Advertisement

Obtaining Atomically Smooth 4H–SiC (0001) Surface by Controlling Balance Between Anodizing and Polishing in Electrochemical Mechanical Polishing

  • Xu Yang
  • Xiaozhe Yang
  • Rongyan Sun
  • Kentaro Kawai
  • Kenta Arima
  • Kazuya YamamuraEmail author
Original Articles
  • 51 Downloads

Abstract

Single-crystal 4H–SiC is a promising next-generation semiconductor material for high-power and low-loss power devices. Electrochemical mechanical polishing (ECMP) is a very promising polishing technique for the manufacture of SiC wafers owing to its high efficiency and low cost. In this study, the effects of the balance between the anodic oxidation rate and the polishing rate of the oxide layer on the polishing performance of slurryless ECMP were studied in an attempt to obtain an atomically smooth surface efficiently. The polishing performance of ECMP was evaluated from the viewpoints of surface roughness, residual oxide, and material removal rate. It was found that the balance between the anodic oxidation rate and the polishing rate of the oxide layer strongly affects the surface roughness; the polishing rate of the oxide layer should be greater than the anodic oxidation rate to obtain an atomically smooth surface. By controlling this balance at a current density of 10 mA/cm2, we were able to decrease the surface roughness of a diamond-lapped 4H–SiC (0001) surface from Sq 4.290 to 0.577 nm and obtained a high material removal rate of about 10 μm/h. This study provides a promising way of obtaining atomically flat surfaces by slurryless ECMP.

Keywords

SiC Anodic oxidation Electrochemical mechanical polishing Slurryless Anodizing and polishing balance High efficiency 

1 Introduction

Single-crystal 4H–SiC is a promising next-generation semiconductor material that can improve the energy conversion efficiency and reliability of electronic devices owing to its excellent electronic and thermal properties. Therefore, it is widely used in the fields of photovoltaics [1], hybrid electric vehicles [2, 3], high-power applications [4, 5], and so forth. However, the difficulty of manufacturing SiC wafers has limited the spread of use of SiC electronic devices. Many polishing techniques, such as chemical mechanical polishing (CMP) [6], plasma-assisted polishing (PAP) [7], catalyst-referred etching (CARE) [8, 9], mechanical chemical polishing (MCP) [10], and UV-assisted polishing [11], have been developed to polish SiC wafers, but their material removal rate (MRR) is unsatisfactory for the industrial production of SiC wafers. Furthermore, the cost and environmental problems associated with currently industrially used CMP limit its application [12].

Electrochemical mechanical polishing (ECMP), which combines surface anodic oxidation and mechanical polishing of the oxide layer, has been proposed for the highly efficient polishing of SiC wafers [13, 14, 15]. The application of ECMP to 4H–SiC wafers was first attempted by Li et al. [14] using a KNO3 and H2O2 solution as an electrolyte and a silica slurry as a polishing medium. Large-area flattening of a SiC (0001) surface was obtained by both two-step and simultaneous ECMP, but many etch pits were generated on the polished surface that significantly increased the surface roughness, and additional hydrogen etching was applied to finish the surface to atomic-scale roughness. In our previous study [15], a ceria slurry was used as both an electrolyte and a polishing medium. A damage-free surface was obtained, but not an atomically flat surface. It has been reported that the balance between chemical removal and mechanical removal has a significant effect on the surface roughness in the CMP of electronic materials [16]. Since the mechanism of ECMP is similar to that of CMP, the quality of the surface obtained by ECMP is very likely to be affected by the balance between the anodic oxidation rate and the polishing rate of the oxide layer.

In this study, slurryless ECMP, in which fixed soft abrasives are used as the polishing medium, was conducted on diamond-lapped 4H–SiC (0001) surfaces. During the ECMP process, the balance between the anodic oxidation and the mechanical removal of the oxide layer was controlled by the feeding rate of the SiC wafer. The effects of this balance on the surface quality were investigated.

