ni2+ ions, which in turn aggravates the detachment, causing a self-accelerated cycle. Eventually the plating solution will be subject to an overwhelming precipitation of Ni black. This paper proposes a one-dimensional diffusion model to explain the dependence of the bath stability on the plating time under different bath loadings. This mathematical model contains Vd , defined as the decomposition volume, a measure to judge chemical stability of a plating solution. To obtain Vd experimentally, a PdCl2 solution was purposely introduced into a model solution (the addition leads to immediate generation of metal particles) until the very moment of onset of massive deposition of colloidal Ni. The Vd data from the experiment were then used to perform simulation in order to complete the model proposed. Other than the effects of bath loadings and plating time, an adsorption model was also created to describe the temperature effect. To coordinate the adsorption model, l-cysteine was used as an adsorbate that plays a deactivation role. The under bump metallization process on patterned silicon wafers has been used to support the main theme of this study. 2003 Elsevier Science (USA). All rights reserved. Keywords: Electroless nickel plating; Metal colloidal particles; Diffusion; Solution stability; Modeling
1. Introduction
2+ 2H2 PO− 2 + 2H2 O + Ni
−→ Ni0 + 2H+ + 2H2 PO− 3 + H2 . Cat.
Since its discovery by Brenner and Riddell in 1946 [1], the electroless nickel (EN) plating technology has been experiencing fast development because of the demand for various surface engineering processes [2,3]. Recently the application of the EN plating technology in the microelectronicpackaging industry, i.e., metallization of semiconductor chips to create via/conductor hole interconnections [4,5] and chip-on-board solder bumps [6,7], requires more precise understanding of the chemistry of plating. The constituents of an EN solution include a Ni(II) salt, a reducing agent, suitable metal coordination ligands, stabilizers, and additives for particular needs. The overall reaction in an acidic hypophosphite bath can be written as * Corresponding author.
E-mail address: [email protected] (L. Hong).
(1)
The early studies on the chemical mechanism of this reaction suggested that the anodic process (the oxidation of hypophosphite) is a slow step [8]. This viewpoint was further consolidated within the past decade typically through the work of Jusys et al. [9] with electrochemical mass spectra and the work of Abrantes et al. [10] with probe beam deflection. Their results all come to the conclusion that the homolytic cleavage of phosphorus–hydrogen bonds of the adsorbed hypophosphite anion on nickel metal is a catalytic and rate-determining step. Most investigations supported the view that atomic hydrogen rather than hydrogen molecules is the real reducing agent of nickel ions [11–13]. As a matter of fact, the EN plating solution is a kinetically stabilized system, that is, Ni2+ ions cannot be reduced even at the plating temperature (∼90 ◦ C) as long as there is not a solid surface at which the P–H bond can be broken down to supply hydro-
0021-9797/03/$ – see front matter 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S0021-9797(03)00191-7
90
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
gen atoms. It is now known that only a few types of metals possess catalytic activity for the cleavage of P–H bonds; they include Pd, Ni, and Co [14]. Therefore as soon as the first layer of Ni metal atoms is deposited, the P–H bonds can undergo dissociation on it to yield hydrogen atoms, and the reduction of Ni2+ ions afterward becomes self-sustainable. An EN plating bath without a proper stabilization system will produce a lot of fine nickel flakes shortly after the plating is initiated. There has been a notable lack of insight into the origin of this phenomenon. On the basis of the present investigation, we propose that the detachment of colloidal nickel particles from the plating frontier, followed by their diffusion into the bulk of the solution, is responsible for the quick growth of Ni flakes. This is due to the presence of many coordination-unsaturated Ni atoms on these tiny metal nuclei; they accelerate homolysis of the P–H bond. The Ni deposition gains momentum at these particles more than on the desired plating object. The growing particles themselves will soon become the parents of new colloidal particles. As a result, the plating bath becomes contaminated due to this “self-propagation reaction.” Usually, an EN plating bath contains a trace amount of stabilizers. Most of the stabilizers show the common identity of the poisons with the catalysts of either hydrogenation or dehydrogenation reactions [14]. Those that have been identified to be effective stabilizers fell into the following categories: (a) Compounds of group VI elements, such as S, Se, and Te. − 2− (b) Some oxide anions, such as AsO− 2 , IO3 , and MoO4 . (c) Heavy metal ions, such as Sn2+ , Pb2+ , Hg+ , and Sb3+ . Several methods have been developed to assess the effectiveness of stabilizers, such as plotting the deposition (mixed) potential of EN solution vs the concentrations of a stabilizer [15], measuring the deposition rate vs the stabilizer concentration, and testing the buffering capacity of an EN plating bath during an increase in the concentration of metal nuclei by deliberately introducing a PdCl2 solution (0.1 g/l) into the bath until decomposition takes place. Any measure that is capable of restricting the generation of Ni metal particles in an EN solution contributes to the stabilization of the bath. Among the stabilizers listed above, sulfur-containing organic compounds and heavy inorganic cations are most often used. The former type would form Ni–S adsorption bonds, which are believed to impair the catalytic activity of Ni atoms to break down the P–H bond [16,17], and are regarded as anodic-type stabilizers. The second type may function as blocks to the occupancy of those catalytic reactive Ni sites [18,19]. In this work, l-cysteine was employed as an anodic-type stabilizer. This paper proposes a mathematical model to predict the dependence of bath stability on bath loading, plating time, and bath temperature. In this model the decomposition volume Vdmod is defined to express the stability of an EN plating
bath. Experimentally, the decomposition volume was measured by adding an aqueous solution of PdCl2 into the EN solution under investigation at a controlled rate. Unlike Ni2+ ion, Pd2+ ion could be reduced immediately by the hypophosphite ion present in the solution at around 40 ◦ C. The resulting Pd metal particle is an excellent catalyst of the homolysis of H2 PO− 2 that supplies hydrogen atoms. Therefore, Ni deposition will take place forthwith on the Pd particles as long as the catalytic reactivity of the surface Pd atoms is not inhibited by a stabilizer. The volume of the PdCl2 solution consumed at the moment when the EN solution turns into black at a given temperature is taken as the experimenexp tal decomposition volume Vd . Unless specified separately, the volume of EN solution and the concentration of PdCl2 solution used for this study are fixed. exp After having Vd value ready, we started to run a selfconsistent procedure that minimized the difference between exp Vdmod and Vd . In this way, the mathematical model could be specified with respect to a particular set of stabilization conditions implemented in the plating solution. In parexp allel, the dependence of Vd on plating temperature was also investigated using the Gouy–Chapman–Stern–Grahame (GCSG) method [20]. This paper focuses on the diffusion of Ni metal colloidal particles from the plating frontier and attempts to describe its effect on the instability of an EN bath by the two simplified mathematical models.
2. Experimental 2.1. Pretreatment and electroless nickel plating Two different kinds of aluminum substrates supplied by Chartered Semiconductor Manufacturing Ltd. (Singapore) were used in this study; they are E-beam evaporated Al (ca. 1 µm thickness) on silicon wafer and Al bond pads on chip (size 80 × 80 µm, pitch 155 µm). The EN plating cannot take place directly on the aluminum substrates because there is an Al2 O3 thin film on top. To make the aluminum surface ready for plating, a double zincation process is necessary. Commercial zincation solution (Plaschem, Singapore) was employed. After soaking in it, a zinc layer replaced the oxide layer on the Al base: − 0 − 3Zn(OH)2− 4 + 2Al → 2Al(OH)4 + 3Zn + 4OH .
