- Email: [email protected]
Effect of bath additives on copper electrodeposited directly on diffusion barrier for integrated silicon devices
Thin Solid Films 546 (2013) 263–270
Contents lists available at ScienceDirect
Thin Solid Films journal homepage: www.elsevier.com/locate/tsf
Effect of bath additives on copper electrodeposited directly on diffusion barrier for integrated silicon devices Byoungyong Im, Sunjung Kim ⁎ School of Materials Science and Engineering, University of Ulsan, Ulsan 680-749, Republic of Korea
a r t i c l e
i n f o
Available online 4 April 2013 Keywords: Thin film formation Copper interconnect Seedless electrodeposition Nucleation and growth Additive PEG JGB
a b s t r a c t Polyethylene glycol (PEG) and Janus Green B (JGB) have been widely used as suppressor and leveler, respectively, for copper electrodeposition to fill damascene structures of highly integrated silicon devices. In this study, we investigated fundamental nucleation behavior of copper electrochemically deposited on tungsten diffusion barrier without using a copper seed layer in a citrate-based, neutral electrolyte. Concentrations of PEG and JGB were varied to optimize the nucleation and growth of a thin and uniform copper film on tungsten, ultimately considering the application of this copper electrodeposition method to copper fill of sub-45 nm damascene structures. Important fundamental properties of copper nucleation directly on tungsten such as the nuclei density, size distribution, 3D nucleation mode, and surface roughness were investigated when forming a uniform copper film thinner than 30 nm on tungsten by manipulating additive composition in electrolyte. In consequence, a thin and smooth copper film was electrodeposited resulting from instantaneous nucleation and high area density of copper clusters in the electrolyte including 10 μM PEG and 10 μM JGB. © 2013 Elsevier B.V. All rights reserved.
1. Introduction Recently, electronic devices such as cellular phone, personal computer, and electronic camera become smaller for their portability, but they also require very high and multifunctional performance. In order to reduce their size even with improved performance, more transistors and capacitors with decreased node width should be manufactured within an ultra-large-scale-integrated (ULSI) silicon chip. Transistors, capacitors, and other components of the ULSI silicon chip are interconnected with copper circuit [1]. Trenches and vias, which are main components of copper interconnect, are also scaled down to the width of tens of nanometers for more device integration. Problems for copper interconnect include the difficulty of filling narrow trenches and vias of high aspect-ratio with copper by electrodeposition method. Thus, we have suggested a method of copper electrodeposition directly on metallic and non-metallic diffusion barriers without using a copper seed layer to have more inner space for copper fill [2,3]. Recently transition metals such as W, Ta, Ti, Ru, Cr, Ni, Co, Pt, Pd, Nb, Mo, Ir, and Os have been promisingly considered as diffusion barrier material due to their excellent resistance to copper diffusion as well as high electrical conductivity compared to non-metallic diffusion barriers [4–6]. However, it has been evident that it is not easy to superfill nanometer-scale trenches and vias of which dimension decreases continuously according to the reduction of device node width with
⁎ Corresponding author. Tel.: +82 52 259 2230. E-mail address: [email protected] (S. Kim). 0040-6090/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.tsf.2013.03.075
copper [7]. There have been many efforts to achieve void-free copper fill by controlling bath chemistry including additive compounds [8–12]. Leveler and suppressor molecules prohibit abnormally concentrated copper electrodeposition near the entrance region of damascene structures by being absorbed preferentially at their top-corners. Accelerator molecules as a super catalyst expedite copper reduction process at the bottom of damascene structures, which consequently lead to a bottom-up copper fill. It was reported that the presence of suppressor molecules in an electrolyte suppresses an increase in deposition current density significantly even though the rotation of working electrode mounted with wafer specimen increases it generally [8]. In this research, fundamental nucleation study of very thin copper layers directly electrodeposited on tungsten diffusion barrier is conducted according to the composition of additives in a citrate-based electrolyte. Additives such as polyethylene glycol (PEG) and Janus Green B (JGB) were added to the electrolyte as suppressor and leveler, respectively. Accelerator such as bis-(3-sulfopropyl) disulfide was not included in the electrolyte for direct copper growth on tungsten because ultimately we aim to develop a fundamental electrodeposition process for copper fill of damascene structures narrower than 45 nm and less than 200 nm in depth. Void-free copper fill of relatively wide damascene structures using an electrolyte including both PEG and JGB were previously investigated by several researchers in the aspect of theory building and experimental performance [13,14]. However, the effect of additive composition on the nucleation and growth of copper directly reduced on transition metal diffusion barrier was rarely investigated at the basic level. Consequently, we studied the nucleation and growth behavior of very thin copper layers on tungsten diffusion barrier
264
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
fundamentally with changing additive composition in the electrolyte including PEG and JGB. 2. Experimental details A 100-nm thick tungsten diffusion barrier layer was sputterdeposited on thermally grown silicon oxide of 4 in. silicon wafer. The tungsten-deposited silicon wafer was cleaned with ethanol and rinsed with deionized water before copper electrodeposition. The silicon wafer was cut into rectangular specimens with a dimension of 1.5 × 2 cm, and then the wafer specimen was mounted on the Teflon housing, as shown in Fig. 1. It was fixed by Teflon working electrode. An area of tungsten surface exposed to the electrolyte was approximately 1 cm2. The base electrolyte contains 0.05 M CuSO4·5(H2O) and 0.05 M (NH4)2HC6H5O7 in common. pH of the electrolyte was 7, which was controlled using NH4OH solution. PEG and JGB were used as suppressor and leveler, respectively. The concentration of PEG in the electrolyte was varied from 0 to 1000 μM, whereas that of JGB was 10 μM if it was added. Experiments were carried out in a three-electrode cell using a potentiostat (Gamry Inc.). A saturated calomel electrode (SCE) was used as a reference electrode, and a pure copper plate