2 Experimental Section

Figure 1 shows a schematic diagram of our newly developed ECMP machine, which consists of an anodic oxidation system and a polishing unit. In the ECMP machine, a SiC wafer is mounted on a copper plate and set as the working electrode (WE), and a grinding stone is set on the tip of the spindle in contact with the SiC surface. The tip of the spindle is made of aluminum alloy and used as a counter electrode (CE). A direct current (DC) power source is applied to control the anodic oxidation condition. Electrolyte is supplied from the center of the spindle by a peristaltic pump to prevent the electrolyte from being ejected by the centrifugal force of the spindle.
Fig. 1

ECMP machine

In this study, sodium chloride (NaCl) aqueous solution with a weight concentration of 1% was used as the electrolyte. A vitrified bonded ceria grinding stone supplied by Mizuho Co., Ltd. with an average particle size of 1 μm (#8000) was applied to remove the oxide layer. During ECMP, the ceria grinding stone was rotated with the spindle and the SiC wafer underwent reciprocating motion along the x-axis at a set feeding rate. Therefore, on the basis of the polishing motion pattern, when part of the SiC surface was being polished, another area was being oxidized during the ECMP process, and the oxidation duration of the area being oxidized could be controlled by the feeding rate of the X stage. That is, the balance between the anodic oxidation rate and the removal rate of the oxide layer was controlled by the feeding rate. ECMP experiments were conducted at different feeding rates; the experimental parameters are shown in Table 1. Diamond-lapped 4H–SiC wafers (on-axis, n-type) with a thickness of about 360 μm and a resistivity range of 0.015–0.028 Ω cm were used for the ECMP experiments, and all experiments were performed on the (0001) face. Surface topographies after ECMP were observed by scanning white light interferometry (SWLI, NewView 8300, Zygo), residual oxide on the polished surface was observed by X-ray photoelectron spectroscopy (XPS, Quantum 2000, ULVAC-PHI), and MRRs under different ECMP conditions were evaluated by observing the cross-sectional views of the polishing areas using a stylus profiler (Surfcom 1400D, Tokyo Seimitsu), as shown in Fig. 1.
Table 1

ECMP parameters

Current density (mA/cm2)

10

Spindle rotation speed (rpm)

1500

Load (kPa)

140

Feeding rate (mm/s)

① 1

② 2

③ 5

④ 10

Reciprocating distance (mm)

5

Polishing time (min)

30

Flow rate (mL/min)

380

3 Results and Discussion

Figure 2 shows an SWLI image of an as-received diamond-lapped surface used for ECMP, which was lapped with a #4000 diamond grinding plate (average particle size of 3 μm) for 10 min at a pressure of 19.6 kPa. Scratches and pits were observed owing to the mechanical removal of the material by the diamond abrasives. It had a relatively large Sq surface roughness of 4.290 nm.
Fig. 2

SWLI image of as-received diamond-lapped surface, Sq = 4.290 nm, Sz = 61.881 nm

Figure 3 shows SWLI images of surfaces processed by ECMP at different feeding rates. It is clear that scratches on the as-received diamond-lapped surface were removed and the surface roughness was decreased at all feeding rates, but many etch pits were observed on the surface polished with feeding rates of 1 and 2 mm/s. Moreover, the number of pits decreased with increasing feeding rate, and the pits completely disappeared when the feeding rate was increased to 10 mm/s. Figure 4 shows the relationship between surface roughness and feeding rate. The surface roughness decreased with increasing feeding rate. With the disappearance of the pits, both the Sz and Sq surface roughnesses decreased rapidly, indicating that the surface roughness was mainly affected by the pits generated.
Fig. 3

SWLI images of SiC surfaces polished at feeding rates of a 1, b 2, c 5, and d 10 mm/s

Fig. 4

Relationship between surface roughness and feeding rate

To investigate the reason for the generation of pits, the removal depths of the SiC surfaces were observed using a stylus profiler. Figure 5 shows cross-sectional views of polishing spots obtained by ECMP at different feeding rates. Almost the same removal depth of about 5 μm from the surface of the as-received SiC in 30 min was observed at different feeding rates, as shown in Fig. 5. Therefore, the MRR of SiC was about 10 μm/h regardless of the feeding rates. Since the removal depths of all samples were similar, the degrees of subsurface damage removal were similar for all samples. These results indicate that the pits on the surface polished at low feeding rates did not originate from diamond lapping, but from the ECMP process.
Fig. 5