(2)
The EN solution was formulated in house; its composition and the plating conditions are listed in Table 1. The Zn layer was etched out in the acidic environment of the EN bath, which was immediately followed by nickel deposition. The plating time was varied in the range 20–60 min according to the needs of the study. After that, the substrate was taken out and the measurement of Vd was conducted in the solution that remained.
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
91
Table 1 EN bath composition and plating conditions NiSO4 ·6H2 O NaH2 PO2 ·H2 O NaAc DL-Malic acid Lactic acid (90%) Borax Cysteine Temperature pH (adjusted with NaOH) Plating time
26 g/l 28 g/l 10 g/l 15 g/l 15 ml/l 6 g/l Variable 88 ◦ C 4–5 20 min
2.2. Stability measurement An aqueous solution of PdCl2 was prepared by dissolving 0.025 g of PdCl2 in 0.25 ml concentrated hydrochloric acid. The resulting solution was diluted by deionized water to 1 l in a volumetric flask. The decomposition volume Vd was determined by introducing the PdCl2 solution at a rate of 0.05 ml/s into 20 ml of plating solution, keeping the temperature unchanged. Sufficient magnetic stirring was upheld to dissipate any concentration gradient due to the addition of PdCl2 . The Vd is the volume of PdCl2 solution consumed at which the bath suddenly turns black, a sign of large precipitation of metal colloids. It was observed that the Vd alters with the use of different values of the plating loading (PL). The PL is defined as the ratio Aobject/Vbath , where Aobject is the area of the object to be plated in dm2 and Vbath is the volume of plating solution in l. Therefore, PL has unit in dm2 /l. 2.3. Surface morphology and composition measurements The surface morphology of the deposited films was examined by scanning electron microscopy (SEM) (JEOL JSM5600LV). X-ray photon spectroscopy (XPS) was carried out on an instrument (Kratos Axis His, Manchester, UK) with an AlKα X-ray source (at 1486.6 eV) at and 90◦ take-off angle. The pass energy is 40 eV for high-resolution scans.
3. Theoretical background 3.1. The basic assumption for creating the Helmholtz plane in an EN plating system The concept of the Stern–Grahame (S–G) double layer is applied to describe the interface created when a piece of metal, i.e., the Al used in this study, is placed in an EN plating solution. For an active metal, a negatively charged surface in contact with the aqueous solution is expected due to the leaving of a small number of Al atoms from the metal surface in the form of ions, which is regarded as one of the origins of electrode potentials [21]. An adsorption layer consisting of hydrated nickel ions and possibly other complex ions of nickel is to be formed proximately at the fresh surface of Al. This adsorption layer is defined as the inner
Fig. 1. A schematic illustration of the electric double layer likely to form in an electroless nickel plating bath and the structure of the plating frontier.
Helmholtz plane (IHP) [22,23]. Beyond this layer, there is a thicker layer comprising various anions, defined as the outer Helmholtz plane (OHP) (as illustrated by Fig. 1). The electroless plating is essentially an autocatalytic process as elaborated in the preceding section. Following this mechanism, the Ni atoms upon formation on IHP will join the Ni assembly at the surface. However, it could also happen that a handful of Ni metal clusters diffuse outward into the solution through the OHP. The spinning-off depends most probably on the local concentration of Ni–H species, which is governed, in turn, by the coordination environments of the Ni atoms at the surface of the substrate (the plating frontier). If there are a larger number of coordinationunsaturated Ni atoms in a specific geometry, a higher local concentration of Ni–H species will be present, which offers a higher chance of formation of Ni metal clusters (nuclei) because the reducing rate is higher than the packing rate at which Ni atoms find their way into the coating layer. Furthermore, an electrical double layer will be formed surrounding those detached Ni metal nuclei while they diffuse through the S–G layer. Charges built up on these nuclei due to the adsorption of ions would affect their diffusion behavior by electrostatic interactions among the charged particles [24,25] and between them and the solution. The magnitude of the interaction depends upon the potential around a particle. In the present case, the charged adsorption layer on a Ni nucleus could be considered insignificant before it moves into the bulk of the plating solution from the S–G layer. The structure of the S–G layer backs this assumption; both the concentration and activity coefficient of anions in the Gouy layer located between the IHP plane and the OHP plane (the shear surface) are assumed to be lower than in the bulk of solution [23], for they constitute the Stern plane, which is directly in contact with the Ni plating frontier, and as a result of this, cations make up the majority in the Gouy layer. On the other hand, it is expected that the hydration layer surrounding the cations would obstruct their adsorption onto
92
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
the Ni nuclei. As such, the solution to Poisson’s equation for low potential and low ion concentration [22] applies, in which the potential is expressed by q −r/κ −1 e , ΨNi-nucleus = (3) 4πεr where κ −1 characterizes the “thickness” of the adsorption layer on the particle, q the surface charge, and r the radius of the sphere, including the adsorption layer. Based on the above assumption, r κ −1 and q is a small value, which renders the potential around a Ni nanoparticle negligible. As a result, the effect of the charge adsorption layer on the Ni nucleus is omitted in the following mathematical treatment that describes the diffusion behavior of these tiny particles within the S–G layer. Fick’s second equation is applicable to this scenario, ∂C = ∇ · D∇C, (4) ∂t where C is the concentration of Ni nuclei and D is the diffusivity constant. For a planar plating frontier (one-dimensional assumption) with a constant diffusivity, it has the form ∂C ∂ 2C =D 2. ∂t ∂x The initial and boundary conditions are selected as C(x, 0) = 0 C(0, t) = C0
(5)
for t = 0, for t > 0 and x = 0,
where C0 is assumed to be the concentration of Ni nuclei at IHP. Applying these two conditions, the expression for C as a function of x and t is obtained [26]: x x C = C0 1 − erf √ (6) = C0 erfc . 2(Dt)1/2 4Dt Although the concentration of ions in an EN bath increases steadily with plating, it is rationale to assume that the S–G layer undergoes insignificant changes over the period of study because it is a shorter duration compared with the time (4 h) required normally for completing one metal turnover (the time required to deposit all the Ni ions initially present in the solution). Therefore, we can consider the thickness of the double layer (L = IHP + OHP) to be independent of the plating time in the present system: C(L, t) = Cbulk
at x = L.