with a dimension of 5 × 5 × 0.6 cm was used as a counter electrode. Copper deposition was performed by applying a constant cathodic potential of −1.0 V to the specimen-mounted working electrode for 10 s. Cyclic voltammetry (CV) was performed to characterize cathodic and anodic polarization behavior in the range of 2.0 to −2.0 V at a scan rate of 5 mV/s. The morphology of deposits was examined using a field emission scanning electron microscopy (FESEM, Quanta 200, FEI) with an operating voltage of 15.0 kV. SEM images were employed to identify the nucleation mode, which could be confirmed by comparing experimental current transient curves with three-dimensional (3D) nucleation models developed by Scharifker and Hills [15–17] and by using a statistical consideration of size distribution of copper clusters. Atomic force microscope (AFM, XE-100, Park Systems) was used to measure the surface roughness of copper deposits. It was operated in non-contact mode using a non-contact cantilever tip (PPP-NCHR 10 M). 3. Results and discussion CVs in three different baths were conducted for tungsten diffusion barrier, as shown in Fig. 2. The base electrolyte does not contain any additive except copper sulfate and citrate complexing agent, which were effective in electrodepositing thin copper films on non-metallic diffusion barriers [2,3]. The other baths contain only PEG or JGB individually as additive in the base electrolyte. It was previously suggested that JGB of high molecule weight replaces other compounds like PEG existing in a seed surface [18]. Supposing that the suggestion regarding JGB in a multi-additive containing electrolyte is plausible, it is necessary to investigate the role of PEG or JGB independently in order to avoid missing the effect of PEG in the citrate-based electrolyte including both additives. CV curves are shifted to lower current density in cathodic
Fig. 1. Schematic diagram of the cross-sectional structure of working electrode including a wafer specimen.
Fig. 2. Cyclic voltammetry obtained from three electrolytes of pH 7. Its scan rate is 5 mV/s.
potential region when PEG or JGB is added. It implies that larger cathodic polarization occurs by adding 10 μM JGB or 1000 μM PEG to the base electrolyte because the adsorption of JGB or PEG molecules on tungsten hinders copper reduction partly. From our previous report, we found that applying deposition potential near −1.0 V in the citrate-based electrolyte without additives is proper to produce a thin and uniform copper film resulted from a high nuclei density of copper [4]. Potentials more positive than −1.0 V have led to relatively coarse copper deposit, whereas more negative potentials have resulted in rough copper deposit accompanied by hydrogen evolution. Thus, copper depositions in this study were carried out at a constant potential of −1.0 V. Current transient curves obtained from potentiostatic depositions were compared with Scharifker and Hills' theoretical model to determine 3D nucleation mode during copper deposition [15–17]. The theoretical model includes instantaneous and progressive nucleation. Instantaneous nucleation is preferred to form a thin and uniform copper film. However, any experimental result cannot correspond with exactly instantaneous nucleation, which indicates that no more nuclei should form after initial nuclei are created at the beginning of electrodeposition. The 3D instantaneous nucleation is expressed as follows. 2 2 i t t 1− exp −1:2564 ¼ 1:9542 m im tm t
ð1Þ
The 3D progressive nucleation can be identified by the following expression as 2 2 2 i t t 1− exp −2:3367 ¼ 1:2254 m im tm t
ð2Þ
where im is a maximum current density and tm is a time at im. Fig. 3 shows the normalized curves of current transient curves in three baths according to Scharifker and Hills' model. When PEG or JGB is added to the base electrolyte, the curves are shifted from progressive nucleation to instantaneous nucleation because of the inhibition effect of PEG and JGB. The curve of the electrolyte containing 1000 μM PEG was closer to that of instantaneous nucleation. PEG as suppressor works effectively in suppressing further copper nucleation once copper nuclei are formed initially. Fig. 4 shows SEM images of copper deposits on tungsten diffusion barrier according to additive condition in the electrolyte. We are concerned on the area density, size distribution, surface roughness, and surface coverage of copper clusters considering the effect of individual additive on copper nucleation. Forming a thin and uniform copper film, which is proper for void-free copper fill of nanometer scale damascene structures, needs a copper nuclei density higher than 2 × 1010 cm−2 from theoretical calculation. The value of 1.8 × 1010 cm−2 was achieved
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 3. Comparison of experimental current transient curves created in three electrolytes and theoretical curves of Scharifker and Hills' model to investigate a nucleation mode of copper deposits.
265
by copper electrodeposition on tungsten at −1.0 V in the base electrolyte without additives. It is lower than the required copper nuclei density for void-free copper fill. However, adding 1000 μM PEG to the base electrolyte increases the copper nuclei density up to 2.58 × 1010 cm−2. Furthermore, the addition of 10 μM JGB brings out the value of 2.74 × 1010 cm−2. These values are enough to form very thin copper films proper for void-free fill in nanometer scale damascene structures. The size of copper clusters was previously determined by measuring the cluster diameter [4]. However, this method is not proper to measure the typical size of individual clusters precisely because copper clusters shown in 2D SEM images are not perfect circles. Thus, we developed a more precise method of determining the cluster size by measuring the cross-sectional area of copper clusters using an image analysis software (Pixcavator 2.3, Intelligent Perception), as shown in Fig. 4. Fig. 5 shows the size distribution of copper clusters deposited on tungsten at −1.0 V for 10 s in each electrolyte. The average size of copper clusters is 0.0035 μm 2 in the base electrolyte without additives. It is decreased to 0.0018 and 0.0026 μm 2 by adding 1000 μM PEG and 10 μM JGB, respectively, to the base electrolyte. In addition, a wide distribution in the size of copper clusters for additive-free electrolyte is significantly reduced by adding PEG or JGB. The shrinkage of the size distribution of copper
Fig. 4. SEM images (left) of copper clusters deposited on tungsten diffusion barrier at −1.0 V for 10 s in three electrolytes. 2D cluster maps (right) were converted from the SEM images for the measurement of size distribution and area density of copper clusters.