Cross-sectional views of polishing spots obtained at different feeding rates

It was also observed that the overall cross-sectional views of the polishing spots were different, and the flatness of the polishing areas became worse with increasing feeding rate, as shown in Fig. 5. The differences in overall cross-sectional views of the polishing spots were caused by the warp of the thin SiC wafers inevitably induced by their manufacturing process. On the other hand, the changes in the flatness of the polishing areas are attributed to the polishing motion in the ECMP process. During ECMP, the ceria grinding stone was rotated with the spindle and the SiC wafer underwent reciprocating motion along the x-axis at a set feeding rate, and a ring-like ceria grinding stone was applied, as shown in Fig. 1. The grinding stone momentarily stopped at both left and right sides of the polishing area owing to the reciprocating motion of the X stage, and thus, a relatively higher removal amount with a ring-like shape would be obtained at both left and right sides of the polishing area owing to a longer polishing time. Therefore, the center of the cross-sectional views would be a little higher than the two sides. With the increase in feeding rate, the moving time between the two stopping sites decreased, which increased the difference in the removal amount between the two stopping sites and other areas, deteriorating the flatness. Furthermore, the contact status between the SiC surface and the grinding stone also affects the polishing result. These problems are solved when the ECMP machine will be improved to polish the entire wafer.

Figure 6a shows the XPS measurement results of the as-received diamond-lapped surface; a small amount of oxide was observed. Figure 6b–e shows the XPS measurement results of the SiC surfaces polished at different feeding rates. Strong peaks corresponding to Si–C bonds and weak peaks corresponding to Si–C–O bonds were observed. This indicates that a small amount of silicon oxycarbide still remained on the polished surface, but the removal states of the anodic oxide of the four surfaces polished at different feeding rates were almost same. Silicon oxycarbide is an intermediate product of the oxidation of SiC. Although it is more difficult to remove than silicon oxide, it can be completely removed by additional mechanical polishing [17].
Fig. 6

Si 2p spectra of a as-received diamond-lapped surface, and surfaces polished at feeding rates of b 1, c 2, d 5, and e 10 mm/s obtained by XPS

On the basis of the above analysis, the removal depth and the removal states of the oxide of the four surfaces were almost same. Therefore, the difference in the ECMP of the four SiC surfaces can be attributed to the difference in the feeding rate, as modeled in Fig. 7. In the case of a low feeding rate, the oxidation time of the SiC surface was long because a longer time was required to complete the reciprocating motion of the SiC wafer. Since a constant current was applied in the ECMP experiments, the oxidation rate is considered to be same at different feeding rates on the basis of Faraday’s electrolysis law. Therefore, the thickness of the oxide layer before being removed by the grinding stone was proportional to the oxidation time and inversely proportional to the feeding rate. Thus, the thicknesses of the oxide layer at a feeding rate of 1 mm/s can be estimated to be 2, 5, and 10 times those at feeding rates of 2, 5, and 10 mm/s, respectively, as shown in Fig. 7. Chemical oxidation/etching of a 4H–SiC surface is an anisotropic process. In particular, in the electrochemical process, the distribution of the current density on the SiC surface also affects the oxidation uniformity [18]. In addition, the anodic oxidation of SiC does not stop with the progress of oxidation owing to the porosity of the oxide layer [19]. Therefore, it is assumed that the interface between the bulk SiC and the oxide layer becomes rough with the thickening of the oxide layer because the ceria grinding stone can only remove the oxide layer, as shown in Fig. 7. After removing the oxide layer, a rough surface with pits was obtained at the feeding rate of 1 mm/s owing to the formation of the thick oxide layer. With increasing feeding rate, the thickness of the oxide layer decreased and the interface between the bulk SiC and the oxide layer became smoother, and a smooth surface without pits was obtained at the feeding rate of 10 mm/s.
Fig. 7