Given that the concentration of Ni nuclei in the bulk of solution equals Cbulk at plating time t, the following equation is obtained by plugging the above boundary condition into Eq. (6): L . Cbulk(t) = C0 erfc (7) 2(Dt)1/2 Vd has been defined as the volume of the PdCl2 solution added into an EN plating bath to test the buffering capacity of the solution to the concentration of Pd metal colloidal
particles. The metal nuclei in the bulk of the solution should come from two sources: (1) reversed diffusion from OHP of the plating substrate, which is expressed by the product aCbulk, where a is a parameter describing the reactivity of the Ni nuclei originated from the interface, and (2) reduction of Pd2+ ions that are purposely introduced into the solution. To describe the effect of the Pd particles on failure of the plating solution, the minimum volume b of the PdCl2 solution that is needed to trigger the solution decomposition in an unloaded EN bath (the absence of a plating substrate) must be defined. As a short plating time was used in the present study, Cbulk is considerably smaller than b, and so is the aCbulk value. Therefore, the dependence of the decomposition volume Vd on Cbulk can be expressed approximately as L +b Vd = aCbulk + b = a C0 erfc 2(Dt)1/2 K2 = K1 erfc 1/2 + b, (8) t where K1 = aC0 and K2 = L/(2D 1/2 ). According to the physical meaning aforementioned, parameter a must be negative and b greater than zero. Equation (8) gives the dependence of the decomposition volume (Vd ) on the plating time (t). 3.2. The temperature effect on the decomposition volume The Gouy–Chapman–Stern–Grahame (GCSG) model [20] is utilized as a theoretical model to describe temperature effects on the adsorption of l-cysteine molecules on reactive metal colloidal particles. The adsorption of l-cysteine molecules at reactive surface sites of colloidal metal particles will prevent dissociation of the P–H bond that produces hydrogen radicals required for reducing Ni2+ ions. Increasing the plating bath temperature would cause desorption of l-cysteine and therefore increase the reactivity of colloidal metal particles, which brings about instability of the bath. The GCSG equation has the form 1/2 V (x0 ) 2kT π exp − . n = n0 − 2
(9) ∂ V kT 2 ∂x
x=x0
Here n is assumed to be the number of l-cysteine molecules adsorbed at the interface between metal particles and liquid, n0 is the number of the same kind of molecules in the solution, k is the Boltzmann constant, T (K) is the temperature of plating bath, V (x) is the average electric potential of an ion (adsorbate) at distance x from the planar interface (adsorbent), and x0 is the position with a minimum V (x). Equation (9) can be rewritten as 1/2 2kπ V (x0 ) n = n0 − 2
T 1/2 exp − ∂ V kT ∂x 2 x=x0 B2 , = n0 B1 T 1/2 exp (10) T
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
93
or ln
B2 n , = ln B1 + 0.5 ln T + n0 T
(11)
in which B1 and B2 must have values >0 from the perspective of their physical meanings. 4. Results and discussion (a)
4.1. Dependence of the decomposition volume on the plating time On the basis of the typical experimental conditions given in Table 1 and in Section 2.2, the b value in Eq. (8) could be determined experimentally. Here a Z value is defined in order to minimize the discrepancy between the theoretical decomposition volume (Vdmod ) defined in Eq. (8) and the exexp perimental decomposition volume (Vd ) that was obtained under various plating loadings (defined in Section 2.2) and exp plating times, where N represents the number of Vd values taken with varying time and PL, respectively: Z=
N mod exp V − Vd i = f (K1 , K2 ). d
(12)
i=1 exp
In the simulation, the Vdmod will converge on Vd ; a set of most apt K1 and K2 values at a given temperature can then be figured out. The Z value is minimized by using the Nelder–Mead method [27,28]; the minimization is conducted using MATLAB 5.3 with a tolerance value of 1 × 10−4 . The simulation used arbitrarily −1.2 and 2 as the initial values of K1 and K2 , respectively. Figure 2 shows the effects of plating time on the decomposition volume at plating loadings of 1.042, 2.083, and 4.17 dm2 /l, respectively. With respect to a particular platexp ing time, both Vd (dots) and Vdmod (solid line) decrease with increasing plating loading. A larger PL means a greater area of the plating frontier, and hence a greater number of Ni metal colloidal particles were released within the same period of time. The bath stability also decreases with the ex-
Fig. 2. The dependence of decomposition volume on plating time at different plating loadings (PL).
(b) Fig. 3. The change of parameters (a) K1 and (b) K2 with PL as a result of simulation.
tension of plating time due to the fact that a larger number of reactive nickel particles enter into the solution. In a referexp ence experiment, we found that the Vd value changes very little with time in the unloaded bath with other conditions kept unchanged. Compared to the loaded bath, this observation substantiates that the tiny particles coming from the plating frontier contribute to the instability of an EN plating bath. The one-dimensional diffusion model (Eq. (8)) is established on the assumption that the transport of Ni nuclei along the other two dimensions (y and z) is insignificant to the contribution of Ni particles in the bulk of plating solution. In light of this one-dimensional diffusion model, C0 stands for, in effect, the plane concentration of Ni metal particles at OHP. Figure 3a shows a nearly linear dependence of K1 upon PL, with K1 becoming more negative with increasing PL. Since K1 = aC0 (a < 0), this outcome suggests that C0 increases with increased plating loading, which is consistent with the trend expressed by the experiments shown in Fig. 2. In contrast, the K2 value decreases first and then lay out with the increase in PL as shown in Fig. 3b. Knowing that K2 = L/(2D 1/2 ), it is rational to propose that the initial decrease of K2 is due to the increase of the diffusivity value D versus a finite change in the thickness of the double layer. In Fig. 3b, a further increase of PL from 2 to 4 causes a negligible change in the K2 value, which implies that D does not follow C0 when PL is greater than 2 in the present system of investigation. It would be appropriate to think that, at higher levels of PL, a one-dimensional diffusion model may very likely depart from the real situation; namely, two- or three-dimensional diffusion makes the actual number of Ni particles entering the solution less than the one-dimensional model predicts.
94
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
Fig. 5. The plot based on the GCSG model which gives the dependence of the number of l-cysteine molecules adsorbed on Pd colloidal particles upon the testing temperature.
It turns out that the equation gives a straight line in the temperature range of study (Fig. 5). In the future, B1 and B2 values can be determined through the measurement of the n and n0 values by an instrumental method at different temperatures. 4.3. Under bump metallurgy (UBM)—an example of the reversed diffusion phenomenon As shown in Fig. 6, the under bump metallization is completed via several steps including EN plating. The Al pad has a square shape (∼80 × 80 µm). The edge atoms of Al pad
Fig. 4. The change of decomposition volume Vd with temperature in the three plating solutions containing different concentrations of l-cysteine.
4.2. Temperature effect The Vd can also be interpreted as the critical reactive surface needed to trigger an overwhelming deposition of metal colloids in plating solution. Similarly, the n/n0 represents the equilibrium molar ratio of the adsorbed stabilizer molecules to those that are in the bulk of the solution. It is logical that the Vd and the n/n0 express the same concept in the sense that the n/n0 decreases with increasing temperature because of desorption, which gives rise to more reactive surface sites and raises the instability of the solution. Hence, a lower Vd value is the result of using a higher plating temperature. It is expected, therefore, that Vd and n/n0 should have a similar response to the temperature. Experimentally (Fig. 4), it was found that for initial concentrations of l-cysteine ranging from 2 to 4 ppm, Vd vs T follows a linear relationship. Similarly, n/n0 vs T should also exhibit a linear relationship. Although in Eq. (10) the two parameters, B1 and B2 , are unknown quantities, we can approximately take (∂ 2 V /∂x 2)x=x0 = 0.1 (eV/Å2 ) and V (x0 ) = −10−1 (eV) by utilizing the Lennard–Jones potential [29]. Using these two values, Eq. (10) becomes ln
1159 n . = 43 + 0.5 ln T + n0 T
(13)
Fig. 6. Schematic illustration of the wafer bumping process.