266
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 7. Surface roughness of copper deposits according to additive condition in electrolyte.
Fig. 5. Size distribution of copper clusters, according to their 2D cross-sectional area, deposited on tungsten diffusion barrier at −1.0 V for 10 s in three electrolytes.
clusters toward the average value indicates that copper nucleation tends to be more instantaneous in the electrolyte containing PEG or JGB. This tendency to instantaneous nucleation of copper clusters in additive-containing electrolytes is coincident well with the result presented in Fig. 3. AFM analysis could provide the effect of JGB as leveler, as shown in Fig. 6. Scanned surface area by an AFM tip was 1.25 × 1.25 μm, which was equal to the area taken for measuring the size distribution and area density of copper clusters. Roughness of copper deposits depends on additive condition in electrolyte, as shown Fig. 7. Error bars of ±10%
Fig. 6. AFM images of copper deposited on tungsten diffusion barrier at −1.0 V for 10 s in three electrolytes.
were added to data points. Roughness of copper clusters decreases with adding additive. The average roughness of copper clusters is reduced to about 3.2 and 1.9 nm for PEG- and JGB-added electrolyte, respectively. JGB as leveler is very effective to reduce the surface roughness of copper deposits. PEG also helps in leveling copper deposit surface because it suppresses non-uniform copper reduction especially concentrating on relatively large clusters even though its leveling effect is not so significant as JGB. As a second part of our research, we conducted nucleation study of copper on tungsten diffusion barrier in electrolytes with both additives. The concentration of PEG was varied in the electrolyte, which basically contained 10 μM JGB. CVs in Fig. 8 show that current density decreases with increasing PEG concentration in electrolyte. It implies that more tungsten surface is covered by PEG with hindering copper reduction as PEG concentration increases. This phenomenon is contrary to the previous expectation that JGB replaces PEG mostly [18]. Another report suggesting that PEG molecules tend to replace previously adsorbed JGB molecules is more persuasive to explain the variation of CV curves in cathodic region [19]. Therefore, the influence of PEG on the nucleation and growth of copper may be more critical than that of JGB when both additives exist within electrolyte. Chronoamperometries were conducted at − 1.0 V for 10 s according to PEG concentration in electrolyte, as shown in Fig. 9. Cathodic current density decreases with increasing PEG concentration due to its suppressing ability of copper reduction. However, it does
Fig. 8. Cyclic voltammetry obtained citrate-based, neutral electrolytes containing 10 μM JGB with varying PEG concentration. Its scan rate is 5 mV/s.
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 9. Current transient curves obtained from potentiostatic depositions at −1.0 V for 10 s in citrate-based, neutral electrolytes containing 10 μM JGB with varying PEG concentration.
not decrease further beyond 100 μM PEG because tungsten surface available for PEG adsorption is fully occupied by PEG molecules in the electrolyte with 100 μM PEG while PEG adsorption and copper reduction are competitive. More PEG molecules than 100 μM may exist as impurities within copper films. Scharifker and Hills' model was employed again to determine the 3D nucleation mode of copper according to PEG concentration in electrolyte, as shown in Fig. 10. However, it is difficult to verify the nucleation mode of experimental curves by comparing with the theoretical model because the curves are not sufficiently constructed beyond tm. The nucleation mode of copper according to PEG concentration is rather to be determined by image analysis of copper deposits. We analyzed nucleation mode by using SEM images, as shown in Fig. 11. For more precise determination of size distribution and area density of copper clusters, the SEM images were converted to 2D cluster maps distinctly showing individual copper clusters using the image analysis software. Addition of 10 μM PEG to JGB-containing electrolyte results in a similar copper nuclei density to the value of 2.74 × 1010 cm−2, which was obtained in the electrolyte with only 10 μM JGB, whereas it seems to slightly help decreasing the size of copper clusters. More PEG than 10 μM in the JGB-containing electrolyte
267
leads to less populated copper deposits because the surface passivating effect of PEG is excessive. The size distribution of copper clusters deposited on tungsten at −1.0 V for 10 s in each electrolyte is shown in Fig. 12. Copper clusters formed in the electrolyte with both 10 μM PEG and 10 μM JGB have a size distribution slightly narrower than that produced in the electrolyte containing JGB only. In addition, the average size of copper clusters is decreased from 0.0026 to 0.0018 μm2 by adding 10 μM JGB to the JGB-containing electrolyte with maintaining a high nuclei density of 2.59 × 1010 cm−2. On the other hand, the addition of excessive PEG such as 100 and 1000 μM increases the width of size distribution of copper clusters. Furthermore, copper nuclei density is significantly decreased to 1.77 × 1010 and 1.59 × 1010 cm−2 for 100 and 1000 μM PEG addition, respectively. Consequently, a low amount of PEG such as 10 μM in the JGB-containing electrolyte replaces JGB molecules partly so that it decreases the average size of copper clusters and, at the same time, it is helpful to maintain the nature of instantaneous nucleation of copper. JGB molecules unreplaced by PEG molecules perform as leveler to keep the roughness of copper deposit as low as 2.1 nm. Fig. 13 shows the cross-sectional SEM images of copper-deposited wafer specimens according to PEG concentration after 20 s deposition at − 1.0 V. The thickness of copper deposits was in the range of 25 to 30 nm. For electrolytes with 0 and 10 μM PEG, the full coverage of tungsten surface by copper indicates that a complete, thin film of copper is formed. Increasing PEG concentration leads to particulate-like copper deposits, which has relatively large surface roughness in spite of lateral contact of individual clusters. Whether the formation of a thin and uniform copper film is completed can also be verified by measuring sheet resistance of copper deposits. A sheet resistance of a 28-nm thick copper film sputter-deposited on tungsten surface was measured, for reference, as 21.0 Ω/sq, which is a value influenced by the underlying tungsten layer of which sheet resistance was 27.8 Ω/sq. Copper deposits formed in the electrolytes with PEG less than 100 μM have reasonable sheet resistances between 21.5 and 26.2 Ω/sq, as shown in Fig. 13, from which we determine that complete copper films were formed. The sheet resistance, 34.1 Ω/sq, of copper deposit obtained for 1000 μM PEG is even higher than that of tungsten layer, which possibly indicates that partly uncompleted film formation and excessive PEG molecule residue may interrupt current flow along the copper deposit.
