Analysis of Nitrogen State on MOS Interface of 4H-SiC m-face after Nitric Oxide Post Oxidation Annealing (NO-POA)

Transcription

1 e-journal of Surface Science and Nanotechnology 31 October 2017 e-j. Surf. Sci. Nanotech. Vol. 15 (2017) Regular Paper Analysis of Nitrogen State on MOS Interface of 4H-SiC m-face after Nitric Oxide Post Oxidation Annealing (NO-POA) Kimimori Hamada Power Electronics Development Division, Toyota Motor Corporation, 543, Kirigahora, Nishihirose-cho, Toyota, Aichi , Japan, and Doctoral Program in Nano-Science and Nano-Technology, Graduate School of Pure and Applied Sciences, University of Tsukuba, Tennodai, Tsukuba, Ibaraki , Japan Akira Mikami and Hideki Naruoka Power Electronics Development Division, Toyota Motor Corporation, 543, Kirigahora, Nishihirose-cho, Toyota, Aichi , Japan Kikuo Yamabe Graduate School of Pure and Applied Sciences, University of Tsukuba, Tennodai, Tsukuba, Ibaraki , Japan (Received 1 August 2017; Accepted 23 August 2017; Published 31 October 2017) It is known that the interface nitrogen density of the 4H-SiC Si-face, C-face, and a-face increases as a result of the NO-POA process, that the electron mobility increases as the nitrogen density increases, and that each face has a different interface nitrogen saturation density. In contrast, the anisotropy of the nitridation characteristics of the m-face, which is regarded as a promising channel for high-performance trench MOSFETs, is not well known. To identify the nitridation status of the m-face after NO-POA, the MOS interface structures with a SiO 2 formed on m-face by CVD and treated by NO-POA was investigated by SIMS, HAXPES, XPS, and XAFS. In the same way as the other faces, it was found that the nitrogen segregates on the SiO 2/SiC MOS interface, that most of the nitrogen combines with Si, and that the interface nitrogen density has a unique saturation value. The nitrogen density saturation value on the m-face measured by SIMS was cm 2. This value is approximately 1.5 times the exposed carbon density on the top surface of the m-face. [DOI: /ejssnt ] Keywords: Silicon carbide; Silicon oxides; Nitrides; Nitrogen oxides; Semiconductor-insulator interfaces I. INTRODUCTION Silicon carbide (SiC) has excellent material characteristics for power semiconductor devices, such as high electric field breakdown strength and high thermal conductivity [1, 2]. However, one issue of SiC devices is the difficulty of obtaining the expected low on-resistance characteristics because of the low electron mobility of the MOS interface due to its high interface trap density (D it ) [3]. Recently, the nitric oxide post-oxidation annealing (NO- POA) process has been widely recognized as a very effective approach to reduce D it and improve on-resistance [4 8]. This process increases the MOS interface electron mobility by segregating nitrogen on the MOS interface and reducing the D it near the conduction band, which is known to depress mobility. Consequently, it is very important to know the nitridation status at the MOS interface. The adoption of a trench MOS gate structure is also regarded as another promising way to reduce on-resistance because of its larger per-unit area gate width compared to a normal planar gate structure [9 12]. When a trench MOSFET is fabricated on a mass production commercial Si-faced SiC wafer, the MOS channel surface is formed at one of the a-face, m-face, or the face between them. Although the a-face and m-face reportedly show higher mobility than the Si or C faces, less is known about the characteristics of the gate interface compared to the Si or C faces [13]. Especially, there are no reports describing Corresponding author: kimimori hamada@mail.toyota.co.jp the anisotropy of the nitridation status of the m-face after the NO-POA process, which has a direct impact on mobility [14]. The research described in this paper investigated the nitridation status of the MOS interface structures on the m-face by SIMS, HAXPES, XPS, and XAFS. This was accomplished by carrying out NO-POA after forming a SiO 2 film by low-pressure chemical vapor deposition(lp- CVD). This paper also considers the NO-POA nitridation mechanism of the m-face MOS interface. II. EXPERIMENTAL DETAILS 1.5-inch m-faced heavy doped n-typed 4H-SiC wafers with 5 µm epitaxial layers doped to a concentration of cm 2 were used in this study. A Si-faced 4-inch wafer with the same doping concentrations and epitaxial layer thickness was used as a reference. After forming a 75 nm SiO 2 film by LP-CVD on the surface of the wafers, NO- POA was carried out in NO (10%) gas at 1 atm at 1000, 1150 and 1300 C for 30 minutes on the m-face wafers, and under the same gas conditions at 1300 C for 30 minutes on the Si-face wafer. The interface structure was analyzed by secondary ion mass spectroscopy (SIMS), hard X-ray photoelectron spectroscopy (HAXPES), X-ray photoelectron spectroscopy (XPS), and X-ray absorption fine structure (XAFS). SIMS was used to obtain the depth profile of N from the SiO 2 film to the SiC substrate. This study defined the MOS interface nitrogen density as the total number of ni- ISSN c 2017 The Surface Science Society of Japan ( 109