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
95
Fig. 7. SEM images of the Ni(P) deposits from the solutions containing different initial cysteine concentrations: (a) 1 ppm, (b) 2 ppm, (c) 3 ppm, (d) 4 ppm, (e) 5 ppm.
possess much higher reactivity than the atoms in the plane because of the presence of more coordination-unsaturated Al atoms. Consequently, a faster Ni deposition was observed at the edge area than on the planar area. According to the reverse diffusion phenomenon discussed previously, the chance of detachment of Ni colloidal particles from the edge area is thus higher than that from the planar area. Naturally, the Ni particles will be adsorbed on the passivation layer (SiO2 ) constituting the perimeter of the Al pad and become the seeds of plating on the passivation area. This is what the morphology displayed in Fig. 7a suggests, the whole substrate being covered by a Ni–P alloy layer. However, the nonselective metal deposition can be constrained gradually with the increase in concentration of l-cysteine, as shown in Fig. 7. Moreover, the morphology of the deposition on the passivation zone is a packing of grains with sizes of about 10 µm. This grain-boundary structure likely resulted from the seed-propagation process as aforementioned: while one nanosized Ni particle grew into a micrometer-sized grain, it
must have released a number of nanosized nickel particles. These particles then become the seeds of the other grains. As far as the stabilization of l-cysteine is concerned, an interesting phenomenon has been observed through the comparison of the XPS spectra of the three different Ni(P) deposition layers on the Al pad, which were obtained by using l-cysteine, mercaptoacetic acid, and thiourea, respectively, as stabilizer (Fig. 8). Compared to the other two stabilizers, l-cysteine exhibits a very minor affinity to the Ni surface even though its concentration in the bulk of the plating solution is higher than theirs. We are inclined to judge that l-cysteine has a stronger affinity only to those Ni atoms that are placed in the highly coordination-unsaturated environment. In principle, l-cysteine has two strong hydrophilic groups that would pull the molecule out of binding with Ni atoms by its thiol group. It is because of this counteraction, that the “poisoning effect” of l-cysteine is more selective, and l-cysteine is therefore a more suitable stabilizer than the other two types of structures.
96
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
Acknowledgments The authors thank Plaschem Co. for their kindly supplying the soak cleaning powder and zincate solution and Chartered Semiconductor Manufacturing Co. for their supply of chip dies. References [1] [2] [3] [4]
Fig. 8. The XPS spectra of the sulfur-containing species staying at the deposited Ni(P) films prepared from solutions containing different stabilizers.
5. Conclusions This paper creates a theoretical decomposition volume, Vdmod , which is derived from the one-dimension diffusion model. The model assumes that nickel atoms are formed at the inner Helmholtz plane (IHP), and a handful of them may diffuse into the solution in the form of colloidal particles. These tiny particles will consume the stabilizer and make the bath unstable. This process was simulated experimentally by dropping a PdCl2 solution into a prototype EN bath. The volume of PdCl2 solution added till the occurrence of a heavy deposition of Ni particles is taken as a measure of the buffering capacity of the EN bath, called exp the decomposition volume Vd . Through the simulation of exp Vd , the one-dimensional diffusion model can be used to describe the dependence of the decomposition volume on the plating time under different plating loadings, respectively. Similarly, a linear relation between the decomposition volume and the temperature of the plating bath is derived from the Gouy–Chapman–Stern–Grahame (GCSG) adsorption model, which states that the temperature has a negative effect on the stability of the bath due to the desorption of the stabilizer (l-cysteine in the present case) from the Ni colloidal particles. The under bump metallization (UBM) on patterned silicon wafer was taken as an example to demonstrate the reverse diffusion of Ni nuclei and the effect of stabilizer in impairing this process.
[5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]
[21] [22] [23] [24] [25] [26] [27] [28]
[29]
A. Brenner, G. Riddell, J. Res. Natl. Bur. Stand. 37 (1946) 31. H.G. Schenzel, H. Kreye, Plat. Surf. Finish. 77 (10) (1990) 50. G.A. Di Bari, Plat. Surf. Finish. 85 (11) (1998) 54. C.H. Ting, M. Paunovic, P.L. Pai, G. Chiu, J. Electrochem. Soc. 136 (1989) 462. V.M. Dubin, J. Electrochem. Soc. 139 (1992) 633. H. Honma, H. Watanabe, T. Kobayashi, J. Electrochem. Soc. 141 (1994) 1791. H. Watanabe, H. Honma, J. Electrochem. Soc. 144 (1997) 471. F.M. Onahue, C.U. Yu, Electrochim. Acta 15 (1970) 237. Z. Jusys, J. Liankonis, A. Vaskelis, J. Electrochem. Soc. 307 (1991) 87. L.M. Abrates, M.C. Oliveira, E. Vieil, Electrochim. Acta 41 (9) (1996) 1515. V.D. Meerakker, J. Appl. Electrochem. 11 (1981) 395. L. Li, W. Chen, Z. Gui, Ferroelectrics 231 (1–4) (1999) 793–798. J.J. Podesta, R.C.V. Piatti, A.J. Arvia, J. Appl. Electrochem. 20 (2) (1990) 245. I. Ohno, O. Wakabayashi, S. Haruyama, J. Electrochem. Soc. 132 (10) (1985) 2323. G. Gutzeit, Plating 46 (1959) 1158. N. Feldstein, T.S. Lancsek, J. Electrochem. Soc. 118 (1971) 869. K. Han, L.F. Jing, Met. Finish. 95 (2) (1997) 73. J. Kelber, G. Seshadri, Surf. Interface Anal. 31 (2001) 431. S. Zhang, J. De Baets, M. Vereeken, A. Vervaet, A. Van Calster, J. Electrochem. Soc. 146 (8) (1999) 2870–2875. S. Yiacoumi, C. Tien, Kinetics of Metal Ion Adsorption from Aqueous Solutions: Models, Algorithms, and Applications, Kluwer Academic, Boston, 1995, p. 11. P.H. Rieger, Electrochemistry, Chapman & Hall, London, 1994. P.C. Hiemenz, R. Rajagopalan, Principles of Colloid and Surface Chemistry, 3rd ed., Dekker, New York, 1997. R.J. Hunter, Foundations of Colloid Science, Vol. 1, Clarendon, Oxford, 1986. W.J. Hunt, C.F. Zukoski, J. Colloid Interface Sci. 195 (1997) 32. R.B. McClurg, C.F. Zukoski, J. Colloid Interface Sci. 208 (1998) 529. J. Crank, The Mathematics of Diffusion, 2nd ed., Clarendon, Oxford, 1975. J.A. Nelder, R. Mead, Comput. J. 7 (1964) 308. D.E. Dennis Jr., D.J. Woods, in: A. Wouk (Ed.), New Computing Environments: Microcomputers in Large-Scale Computing, SIAM, Philadelphia, 1987, p. 116. I.N. Levine, Physical Chemistry, 4th ed., McGraw–Hill, New York, 1995.