4. Conclusion Copper was deposited directly on tungsten diffusion barrier at − 1.0 V in a citrate-based, neutral electrolyte, considering its application to copper interconnect in sub-45 nm silicon devices. Nucleation behavior of copper on tungsten was fundamentally investigated according to the composition of additives such as PEG and JGB in the citrate-based electrolyte. Adding PEG or JGB to the electrolyte influence positively the high-density nucleation of copper on tungsten and even can shift its 3D nucleation mode toward desirable instantaneous nucleation. Instantaneous nucleation is preferred to form a thin and uniform copper film directly on tungsten diffusion barrier. However, adding both PEG and JGB to the electrolyte should be more careful because PEG is predominantly adsorbed on tungsten replacing JGB if PEG is added excessively. The electrolyte containing 10 μM PEG and 10 μM JGB is proper to form thin and uniform copper films on tungsten considering a void-free copper fill in nanometer scale damascene structures.
Acknowledgment Fig. 10. Comparison of experimental current transient curves created in citrate-based, neutral electrolytes containing 10 μM JGB and theoretical curves of Scharifker and Hills' model to investigate a nucleation mode of copper deposits.
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2012-0006887).
268
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 11. SEM images (left) of copper clusters deposited on tungsten diffusion barrier at −1.0 V for 10 s in citrate-based, neutral electrolytes containing 10 μM JGB. 2D cluster maps (right) were converted from the SEM images for the measurement of size distribution and area density of copper clusters.
References [1] [2] [3] [4] [5] [6]
P.C. Andricacos, ECS Interface 8 (1999) 6. S. Kim, D.J. Duquette, Surf. Coat. Technol. 201 (2006) 2712. S. Kim, D.J. Duquette, J. Electrochem. Soc. 153 (2006) C417. K. Park, S. Kim, J. Electrochem. Soc. 157 (2010) D609. D. Starosvetsky, N. Sezin, Y. Ein-Eli, Electrochim. Acta 55 (2010) 1656. T.E. Hong, K. Mun, S. Choi, J. Park, S. Kim, T. Cheon, W.K. Kim, B. Lim, S. Kim, Thin Solid Films 520 (2012) 6100. [7] D. Wheeler, D. Josell, T. Moffat, 2002, p. 827. [8] F.I. Lizama-Tzec, L. Canché-Canul, G. Oskam, Electrochim. Acta 56 (2011) 9391. [9] W. Dow, C. Li, Y. Su, S. Shen, C. Huang, C. Lee, B. Hsu, S. Hsu, Electrochim. Acta 54 (2009) 5894.
[10] A. Radisic, O. Lühn, H.G.G. Philipsen, Z. El-Mekki, M. Honore, S. Rodet, S. Armini, C. Drijbooms, H. Bender, W. Ruythooren, Microelectron. Eng. 88 (2011) 701. [11] K. Kondo, N. Yamakawa, Z. Tanaka, K. Hayashi, J. Electroanal. Chem. 559 (2003) 137. [12] O. Lühn, C. Van Hoof, W. Ruythooren, Electrochim. Acta 54 (2009) 2504. [13] P. Taephaisitphongse, Y. Cao, A.C. West, J. Electrochem. Soc. 148 (2001) C492. [14] W.-P. Dow, H.-S. Huang, M.-Y. Yen, H.-C. Huang, J. Electrochem. Soc. 152 (2005) C425. [15] G. Gunawardena, G. Hills, I. Montenegro, B. Scharifker, J. Electroanal. Chem. Interfacial Electrochem. 138 (1982) 225. [16] B.R. Scharifker, J. Mostany, J. Electroanal. Chem. Interfacial Electrochem. 177 (1984) 13. [17] B. Scharifker, G. Hills, Electrochim. Acta 28 (1983) 879. [18] T.P. Moffat, D. Wheeler, S.-K. Kim, D. Josell, J. Electrochem. Soc. 153 (2006) C127. [19] J.D. Adolf, Master Thesis, Case Western Reserve University, 2008.
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 12. Size distribution of copper clusters, according to their 2D cross-sectional area, deposited on tungsten diffusion barrier at −1.0 V for 10 s in citrate-based, neutral electrolytes containing 10 μM JGB.
269
270
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 13. Cross-sectional SEM images (top) showing copper layers deposited at −1.0 V for 20 s in citrate-based, neutral electrolytes containing 10 μM JGB. The variation of sheet resistance (bottom) according to PEG concentration is also shown.