Modeling of ECMP process at feeding rates of a 1, b 2, c 5, and d 10 mm/s

An anodic oxidation experiment was conducted to verify the proposed model in Fig. 7. Diamond-lapped surfaces were anodically oxidized at a constant current density of 10 mA/cm2 in 1 wt% NaCl aqueous solution for 2, 10, and 30 min. Then, the oxide layers were removed by dipping into 50 wt% hydrofluoric acid (HF) solution. Five areas on the surfaces were observed by SWLI after removing the oxide layer. Figure 8 shows typical SWLI images of these surfaces, and Fig. 9 shows the surface roughness of these surfaces calculated from the five observed areas for each sample. The surfaces oxidized for 2 and 10 min still had a similar topography pattern to the diamond-lapped surface as shown in Fig. 8a, b, but the Sz and Sq surface roughnesses increased, as shown in Fig. 9. This is attributed to the preferential oxidation of the surface damage on the diamond-lapped surface; the regions of scratches and pits on the surface were preferentially oxidized, which resulted in the deepening and widening of the scratches and pits and an increase in the Sz surface roughness [20]. Figure 8c shows the surface oxidized for 30 min after removing the oxide layer by HF dipping. The surface structure found on the diamond-lapped surface was not observed, whereas many pits and fibrous protrusions were randomly distributed on the surface. This is attributed to the anisotropic oxidation of the SiC surface [18]. Therefore, the initial oxidation period of the SiC surface was dominated by the distribution of subsurface damage on the SiC wafer and then by the anisotropy of single-crystal 4H–SiC after the subsurface damage had been completely oxidized. For both types of oxidation, the surface roughness increased with increasing oxidation time.
Fig. 8

SWLI images of SiC surfaces oxidized at a current density of 10 mA/cm2 at oxidation times of a 2, b 10, and c 30 min after removing the oxide layer by HF etching

Fig. 9

Relationship of Sq and Sz surface roughnesses with oxidation time

The results in Figs. 8 and 9 verify the proposed model in Fig. 7. A smooth surface was obtained when the oxide layer was removed immediately before the interface between the bulk SiC and oxide layer became rough. This suggests that the removal rate of the oxide layer should be greater than the anodic oxidation rate of SiC to obtain a smooth surface by ECMP.

4 Conclusions

In this study, the effects of the balance between the anodic oxidation rate and the removal rate of the oxide layer on the polishing performance of slurryless ECMP were investigated. This balance was controlled by the feeding rate in the ECMP of SiC wafers using the prototype apparatus developed by us, and a higher feeding rate resulted in the oxide layer being removed more rapidly. The roughness of the SiC surface after ECMP was found to decrease with increasing feeding rate, and the pits that were generated in ECMP at low feeding rates disappeared at a high feeding rate. A smooth surface with an Sq surface roughness of 0.577 nm and an Sz surface roughness of 3.876 nm was obtained at an MRR of about 10 μm/h. By performing anodic oxidation experiments with different oxidation times, we found that the interface between the bulk SiC and the oxide layer became rough with the thickening of the oxide layer. The results of this study show that the surface roughness obtained by ECMP is mainly determined by the balance between the anodic oxidation rate and the removal rate of the oxide layer. A smooth surface can be obtained if the oxide layer can be removed immediately after its generation. It is concluded that the removal rate of the oxide layer should be greater than the anodic oxidation rate to obtain a smooth surface in the ECMP of SiC wafers.

Notes

Acknowledgements

This work was partially supported by a Grant-in-Aid for Challenging Research (Exploratory) (18K18810) from the MEXT, Japan, a research grant from the Mitsutoyo Association for Science, and a research grant from the Technology and Machine Tool Engineering Foundation.