Modeling the stability of electroless plating bath—diffusion of nickel colloidal particles from the plating frontier X. Yin, L. Hong,a,∗ B.-H. Chen,b and T.-M. Ko c a Department of Chemical and Environmental Engineering, National University of Singapore, 10 Kent Ridge Crescent, 119260 Singapore b Department of Chemical Engineering, National Cheng Kung University, 1 University Road, Tainan 70101, Taiwan c IBM Microelectronics, 2070 Route 52, Mail Stop E40, Hopewell Junction, NY 12533, USA
Received 16 May 2002; accepted 11 February 2003
Abstract Electroless nickel (EN) plating is a process in which Ni2+ ions are reduced by hydrogen atoms adsorbed at a fresh Ni surface. However, detaching of a handful of tiny Ni metal particles from a substrate causes the entrance of these particles into the plating solution. The metal particles offer very reactive surfaces for the reduction of Ni2+ ions, which in turn aggravates the detachment, causing a self-accelerated cycle. Eventually the plating solution will be subject to an overwhelming precipitation of Ni black. This paper proposes a one-dimensional diffusion model to explain the dependence of the bath stability on the plating time under different bath loadings. This mathematical model contains Vd , defined as the decomposition volume, a measure to judge chemical stability of a plating solution. To obtain Vd experimentally, a PdCl2 solution was purposely introduced into a model solution (the addition leads to immediate generation of metal particles) until the very moment of onset of massive deposition of colloidal Ni. The Vd data from the experiment were then used to perform simulation in order to complete the model proposed. Other than the effects of bath loadings and plating time, an adsorption model was also created to describe the temperature effect. To coordinate the adsorption model, l-cysteine was used as an adsorbate that plays a deactivation role. The under bump metallization process on patterned silicon wafers has been used to support the main theme of this study. 2003 Elsevier Science (USA). All rights reserved. Keywords: Electroless nickel plating; Metal colloidal particles; Diffusion; Solution stability; Modeling
1. Introduction
2+ 2H2 PO− 2 + 2H2 O + Ni
−→ Ni0 + 2H+ + 2H2 PO− 3 + H2 . Cat.
Since its discovery by Brenner and Riddell in 1946 [1], the electroless nickel (EN) plating technology has been experiencing fast development because of the demand for various surface engineering processes [2,3]. Recently the application of the EN plating technology in the microelectronicpackaging industry, i.e., metallization of semiconductor chips to create via/conductor hole interconnections [4,5] and chip-on-board solder bumps [6,7], requires more precise understanding of the chemistry of plating. The constituents of an EN solution include a Ni(II) salt, a reducing agent, suitable metal coordination ligands, stabilizers, and additives for particular needs. The overall reaction in an acidic hypophosphite bath can be written as * Corresponding author.
E-mail address: [email protected] (L. Hong).
(1)
The early studies on the chemical mechanism of this reaction suggested that the anodic process (the oxidation of hypophosphite) is a slow step [8]. This viewpoint was further consolidated within the past decade typically through the work of Jusys et al. [9] with electrochemical mass spectra and the work of Abrantes et al. [10] with probe beam deflection. Their results all come to the conclusion that the homolytic cleavage of phosphorus–hydrogen bonds of the adsorbed hypophosphite anion on nickel metal is a catalytic and rate-determining step. Most investigations supported the view that atomic hydrogen rather than hydrogen molecules is the real reducing agent of nickel ions [11–13]. As a matter of fact, the EN plating solution is a kinetically stabilized system, that is, Ni2+ ions cannot be reduced even at the plating temperature (∼90 ◦ C) as long as there is not a solid surface at which the P–H bond can be broken down to supply hydro-
0021-9797/03/$ – see front matter 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S0021-9797(03)00191-7
90
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
gen atoms. It is now known that only a few types of metals possess catalytic activity for the cleavage of P–H bonds; they include Pd, Ni, and Co [14]. Therefore as soon as the first layer of Ni metal atoms is deposited, the P–H bonds can undergo dissociation on it to yield hydrogen atoms, and the reduction of Ni2+ ions afterward becomes self-sustainable. An EN plating bath without a proper stabilization system will produce a lot of fine nickel flakes shortly after the plating is initiated. There has been a notable lack of insight into the origin of this phenomenon. On the basis of the present investigation, we propose that the detachment of colloidal nickel particles from the plating frontier, followed by their diffusion into the bulk of the solution, is responsible for the quick growth of Ni flakes. This is due to the presence of many coordination-unsaturated Ni atoms on these tiny metal nuclei; they accelerate homolysis of the P–H bond. The Ni deposition gains momentum at these particles more than on the desired plating object. The growing particles themselves will soon become the parents of new colloidal particles. As a result, the plating bath becomes contaminated due to this “self-propagation reaction.” Usually, an EN plating bath contains a trace amount of stabilizers. Most of the stabilizers show the common identity of the poisons with the catalysts of either hydrogenation or dehydrogenation reactions [14]. Those that have been identified to be effective stabilizers fell into the following categories: (a) Compounds of group VI elements, such as S, Se, and Te. − 2− (b) Some oxide anions, such as AsO− 2 , IO3 , and MoO4 . (c) Heavy metal ions, such as Sn2+ , Pb2+ , Hg+ , and Sb3+ . Several methods have been developed to assess the effectiveness of stabilizers, such as plotting the deposition (mixed) potential of EN solution vs the concentrations of a stabilizer [15], measuring the deposition rate vs the stabilizer concentration, and testing the buffering capacity of an EN plating bath during an increase in the concentration of metal nuclei by deliberately introducing a PdCl2 solution (0.1 g/l) into the bath until decomposition takes place. Any measure that is capable of restricting the generation of Ni metal particles in an EN solution contributes to the stabilization of the bath. Among the stabilizers listed above, sulfur-containing organic compounds and heavy inorganic cations are most often used. The former type would form Ni–S adsorption bonds, which are believed to impair the catalytic activity of Ni atoms to break down the P–H bond [16,17], and are regarded as anodic-type stabilizers. The second type may function as blocks to the occupancy of those catalytic reactive Ni sites [18,19]. In this work, l-cysteine was employed as an anodic-type stabilizer. This paper proposes a mathematical model to predict the dependence of bath stability on bath loading, plating time, and bath temperature. In this model the decomposition volume Vdmod is defined to express the stability of an EN plating
bath. Experimentally, the decomposition volume was measured by adding an aqueous solution of PdCl2 into the EN solution under investigation at a controlled rate. Unlike Ni2+ ion, Pd2+ ion could be reduced immediately by the hypophosphite ion present in the solution at around 40 ◦ C. The resulting Pd metal particle is an excellent catalyst of the homolysis of H2 PO− 2 that supplies hydrogen atoms. Therefore, Ni deposition will take place forthwith on the Pd particles as long as the catalytic reactivity of the surface Pd atoms is not inhibited by a stabilizer. The volume of the PdCl2 solution consumed at the moment when the EN solution turns into black at a given temperature is taken as the experimenexp tal decomposition volume Vd . Unless specified separately, the volume of EN solution and the concentration of PdCl2 solution used for this study are fixed. exp After having Vd value ready, we started to run a selfconsistent procedure that minimized the difference between exp Vdmod and Vd . In this way, the mathematical model could be specified with respect to a particular set of stabilization conditions implemented in the plating solution. In parexp allel, the dependence of Vd on plating temperature was also investigated using the Gouy–Chapman–Stern–Grahame (GCSG) method [20]. This paper focuses on the diffusion of Ni metal colloidal particles from the plating frontier and attempts to describe its effect on the instability of an EN bath by the two simplified mathematical models.