Contents lists available at ScienceDirect
Thin Solid Films journal homepage: www.elsevier.com/locate/tsf
Effect of bath additives on copper electrodeposited directly on diffusion barrier for integrated silicon devices Byoungyong Im, Sunjung Kim ⁎ School of Materials Science and Engineering, University of Ulsan, Ulsan 680-749, Republic of Korea
a r t i c l e
i n f o
Available online 4 April 2013 Keywords: Thin film formation Copper interconnect Seedless electrodeposition Nucleation and growth Additive PEG JGB
a b s t r a c t Polyethylene glycol (PEG) and Janus Green B (JGB) have been widely used as suppressor and leveler, respectively, for copper electrodeposition to fill damascene structures of highly integrated silicon devices. In this study, we investigated fundamental nucleation behavior of copper electrochemically deposited on tungsten diffusion barrier without using a copper seed layer in a citrate-based, neutral electrolyte. Concentrations of PEG and JGB were varied to optimize the nucleation and growth of a thin and uniform copper film on tungsten, ultimately considering the application of this copper electrodeposition method to copper fill of sub-45 nm damascene structures. Important fundamental properties of copper nucleation directly on tungsten such as the nuclei density, size distribution, 3D nucleation mode, and surface roughness were investigated when forming a uniform copper film thinner than 30 nm on tungsten by manipulating additive composition in electrolyte. In consequence, a thin and smooth copper film was electrodeposited resulting from instantaneous nucleation and high area density of copper clusters in the electrolyte including 10 μM PEG and 10 μM JGB. © 2013 Elsevier B.V. All rights reserved.
1. Introduction Recently, electronic devices such as cellular phone, personal computer, and electronic camera become smaller for their portability, but they also require very high and multifunctional performance. In order to reduce their size even with improved performance, more transistors and capacitors with decreased node width should be manufactured within an ultra-large-scale-integrated (ULSI) silicon chip. Transistors, capacitors, and other components of the ULSI silicon chip are interconnected with copper circuit [1]. Trenches and vias, which are main components of copper interconnect, are also scaled down to the width of tens of nanometers for more device integration. Problems for copper interconnect include the difficulty of filling narrow trenches and vias of high aspect-ratio with copper by electrodeposition method. Thus, we have suggested a method of copper electrodeposition directly on metallic and non-metallic diffusion barriers without using a copper seed layer to have more inner space for copper fill [2,3]. Recently transition metals such as W, Ta, Ti, Ru, Cr, Ni, Co, Pt, Pd, Nb, Mo, Ir, and Os have been promisingly considered as diffusion barrier material due to their excellent resistance to copper diffusion as well as high electrical conductivity compared to non-metallic diffusion barriers [4–6]. However, it has been evident that it is not easy to superfill nanometer-scale trenches and vias of which dimension decreases continuously according to the reduction of device node width with
⁎ Corresponding author. Tel.: +82 52 259 2230. E-mail address: [email protected] (S. Kim). 0040-6090/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.tsf.2013.03.075
copper [7]. There have been many efforts to achieve void-free copper fill by controlling bath chemistry including additive compounds [8–12]. Leveler and suppressor molecules prohibit abnormally concentrated copper electrodeposition near the entrance region of damascene structures by being absorbed preferentially at their top-corners. Accelerator molecules as a super catalyst expedite copper reduction process at the bottom of damascene structures, which consequently lead to a bottom-up copper fill. It was reported that the presence of suppressor molecules in an electrolyte suppresses an increase in deposition current density significantly even though the rotation of working electrode mounted with wafer specimen increases it generally [8]. In this research, fundamental nucleation study of very thin copper layers directly electrodeposited on tungsten diffusion barrier is conducted according to the composition of additives in a citrate-based electrolyte. Additives such as polyethylene glycol (PEG) and Janus Green B (JGB) were added to the electrolyte as suppressor and leveler, respectively. Accelerator such as bis-(3-sulfopropyl) disulfide was not included in the electrolyte for direct copper growth on tungsten because ultimately we aim to develop a fundamental electrodeposition process for copper fill of damascene structures narrower than 45 nm and less than 200 nm in depth. Void-free copper fill of relatively wide damascene structures using an electrolyte including both PEG and JGB were previously investigated by several researchers in the aspect of theory building and experimental performance [13,14]. However, the effect of additive composition on the nucleation and growth of copper directly reduced on transition metal diffusion barrier was rarely investigated at the basic level. Consequently, we studied the nucleation and growth behavior of very thin copper layers on tungsten diffusion barrier
264
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
fundamentally with changing additive composition in the electrolyte including PEG and JGB. 2. Experimental details A 100-nm thick tungsten diffusion barrier layer was sputterdeposited on thermally grown silicon oxide of 4 in. silicon wafer. The tungsten-deposited silicon wafer was cleaned with ethanol and rinsed with deionized water before copper electrodeposition. The silicon wafer was cut into rectangular specimens with a dimension of 1.5 × 2 cm, and then the wafer specimen was mounted on the Teflon housing, as shown in Fig. 1. It was fixed by Teflon working electrode. An area of tungsten surface exposed to the electrolyte was approximately 1 cm2. The base electrolyte contains 0.05 M CuSO4·5(H2O) and 0.05 M (NH4)2HC6H5O7 in common. pH of the electrolyte was 7, which was controlled using NH4OH solution. PEG and JGB were used as suppressor and leveler, respectively. The concentration of PEG in the electrolyte was varied from 0 to 1000 μM, whereas that of JGB was 10 μM if it was added. Experiments were carried out in a three-electrode cell using a potentiostat (Gamry Inc.). A saturated calomel electrode (SCE) was used as a reference electrode, and a pure copper plate with a dimension of 5 × 5 × 0.6 cm was used as a counter electrode. Copper deposition was performed by applying a constant cathodic potential of −1.0 V to the specimen-mounted working electrode for 10 s. Cyclic voltammetry (CV) was performed to characterize cathodic and anodic polarization behavior in the range of 2.0 to −2.0 V at a scan rate of 5 mV/s. The morphology of deposits was examined using a field emission scanning electron microscopy (FESEM, Quanta 200, FEI) with an operating voltage of 15.0 kV. SEM images were employed to identify the nucleation mode, which could be confirmed by comparing experimental current transient curves with three-dimensional (3D) nucleation models developed by Scharifker and Hills [15–17] and by using a statistical consideration of size distribution of copper clusters. Atomic force microscope (AFM, XE-100, Park Systems) was used to measure the surface roughness of copper deposits. It was operated in non-contact mode using a non-contact cantilever tip (PPP-NCHR 10 M). 