References

  1. 1.
    Stalter O, Kranzer D, Rogalla S, Burger B (2010) Advanced solar power electronics. In: 2010 22nd International symposium on power semiconductor devices and IC’s (ISPSD). IEEE, pp 3–10Google Scholar
  2. 2.
    Zhang H, Tolbert LM, Ozpineci B (2011) Impact of SiC devices on hybrid electric and plug-in hybrid electric vehicles. IEEE Trans Ind Appl 47(2):912–921CrossRefGoogle Scholar
  3. 3.
    Wrzecionko B, Bortis D, Kolar JW (2014) A 120 °C ambient temperature forced air-cooled normally-off SiC JFET automotive inverter system. IEEE Trans Power Electr 29(5):2345–2358CrossRefGoogle Scholar
  4. 4.
    Friedli T, Round SD, Hassler D, Kolar JW (2009) Design and performance of a 200-kHz all-SiC JFET current DC-link back to back converter. IEEE Trans Ind Appl 45(5):1868–1878CrossRefGoogle Scholar
  5. 5.
    Peftitsis D, Tolstoy G, Antonopoulos A, Rabkowski J, Lim JK, Bakowski M, Ängquist L, Nee HP (2012) High-power modular multilevel converters with SiC JFETs. IEEE Trans Power Electr 27(1):28–36CrossRefGoogle Scholar
  6. 6.
    Zhai W, Gao B, Chang J, Wang H (2019) Optimization of ultrasonic-assisted polishing SiC through CFD simulation. Nanomanuf Metrol 2(1):36–44CrossRefGoogle Scholar
  7. 7.
    Yamamura K, Takiguchi T, Ueda M, Deng H, Hattori AN, Zettsu N (2011) Plasma assisted polishing of single crystal SiC for obtaining atomically flat strain-free surface. CIRP Ann 60(1):571–574CrossRefGoogle Scholar
  8. 8.
    Van Bui P, Sano Y, Morikawa Y, Yamauchi K (2018) Characteristics and mechanism of catalyst-referred etching method: application to 4H–SiC. Int J Autom Technol 12(2):154–159CrossRefGoogle Scholar
  9. 9.
    Hara H, Sano Y, Mimura H, Arima K, Kubota A, Yagi K, Murata J, Yamauchi K (2007) Damage-free planarization of 4H–SiC (0001) by catalyst-referred etching. Mater Sci Forum 556–557:749–751CrossRefGoogle Scholar
  10. 10.
    Kikuchi M, Takahashi Y, Suga T, Suzuki S, Bando Y (1992) Mechanochemical polishing of silicon carbide single crystal with chromium (III) oxide abrasive. J Am Ceram Soc 75(1):189–194CrossRefGoogle Scholar
  11. 11.
    Kubota A, Kurihara K, Touge M (2012) Fabrication of smooth surface on 4H–SiC substrate by ultraviolet assisted local polishing in hydrogen peroxide solution. Key Eng Mater 523–524:24–28CrossRefGoogle Scholar
  12. 12.
    Lai CL, Lin SH (2003) Electrocoagulation of chemical mechanical polishing (CMP) wastewater from semiconductor fabrication. Chem Eng J 95(1–3):205–211CrossRefGoogle Scholar
  13. 13.
    Murata J, Yodogawa K, Ban K (2017) Polishing-pad-free electrochemical mechanical polishing of single-crystalline SiC surfaces using polyurethane–CeO2 core–shell particles. Int J Mach Tools Manuf 114:1–7CrossRefGoogle Scholar
  14. 14.
    Li C, Bhat IB, Wang R, Seiler J (2004) Electro-chemical mechanical polishing of silicon carbide. J Electron Mater 33(5):481–486CrossRefGoogle Scholar
  15. 15.
    Deng H, Hosoya K, Imanishi Y, Endo K, Yamamura K (2015) Electro-chemical mechanical polishing of single-crystal SiC using CeO2 slurry. Electrochem Commun 52:5–8CrossRefGoogle Scholar
  16. 16.
    Lee HS, Jeong HD (2009) Chemical and mechanical balance in polishing of electronic materials for defect-free surfaces. CIRP Ann 58(1):485–490CrossRefGoogle Scholar
  17. 17.
    Deng H, Yamamura K (2013) Atomic-scale flattening mechanism of 4H–SiC (0001) in plasma assisted polishing. CIRP Ann 62(1):575–578CrossRefGoogle Scholar
  18. 18.
    Yang X, Sun R, Kawai K, Arima K, Yamamura K (2018) Surface modification and microstructuring of 4H–SiC (0001) by anodic oxidation with sodium chloride aqueous solution. ACS Appl Mater Interfaces 11(2):2535–2542CrossRefGoogle Scholar
  19. 19.
    Yang X, Yang X, Kawai K, Arima K, Yamamura K (2019) Ultrasonic-assisted anodic oxidation of 4H–SiC (0001) surface. Electrochem Commun 100:1–5CrossRefGoogle Scholar
  20. 20.
    Yang X, Sun R, Ohkubo Y, Kawai K, Arima K, Endo K, Yamamura K (2018) Investigation of anodic oxidation mechanism of 4H–SiC (0001) for electrochemical mechanical polishing. Electrochim Acta 271:666–676CrossRefGoogle Scholar

Copyright information

© International Society for Nanomanufacturing and Tianjin University and Springer Nature Singapore Pte Ltd. 2019

Authors and Affiliations

  • Xu Yang
    • 1
  • Xiaozhe Yang
    • 1
  • Rongyan Sun
    • 1
  • Kentaro Kawai
    • 1
  • Kenta Arima
    • 1
  • Kazuya Yamamura
    • 1
    Email author
  1. 1.Department of Precision Science and Technology, Graduate School of EngineeringOsaka UniversitySuitaJapan

Personalised recommendations