2. Experimental 2.1. Pretreatment and electroless nickel plating Two different kinds of aluminum substrates supplied by Chartered Semiconductor Manufacturing Ltd. (Singapore) were used in this study; they are E-beam evaporated Al (ca. 1 µm thickness) on silicon wafer and Al bond pads on chip (size 80 × 80 µm, pitch 155 µm). The EN plating cannot take place directly on the aluminum substrates because there is an Al2 O3 thin film on top. To make the aluminum surface ready for plating, a double zincation process is necessary. Commercial zincation solution (Plaschem, Singapore) was employed. After soaking in it, a zinc layer replaced the oxide layer on the Al base: − 0 − 3Zn(OH)2− 4 + 2Al → 2Al(OH)4 + 3Zn + 4OH .
(2)
The EN solution was formulated in house; its composition and the plating conditions are listed in Table 1. The Zn layer was etched out in the acidic environment of the EN bath, which was immediately followed by nickel deposition. The plating time was varied in the range 20–60 min according to the needs of the study. After that, the substrate was taken out and the measurement of Vd was conducted in the solution that remained.
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
91
Table 1 EN bath composition and plating conditions NiSO4 ·6H2 O NaH2 PO2 ·H2 O NaAc DL-Malic acid Lactic acid (90%) Borax Cysteine Temperature pH (adjusted with NaOH) Plating time
26 g/l 28 g/l 10 g/l 15 g/l 15 ml/l 6 g/l Variable 88 ◦ C 4–5 20 min
2.2. Stability measurement An aqueous solution of PdCl2 was prepared by dissolving 0.025 g of PdCl2 in 0.25 ml concentrated hydrochloric acid. The resulting solution was diluted by deionized water to 1 l in a volumetric flask. The decomposition volume Vd was determined by introducing the PdCl2 solution at a rate of 0.05 ml/s into 20 ml of plating solution, keeping the temperature unchanged. Sufficient magnetic stirring was upheld to dissipate any concentration gradient due to the addition of PdCl2 . The Vd is the volume of PdCl2 solution consumed at which the bath suddenly turns black, a sign of large precipitation of metal colloids. It was observed that the Vd alters with the use of different values of the plating loading (PL). The PL is defined as the ratio Aobject/Vbath , where Aobject is the area of the object to be plated in dm2 and Vbath is the volume of plating solution in l. Therefore, PL has unit in dm2 /l. 2.3. Surface morphology and composition measurements The surface morphology of the deposited films was examined by scanning electron microscopy (SEM) (JEOL JSM5600LV). X-ray photon spectroscopy (XPS) was carried out on an instrument (Kratos Axis His, Manchester, UK) with an AlKα X-ray source (at 1486.6 eV) at and 90◦ take-off angle. The pass energy is 40 eV for high-resolution scans.
3. Theoretical background 3.1. The basic assumption for creating the Helmholtz plane in an EN plating system The concept of the Stern–Grahame (S–G) double layer is applied to describe the interface created when a piece of metal, i.e., the Al used in this study, is placed in an EN plating solution. For an active metal, a negatively charged surface in contact with the aqueous solution is expected due to the leaving of a small number of Al atoms from the metal surface in the form of ions, which is regarded as one of the origins of electrode potentials [21]. An adsorption layer consisting of hydrated nickel ions and possibly other complex ions of nickel is to be formed proximately at the fresh surface of Al. This adsorption layer is defined as the inner
Fig. 1. A schematic illustration of the electric double layer likely to form in an electroless nickel plating bath and the structure of the plating frontier.
Helmholtz plane (IHP) [22,23]. Beyond this layer, there is a thicker layer comprising various anions, defined as the outer Helmholtz plane (OHP) (as illustrated by Fig. 1). The electroless plating is essentially an autocatalytic process as elaborated in the preceding section. Following this mechanism, the Ni atoms upon formation on IHP will join the Ni assembly at the surface. However, it could also happen that a handful of Ni metal clusters diffuse outward into the solution through the OHP. The spinning-off depends most probably on the local concentration of Ni–H species, which is governed, in turn, by the coordination environments of the Ni atoms at the surface of the substrate (the plating frontier). If there are a larger number of coordinationunsaturated Ni atoms in a specific geometry, a higher local concentration of Ni–H species will be present, which offers a higher chance of formation of Ni metal clusters (nuclei) because the reducing rate is higher than the packing rate at which Ni atoms find their way into the coating layer. Furthermore, an electrical double layer will be formed surrounding those detached Ni metal nuclei while they diffuse through the S–G layer. Charges built up on these nuclei due to the adsorption of ions would affect their diffusion behavior by electrostatic interactions among the charged particles [24,25] and between them and the solution. The magnitude of the interaction depends upon the potential around a particle. In the present case, the charged adsorption layer on a Ni nucleus could be considered insignificant before it moves into the bulk of the plating solution from the S–G layer. The structure of the S–G layer backs this assumption; both the concentration and activity coefficient of anions in the Gouy layer located between the IHP plane and the OHP plane (the shear surface) are assumed to be lower than in the bulk of solution [23], for they constitute the Stern plane, which is directly in contact with the Ni plating frontier, and as a result of this, cations make up the majority in the Gouy layer. On the other hand, it is expected that the hydration layer surrounding the cations would obstruct their adsorption onto
92
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
the Ni nuclei. As such, the solution to Poisson’s equation for low potential and low ion concentration [22] applies, in which the potential is expressed by q −r/κ −1 e , ΨNi-nucleus = (3) 4πεr where κ −1 characterizes the “thickness” of the adsorption layer on the particle, q the surface charge, and r the radius of the sphere, including the adsorption layer. Based on the above assumption, r κ −1 and q is a small value, which renders the potential around a Ni nanoparticle negligible. As a result, the effect of the charge adsorption layer on the Ni nucleus is omitted in the following mathematical treatment that describes the diffusion behavior of these tiny particles within the S–G layer. Fick’s second equation is applicable to this scenario, ∂C = ∇ · D∇C, (4) ∂t where C is the concentration of Ni nuclei and D is the diffusivity constant. For a planar plating frontier (one-dimensional assumption) with a constant diffusivity, it has the form ∂C ∂ 2C =D 2. ∂t ∂x The initial and boundary conditions are selected as C(x, 0) = 0 C(0, t) = C0
(5)
for t = 0, for t > 0 and x = 0,
where C0 is assumed to be the concentration of Ni nuclei at IHP. Applying these two conditions, the expression for C as a function of x and t is obtained [26]: x x C = C0 1 − erf √ (6) = C0 erfc . 2(Dt)1/2 4Dt Although the concentration of ions in an EN bath increases steadily with plating, it is rationale to assume that the S–G layer undergoes insignificant changes over the period of study because it is a shorter duration compared with the time (4 h) required normally for completing one metal turnover (the time required to deposit all the Ni ions initially present in the solution). Therefore, we can consider the thickness of the double layer (L = IHP + OHP) to be independent of the plating time in the present system: C(L, t) = Cbulk
at x = L.