3. Results and discussion CVs in three different baths were conducted for tungsten diffusion barrier, as shown in Fig. 2. The base electrolyte does not contain any additive except copper sulfate and citrate complexing agent, which were effective in electrodepositing thin copper films on non-metallic diffusion barriers [2,3]. The other baths contain only PEG or JGB individually as additive in the base electrolyte. It was previously suggested that JGB of high molecule weight replaces other compounds like PEG existing in a seed surface [18]. Supposing that the suggestion regarding JGB in a multi-additive containing electrolyte is plausible, it is necessary to investigate the role of PEG or JGB independently in order to avoid missing the effect of PEG in the citrate-based electrolyte including both additives. CV curves are shifted to lower current density in cathodic
Fig. 1. Schematic diagram of the cross-sectional structure of working electrode including a wafer specimen.
Fig. 2. Cyclic voltammetry obtained from three electrolytes of pH 7. Its scan rate is 5 mV/s.
potential region when PEG or JGB is added. It implies that larger cathodic polarization occurs by adding 10 μM JGB or 1000 μM PEG to the base electrolyte because the adsorption of JGB or PEG molecules on tungsten hinders copper reduction partly. From our previous report, we found that applying deposition potential near −1.0 V in the citrate-based electrolyte without additives is proper to produce a thin and uniform copper film resulted from a high nuclei density of copper [4]. Potentials more positive than −1.0 V have led to relatively coarse copper deposit, whereas more negative potentials have resulted in rough copper deposit accompanied by hydrogen evolution. Thus, copper depositions in this study were carried out at a constant potential of −1.0 V. Current transient curves obtained from potentiostatic depositions were compared with Scharifker and Hills' theoretical model to determine 3D nucleation mode during copper deposition [15–17]. The theoretical model includes instantaneous and progressive nucleation. Instantaneous nucleation is preferred to form a thin and uniform copper film. However, any experimental result cannot correspond with exactly instantaneous nucleation, which indicates that no more nuclei should form after initial nuclei are created at the beginning of electrodeposition. The 3D instantaneous nucleation is expressed as follows. 2 2 i t t 1− exp −1:2564 ¼ 1:9542 m im tm t
ð1Þ
The 3D progressive nucleation can be identified by the following expression as 2 2 2 i t t 1− exp −2:3367 ¼ 1:2254 m im tm t
ð2Þ
where im is a maximum current density and tm is a time at im. Fig. 3 shows the normalized curves of current transient curves in three baths according to Scharifker and Hills' model. When PEG or JGB is added to the base electrolyte, the curves are shifted from progressive nucleation to instantaneous nucleation because of the inhibition effect of PEG and JGB. The curve of the electrolyte containing 1000 μM PEG was closer to that of instantaneous nucleation. PEG as suppressor works effectively in suppressing further copper nucleation once copper nuclei are formed initially. Fig. 4 shows SEM images of copper deposits on tungsten diffusion barrier according to additive condition in the electrolyte. We are concerned on the area density, size distribution, surface roughness, and surface coverage of copper clusters considering the effect of individual additive on copper nucleation. Forming a thin and uniform copper film, which is proper for void-free copper fill of nanometer scale damascene structures, needs a copper nuclei density higher than 2 × 1010 cm−2 from theoretical calculation. The value of 1.8 × 1010 cm−2 was achieved
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 3. Comparison of experimental current transient curves created in three electrolytes and theoretical curves of Scharifker and Hills' model to investigate a nucleation mode of copper deposits.
265
by copper electrodeposition on tungsten at −1.0 V in the base electrolyte without additives. It is lower than the required copper nuclei density for void-free copper fill. However, adding 1000 μM PEG to the base electrolyte increases the copper nuclei density up to 2.58 × 1010 cm−2. Furthermore, the addition of 10 μM JGB brings out the value of 2.74 × 1010 cm−2. These values are enough to form very thin copper films proper for void-free fill in nanometer scale damascene structures. The size of copper clusters was previously determined by measuring the cluster diameter [4]. However, this method is not proper to measure the typical size of individual clusters precisely because copper clusters shown in 2D SEM images are not perfect circles. Thus, we developed a more precise method of determining the cluster size by measuring the cross-sectional area of copper clusters using an image analysis software (Pixcavator 2.3, Intelligent Perception), as shown in Fig. 4. Fig. 5 shows the size distribution of copper clusters deposited on tungsten at −1.0 V for 10 s in each electrolyte. The average size of copper clusters is 0.0035 μm 2 in the base electrolyte without additives. It is decreased to 0.0018 and 0.0026 μm 2 by adding 1000 μM PEG and 10 μM JGB, respectively, to the base electrolyte. In addition, a wide distribution in the size of copper clusters for additive-free electrolyte is significantly reduced by adding PEG or JGB. The shrinkage of the size distribution of copper
Fig. 4. SEM images (left) of copper clusters deposited on tungsten diffusion barrier at −1.0 V for 10 s in three electrolytes. 2D cluster maps (right) were converted from the SEM images for the measurement of size distribution and area density of copper clusters.
266
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 7. Surface roughness of copper deposits according to additive condition in electrolyte.
Fig. 5. Size distribution of copper clusters, according to their 2D cross-sectional area, deposited on tungsten diffusion barrier at −1.0 V for 10 s in three electrolytes.
clusters toward the average value indicates that copper nucleation tends to be more instantaneous in the electrolyte containing PEG or JGB. This tendency to instantaneous nucleation of copper clusters in additive-containing electrolytes is coincident well with the result presented in Fig. 3. AFM analysis could provide the effect of JGB as leveler, as shown in Fig. 6. Scanned surface area by an AFM tip was 1.25 × 1.25 μm, which was equal to the area taken for measuring the size distribution and area density of copper clusters. Roughness of copper deposits depends on additive condition in electrolyte, as shown Fig. 7. Error bars of ±10%
Fig. 6. AFM images of copper deposited on tungsten diffusion barrier at −1.0 V for 10 s in three electrolytes.
were added to data points. Roughness of copper clusters decreases with adding additive. The average roughness of copper clusters is reduced to about 3.2 and 1.9 nm for PEG- and JGB-added electrolyte, respectively. JGB as leveler is very effective to reduce the surface roughness of copper deposits. PEG also helps in leveling copper deposit surface because it suppresses non-uniform copper reduction especially concentrating on relatively large clusters even though its leveling effect is not so significant as JGB. As a second part of our research, we conducted nucleation study of copper on tungsten diffusion barrier in electrolytes with both additives. The concentration of PEG was varied in the electrolyte, which basically contained 10 μM JGB. CVs in Fig. 8 show that current density decreases with increasing PEG concentration in electrolyte. It implies that more tungsten surface is covered by PEG with hindering copper reduction as PEG concentration increases. This phenomenon is contrary to the previous expectation that JGB replaces PEG mostly [18]. Another report suggesting that PEG molecules tend to replace previously adsorbed JGB molecules is more persuasive to explain the variation of CV curves in cathodic region [19]. Therefore, the influence of PEG on the nucleation and growth of copper may be more critical than that of JGB when both additives exist within electrolyte. Chronoamperometries were conducted at − 1.0 V for 10 s according to PEG concentration in electrolyte, as shown in Fig. 9. Cathodic current density decreases with increasing PEG concentration due to its suppressing ability of copper reduction. However, it does
Fig. 8. Cyclic voltammetry obtained citrate-based, neutral electrolytes containing 10 μM JGB with varying PEG concentration. Its scan rate is 5 mV/s.
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 9. Current transient curves obtained from potentiostatic depositions at −1.0 V for 10 s in citrate-based, neutral electrolytes containing 10 μM JGB with varying PEG concentration.
not decrease further beyond 100 μM PEG because tungsten surface available for PEG adsorption is fully occupied by PEG molecules in the electrolyte with 100 μM PEG while PEG adsorption and copper reduction are competitive. More PEG molecules than 100 μM may exist as impurities within copper films. Scharifker and Hills' model was employed again to determine the 3D nucleation mode of copper according to PEG concentration in electrolyte, as shown in Fig. 10. However, it is difficult to verify the nucleation mode of experimental curves by comparing with the theoretical model because the curves are not sufficiently constructed beyond tm. The nucleation mode of copper according to PEG concentration is rather to be determined by image analysis of copper deposits. We analyzed nucleation mode by using SEM images, as shown in Fig. 11. For more precise determination of size distribution and area density of copper clusters, the SEM images were converted to 2D cluster maps distinctly showing individual copper clusters using the image analysis software. Addition of 10 μM PEG to JGB-containing electrolyte results in a similar copper nuclei density to the value of 2.74 × 1010 cm−2, which was obtained in the electrolyte with only 10 μM JGB, whereas it seems to slightly help decreasing the size of copper clusters. More PEG than 10 μM in the JGB-containing electrolyte
267
leads to less populated copper deposits because the surface passivating effect of PEG is excessive. The size distribution of copper clusters deposited on tungsten at −1.0 V for 10 s in each electrolyte is shown in Fig. 12. Copper clusters formed in the electrolyte with both 10 μM PEG and 10 μM JGB have a size distribution slightly narrower than that produced in the electrolyte containing JGB only. In addition, the average size of copper clusters is decreased from 0.0026 to 0.0018 μm2 by adding 10 μM JGB to the JGB-containing electrolyte with maintaining a high nuclei density of 2.59 × 1010 cm−2. On the other hand, the addition of excessive PEG such as 100 and 1000 μM increases the width of size distribution of copper clusters. Furthermore, copper nuclei density is significantly decreased to 1.77 × 1010 and 1.59 × 1010 cm−2 for 100 and 1000 μM PEG addition, respectively. Consequently, a low amount of PEG such as 10 μM in the JGB-containing electrolyte replaces JGB molecules partly so that it decreases the average size of copper clusters and, at the same time, it is helpful to maintain the nature of instantaneous nucleation of copper. JGB molecules unreplaced by PEG molecules perform as leveler to keep the roughness of copper deposit as low as 2.1 nm. Fig. 13 shows the cross-sectional SEM images of copper-deposited wafer specimens according to PEG concentration after 20 s deposition at − 1.0 V. The thickness of copper deposits was in the range of 25 to 30 nm. For electrolytes with 0 and 10 μM PEG, the full coverage of tungsten surface by copper indicates that a complete, thin film of copper is formed. Increasing PEG concentration leads to particulate-like copper deposits, which has relatively large surface roughness in spite of lateral contact of individual clusters. Whether the formation of a thin and uniform copper film is completed can also be verified by measuring sheet resistance of copper deposits. A sheet resistance of a 28-nm thick copper film sputter-deposited on tungsten surface was measured, for reference, as 21.0 Ω/sq, which is a value influenced by the underlying tungsten layer of which sheet resistance was 27.8 Ω/sq. Copper deposits formed in the electrolytes with PEG less than 100 μM have reasonable sheet resistances between 21.5 and 26.2 Ω/sq, as shown in Fig. 13, from which we determine that complete copper films were formed. The sheet resistance, 34.1 Ω/sq, of copper deposit obtained for 1000 μM PEG is even higher than that of tungsten layer, which possibly indicates that partly uncompleted film formation and excessive PEG molecule residue may interrupt current flow along the copper deposit.