Given that the concentration of Ni nuclei in the bulk of solution equals Cbulk at plating time t, the following equation is obtained by plugging the above boundary condition into Eq. (6): L . Cbulk(t) = C0 erfc (7) 2(Dt)1/2 Vd has been defined as the volume of the PdCl2 solution added into an EN plating bath to test the buffering capacity of the solution to the concentration of Pd metal colloidal
particles. The metal nuclei in the bulk of the solution should come from two sources: (1) reversed diffusion from OHP of the plating substrate, which is expressed by the product aCbulk, where a is a parameter describing the reactivity of the Ni nuclei originated from the interface, and (2) reduction of Pd2+ ions that are purposely introduced into the solution. To describe the effect of the Pd particles on failure of the plating solution, the minimum volume b of the PdCl2 solution that is needed to trigger the solution decomposition in an unloaded EN bath (the absence of a plating substrate) must be defined. As a short plating time was used in the present study, Cbulk is considerably smaller than b, and so is the aCbulk value. Therefore, the dependence of the decomposition volume Vd on Cbulk can be expressed approximately as L +b Vd = aCbulk + b = a C0 erfc 2(Dt)1/2 K2 = K1 erfc 1/2 + b, (8) t where K1 = aC0 and K2 = L/(2D 1/2 ). According to the physical meaning aforementioned, parameter a must be negative and b greater than zero. Equation (8) gives the dependence of the decomposition volume (Vd ) on the plating time (t). 3.2. The temperature effect on the decomposition volume The Gouy–Chapman–Stern–Grahame (GCSG) model [20] is utilized as a theoretical model to describe temperature effects on the adsorption of l-cysteine molecules on reactive metal colloidal particles. The adsorption of l-cysteine molecules at reactive surface sites of colloidal metal particles will prevent dissociation of the P–H bond that produces hydrogen radicals required for reducing Ni2+ ions. Increasing the plating bath temperature would cause desorption of l-cysteine and therefore increase the reactivity of colloidal metal particles, which brings about instability of the bath. The GCSG equation has the form 1/2 V (x0 ) 2kT π exp − . n = n0 − 2
(9) ∂ V kT 2 ∂x
x=x0
Here n is assumed to be the number of l-cysteine molecules adsorbed at the interface between metal particles and liquid, n0 is the number of the same kind of molecules in the solution, k is the Boltzmann constant, T (K) is the temperature of plating bath, V (x) is the average electric potential of an ion (adsorbate) at distance x from the planar interface (adsorbent), and x0 is the position with a minimum V (x). Equation (9) can be rewritten as 1/2 2kπ V (x0 ) n = n0 − 2
T 1/2 exp − ∂ V kT ∂x 2 x=x0 B2 , = n0 B1 T 1/2 exp (10) T
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
93
or ln
B2 n , = ln B1 + 0.5 ln T + n0 T
(11)
in which B1 and B2 must have values >0 from the perspective of their physical meanings. 4. Results and discussion (a)
4.1. Dependence of the decomposition volume on the plating time On the basis of the typical experimental conditions given in Table 1 and in Section 2.2, the b value in Eq. (8) could be determined experimentally. Here a Z value is defined in order to minimize the discrepancy between the theoretical decomposition volume (Vdmod ) defined in Eq. (8) and the exexp perimental decomposition volume (Vd ) that was obtained under various plating loadings (defined in Section 2.2) and exp plating times, where N represents the number of Vd values taken with varying time and PL, respectively: Z=
N mod exp V − Vd i = f (K1 , K2 ). d
(12)
i=1 exp
In the simulation, the Vdmod will converge on Vd ; a set of most apt K1 and K2 values at a given temperature can then be figured out. The Z value is minimized by using the Nelder–Mead method [27,28]; the minimization is conducted using MATLAB 5.3 with a tolerance value of 1 × 10−4 . The simulation used arbitrarily −1.2 and 2 as the initial values of K1 and K2 , respectively. Figure 2 shows the effects of plating time on the decomposition volume at plating loadings of 1.042, 2.083, and 4.17 dm2 /l, respectively. With respect to a particular platexp ing time, both Vd (dots) and Vdmod (solid line) decrease with increasing plating loading. A larger PL means a greater area of the plating frontier, and hence a greater number of Ni metal colloidal particles were released within the same period of time. The bath stability also decreases with the ex-
Fig. 2. The dependence of decomposition volume on plating time at different plating loadings (PL).
(b) Fig. 3. The change of parameters (a) K1 and (b) K2 with PL as a result of simulation.
tension of plating time due to the fact that a larger number of reactive nickel particles enter into the solution. In a referexp ence experiment, we found that the Vd value changes very little with time in the unloaded bath with other conditions kept unchanged. Compared to the loaded bath, this observation substantiates that the tiny particles coming from the plating frontier contribute to the instability of an EN plating bath. The one-dimensional diffusion model (Eq. (8)) is established on the assumption that the transport of Ni nuclei along the other two dimensions (y and z) is insignificant to the contribution of Ni particles in the bulk of plating solution. In light of this one-dimensional diffusion model, C0 stands for, in effect, the plane concentration of Ni metal particles at OHP. Figure 3a shows a nearly linear dependence of K1 upon PL, with K1 becoming more negative with increasing PL. Since K1 = aC0 (a < 0), this outcome suggests that C0 increases with increased plating loading, which is consistent with the trend expressed by the experiments shown in Fig. 2. In contrast, the K2 value decreases first and then lay out with the increase in PL as shown in Fig. 3b. Knowing that K2 = L/(2D 1/2 ), it is rational to propose that the initial decrease of K2 is due to the increase of the diffusivity value D versus a finite change in the thickness of the double layer. In Fig. 3b, a further increase of PL from 2 to 4 causes a negligible change in the K2 value, which implies that D does not follow C0 when PL is greater than 2 in the present system of investigation. It would be appropriate to think that, at higher levels of PL, a one-dimensional diffusion model may very likely depart from the real situation; namely, two- or three-dimensional diffusion makes the actual number of Ni particles entering the solution less than the one-dimensional model predicts.
94
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
Fig. 5. The plot based on the GCSG model which gives the dependence of the number of l-cysteine molecules adsorbed on Pd colloidal particles upon the testing temperature.
It turns out that the equation gives a straight line in the temperature range of study (Fig. 5). In the future, B1 and B2 values can be determined through the measurement of the n and n0 values by an instrumental method at different temperatures. 4.3. Under bump metallurgy (UBM)—an example of the reversed diffusion phenomenon As shown in Fig. 6, the under bump metallization is completed via several steps including EN plating. The Al pad has a square shape (∼80 × 80 µm). The edge atoms of Al pad
Fig. 4. The change of decomposition volume Vd with temperature in the three plating solutions containing different concentrations of l-cysteine.
4.2. Temperature effect The Vd can also be interpreted as the critical reactive surface needed to trigger an overwhelming deposition of metal colloids in plating solution. Similarly, the n/n0 represents the equilibrium molar ratio of the adsorbed stabilizer molecules to those that are in the bulk of the solution. It is logical that the Vd and the n/n0 express the same concept in the sense that the n/n0 decreases with increasing temperature because of desorption, which gives rise to more reactive surface sites and raises the instability of the solution. Hence, a lower Vd value is the result of using a higher plating temperature. It is expected, therefore, that Vd and n/n0 should have a similar response to the temperature. Experimentally (Fig. 4), it was found that for initial concentrations of l-cysteine ranging from 2 to 4 ppm, Vd vs T follows a linear relationship. Similarly, n/n0 vs T should also exhibit a linear relationship. Although in Eq. (10) the two parameters, B1 and B2 , are unknown quantities, we can approximately take (∂ 2 V /∂x 2)x=x0 = 0.1 (eV/Å2 ) and V (x0 ) = −10−1 (eV) by utilizing the Lennard–Jones potential [29]. Using these two values, Eq. (10) becomes ln
1159 n . = 43 + 0.5 ln T + n0 T
(13)
Fig. 6. Schematic illustration of the wafer bumping process.
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
95
Fig. 7. SEM images of the Ni(P) deposits from the solutions containing different initial cysteine concentrations: (a) 1 ppm, (b) 2 ppm, (c) 3 ppm, (d) 4 ppm, (e) 5 ppm.
possess much higher reactivity than the atoms in the plane because of the presence of more coordination-unsaturated Al atoms. Consequently, a faster Ni deposition was observed at the edge area than on the planar area. According to the reverse diffusion phenomenon discussed previously, the chance of detachment of Ni colloidal particles from the edge area is thus higher than that from the planar area. Naturally, the Ni particles will be adsorbed on the passivation layer (SiO2 ) constituting the perimeter of the Al pad and become the seeds of plating on the passivation area. This is what the morphology displayed in Fig. 7a suggests, the whole substrate being covered by a Ni–P alloy layer. However, the nonselective metal deposition can be constrained gradually with the increase in concentration of l-cysteine, as shown in Fig. 7. Moreover, the morphology of the deposition on the passivation zone is a packing of grains with sizes of about 10 µm. This grain-boundary structure likely resulted from the seed-propagation process as aforementioned: while one nanosized Ni particle grew into a micrometer-sized grain, it
must have released a number of nanosized nickel particles. These particles then become the seeds of the other grains. As far as the stabilization of l-cysteine is concerned, an interesting phenomenon has been observed through the comparison of the XPS spectra of the three different Ni(P) deposition layers on the Al pad, which were obtained by using l-cysteine, mercaptoacetic acid, and thiourea, respectively, as stabilizer (Fig. 8). Compared to the other two stabilizers, l-cysteine exhibits a very minor affinity to the Ni surface even though its concentration in the bulk of the plating solution is higher than theirs. We are inclined to judge that l-cysteine has a stronger affinity only to those Ni atoms that are placed in the highly coordination-unsaturated environment. In principle, l-cysteine has two strong hydrophilic groups that would pull the molecule out of binding with Ni atoms by its thiol group. It is because of this counteraction, that the “poisoning effect” of l-cysteine is more selective, and l-cysteine is therefore a more suitable stabilizer than the other two types of structures.
96
X. Yin et al. / Journal of Colloid and Interface Science 262 (2003) 89–96
Acknowledgments The authors thank Plaschem Co. for their kindly supplying the soak cleaning powder and zincate solution and Chartered Semiconductor Manufacturing Co. for their supply of chip dies. References [1] [2] [3] [4]
Fig. 8. The XPS spectra of the sulfur-containing species staying at the deposited Ni(P) films prepared from solutions containing different stabilizers.
5. Conclusions This paper creates a theoretical decomposition volume, Vdmod , which is derived from the one-dimension diffusion model. The model assumes that nickel atoms are formed at the inner Helmholtz plane (IHP), and a handful of them may diffuse into the solution in the form of colloidal particles. These tiny particles will consume the stabilizer and make the bath unstable. This process was simulated experimentally by dropping a PdCl2 solution into a prototype EN bath. The volume of PdCl2 solution added till the occurrence of a heavy deposition of Ni particles is taken as a measure of the buffering capacity of the EN bath, called exp the decomposition volume Vd . Through the simulation of exp Vd , the one-dimensional diffusion model can be used to describe the dependence of the decomposition volume on the plating time under different plating loadings, respectively. Similarly, a linear relation between the decomposition volume and the temperature of the plating bath is derived from the Gouy–Chapman–Stern–Grahame (GCSG) adsorption model, which states that the temperature has a negative effect on the stability of the bath due to the desorption of the stabilizer (l-cysteine in the present case) from the Ni colloidal particles. The under bump metallization (UBM) on patterned silicon wafer was taken as an example to demonstrate the reverse diffusion of Ni nuclei and the effect of stabilizer in impairing this process.
[5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]
[21] [22] [23] [24] [25] [26] [27] [28]
[29]
A. Brenner, G. Riddell, J. Res. Natl. Bur. Stand. 37 (1946) 31. H.G. Schenzel, H. Kreye, Plat. Surf. Finish. 77 (10) (1990) 50. G.A. Di Bari, Plat. Surf. Finish. 85 (11) (1998) 54. C.H. Ting, M. Paunovic, P.L. Pai, G. Chiu, J. Electrochem. Soc. 136 (1989) 462. V.M. Dubin, J. Electrochem. Soc. 139 (1992) 633. H. Honma, H. Watanabe, T. Kobayashi, J. Electrochem. Soc. 141 (1994) 1791. H. Watanabe, H. Honma, J. Electrochem. Soc. 144 (1997) 471. F.M. Onahue, C.U. Yu, Electrochim. Acta 15 (1970) 237. Z. Jusys, J. Liankonis, A. Vaskelis, J. Electrochem. Soc. 307 (1991) 87. L.M. Abrates, M.C. Oliveira, E. Vieil, Electrochim. Acta 41 (9) (1996) 1515. V.D. Meerakker, J. Appl. Electrochem. 11 (1981) 395. L. Li, W. Chen, Z. Gui, Ferroelectrics 231 (1–4) (1999) 793–798. J.J. Podesta, R.C.V. Piatti, A.J. Arvia, J. Appl. Electrochem. 20 (2) (1990) 245. I. Ohno, O. Wakabayashi, S. Haruyama, J. Electrochem. Soc. 132 (10) (1985) 2323. G. Gutzeit, Plating 46 (1959) 1158. N. Feldstein, T.S. Lancsek, J. Electrochem. Soc. 118 (1971) 869. K. Han, L.F. Jing, Met. Finish. 95 (2) (1997) 73. J. Kelber, G. Seshadri, Surf. Interface Anal. 31 (2001) 431. S. Zhang, J. De Baets, M. Vereeken, A. Vervaet, A. Van Calster, J. Electrochem. Soc. 146 (8) (1999) 2870–2875. S. Yiacoumi, C. Tien, Kinetics of Metal Ion Adsorption from Aqueous Solutions: Models, Algorithms, and Applications, Kluwer Academic, Boston, 1995, p. 11. P.H. Rieger, Electrochemistry, Chapman & Hall, London, 1994. P.C. Hiemenz, R. Rajagopalan, Principles of Colloid and Surface Chemistry, 3rd ed., Dekker, New York, 1997. R.J. Hunter, Foundations of Colloid Science, Vol. 1, Clarendon, Oxford, 1986. W.J. Hunt, C.F. Zukoski, J. Colloid Interface Sci. 195 (1997) 32. R.B. McClurg, C.F. Zukoski, J. Colloid Interface Sci. 208 (1998) 529. J. Crank, The Mathematics of Diffusion, 2nd ed., Clarendon, Oxford, 1975. J.A. Nelder, R. Mead, Comput. J. 7 (1964) 308. D.E. Dennis Jr., D.J. Woods, in: A. Wouk (Ed.), New Computing Environments: Microcomputers in Large-Scale Computing, SIAM, Philadelphia, 1987, p. 116. I.N. Levine, Physical Chemistry, 4th ed., McGraw–Hill, New York, 1995.