4. Conclusion Copper was deposited directly on tungsten diffusion barrier at − 1.0 V in a citrate-based, neutral electrolyte, considering its application to copper interconnect in sub-45 nm silicon devices. Nucleation behavior of copper on tungsten was fundamentally investigated according to the composition of additives such as PEG and JGB in the citrate-based electrolyte. Adding PEG or JGB to the electrolyte influence positively the high-density nucleation of copper on tungsten and even can shift its 3D nucleation mode toward desirable instantaneous nucleation. Instantaneous nucleation is preferred to form a thin and uniform copper film directly on tungsten diffusion barrier. However, adding both PEG and JGB to the electrolyte should be more careful because PEG is predominantly adsorbed on tungsten replacing JGB if PEG is added excessively. The electrolyte containing 10 μM PEG and 10 μM JGB is proper to form thin and uniform copper films on tungsten considering a void-free copper fill in nanometer scale damascene structures.
Acknowledgment Fig. 10. Comparison of experimental current transient curves created in citrate-based, neutral electrolytes containing 10 μM JGB and theoretical curves of Scharifker and Hills' model to investigate a nucleation mode of copper deposits.
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2012-0006887).
268
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 11. SEM images (left) of copper clusters deposited on tungsten diffusion barrier at −1.0 V for 10 s in citrate-based, neutral electrolytes containing 10 μM JGB. 2D cluster maps (right) were converted from the SEM images for the measurement of size distribution and area density of copper clusters.
References [1] [2] [3] [4] [5] [6]
P.C. Andricacos, ECS Interface 8 (1999) 6. S. Kim, D.J. Duquette, Surf. Coat. Technol. 201 (2006) 2712. S. Kim, D.J. Duquette, J. Electrochem. Soc. 153 (2006) C417. K. Park, S. Kim, J. Electrochem. Soc. 157 (2010) D609. D. Starosvetsky, N. Sezin, Y. Ein-Eli, Electrochim. Acta 55 (2010) 1656. T.E. Hong, K. Mun, S. Choi, J. Park, S. Kim, T. Cheon, W.K. Kim, B. Lim, S. Kim, Thin Solid Films 520 (2012) 6100. [7] D. Wheeler, D. Josell, T. Moffat, 2002, p. 827. [8] F.I. Lizama-Tzec, L. Canché-Canul, G. Oskam, Electrochim. Acta 56 (2011) 9391. [9] W. Dow, C. Li, Y. Su, S. Shen, C. Huang, C. Lee, B. Hsu, S. Hsu, Electrochim. Acta 54 (2009) 5894.
[10] A. Radisic, O. Lühn, H.G.G. Philipsen, Z. El-Mekki, M. Honore, S. Rodet, S. Armini, C. Drijbooms, H. Bender, W. Ruythooren, Microelectron. Eng. 88 (2011) 701. [11] K. Kondo, N. Yamakawa, Z. Tanaka, K. Hayashi, J. Electroanal. Chem. 559 (2003) 137. [12] O. Lühn, C. Van Hoof, W. Ruythooren, Electrochim. Acta 54 (2009) 2504. [13] P. Taephaisitphongse, Y. Cao, A.C. West, J. Electrochem. Soc. 148 (2001) C492. [14] W.-P. Dow, H.-S. Huang, M.-Y. Yen, H.-C. Huang, J. Electrochem. Soc. 152 (2005) C425. [15] G. Gunawardena, G. Hills, I. Montenegro, B. Scharifker, J. Electroanal. Chem. Interfacial Electrochem. 138 (1982) 225. [16] B.R. Scharifker, J. Mostany, J. Electroanal. Chem. Interfacial Electrochem. 177 (1984) 13. [17] B. Scharifker, G. Hills, Electrochim. Acta 28 (1983) 879. [18] T.P. Moffat, D. Wheeler, S.-K. Kim, D. Josell, J. Electrochem. Soc. 153 (2006) C127. [19] J.D. Adolf, Master Thesis, Case Western Reserve University, 2008.
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 12. Size distribution of copper clusters, according to their 2D cross-sectional area, deposited on tungsten diffusion barrier at −1.0 V for 10 s in citrate-based, neutral electrolytes containing 10 μM JGB.
269
270
B. Im, S. Kim / Thin Solid Films 546 (2013) 263–270
Fig. 13. Cross-sectional SEM images (top) showing copper layers deposited at −1.0 V for 20 s in citrate-based, neutral electrolytes containing 10 μM JGB. The variation of sheet resistance (bottom) according to PEG concentration is also shown.