2 Volume 15 (2017) Hamada, et al. trogen atoms between 6 and +6 nm of the MOS interface per unit area (cm 2 ). Although the detailed conditions of the SIMS measurement have not been disclosed by the SIMS engineering service company, the MOS interface nitrogen saturation density is reported to be cm 2 at 1 atm and 1150 C after a 4-hour NO-POA process on the Si-face of a 4H-SiC wafer. In addition, the error on SIMS data of the nitrogen density was reported to be 20% in comparison to nuclear reaction analysis (NRA) results [15, 16]. This study references this saturation density value and SIMS error when discussing the absolute value of nitrogen density. The HAXPES analysis utilized the SPring-8 BL47XU beam line with the objective of analyzing the chemical bonding state of the N at the interface. The incident X- ray energy was set to 7944 ev and the photoelectron takeoff angle was set to 89. Due to the high kinetic energy of these photoelectrons(at least 6 kev), information can be obtained simultaneously from the MOS interface, SiO 2 bulk, and SiC crystals when the SiO 2 film is less than 10 nm. Therefore, hydrofluoric acid (HF) etching was carried out before the analysis to reduce the maximum SiO 2 thickness to 10 nm. The XPS analysis utilized a PHI-5500 device and the MgKα line X-ray source. The photoelectron take-off angle was set to 45. As XPS has the same measurement principle as HAXPES, the chemical bonding state of N can be analyzed. Because the incident X-ray energy of XPS is smaller than that of HAXPES, all SiO 2 was removed by HF wet etching. Therefore, the measurement depth is around 1 to 2 nm from the surface of the SiC crystal, and this research attempted to separate the information obtained from the SiO 2 bulk and SiC crystals by subtracting the XPS data from the HAXPES data. The XAFS analysis utilized the Aichi SR BL-7U beam line X-ray source and electron yield X-ray absorption spectrum measurement method with the objective of analyzing the chemical bonding state of N by a different analysis principal from HAXPES and XPS. The measured XAFS spectra were compared to XAFS spectra obtained by first-principle calculation, and the bonding states of the N were estimated. All SiO 2 was removed by HF wet etching and the interfaces were directly measured. III. EXPERIMENTAL RESULTS A. SIMS measurements Figure 1(a) shows the N profile SIMS measurement results of the samples annealed in NO at 1000, 1150 and 1300 C for 30 minutes. All the N concentration peaks are set at 0 nm. The measured data of Si, C, and O (not shown) confirmed that N segregates on the MOS interface. When the NO-POA process time was set to 30 minutes, the density and concentration of N in both the SiO 2 film and the interface increased in accordance with the annealing time. Figure 1(b) shows the N profile SIMS measurement results of the m-face samples annealed in NO at 1300 C for 30 and 90 minutes, and the Si-face sample annealed in NO at 1300 C for 30 minutes as a reference. The measured MOS interface N densities are summarized in Table I. When the m-face sample was annealed in NO m-face (a) 1300 C 1150 C 1000 C 1300 C m-face 90 mins 30 mins (b) Si-face 30 mins FIG. 1. SIMS measurement results of samples treated under the conditions of (a) NO-POA in NO (10%) gas at 1 atm at 1000 C (colored in red), 1150 C (colored in blue) and 1300 C (colored in green) for 30 minutes on m-face wafers after forming 75 nm SiO 2 film by LP-CVD and (b) NO-POA in NO (10%) gas at 1 atm at 1300 C for 30 minutes (colored in black) and 90 minutes (colored in orange) on m-face and 30 minutes (colored in blue) on Si-face wafer. at 1300 C, the N density became saturated at 30 minutes and the saturation value was cm 2. In contrast, when the Si-face sample was annealed under the same conditions, the N density reached cm 2, which is about half of the m-face saturation density. This Si-face N density is almost the same as the reference data mentioned in the SIMS description above. Therefore, the SIMS measurement error in this research is also expected to be less than 20%. Dhar et al. reported that the N densities of a Si-face, C-face, and m-face annealed in NO at 1175 C became saturated within 30 minutes and that each face has a different saturation value [16]. This study found that the m-face has the same N density saturation behavior as the other faces and that the m-face satura (J-Stage:

3 e-journal of Surface Science and Nanotechnology Volume 15 (2017) TABLE I. MOS interface nitrogen density dependence on NO- POA conditions. 4H-SiC NO-POA N interface density crystal face temperature and time ( atoms/cm 2 ) m-face (1 100) 1300 C, 30 mins 9.86 m-face (1 100) 1150 C, 30 mins 8.13 m-face (1 100) 1000 C, 30 mins 5.34 m-face (1 100) 1300 C, 90 mins 9.70 Si-face (0001) 1300 C, 30 mins 4.80 tion value of cm 2 is unique. The experimental results in Fig. 1(a) appear to show that the saturation value has temperature dependence. However, since it has been reported that N density saturation values are not temperature-dependent at the Si-face, C-face, and m- face [16], it is likely that the N densities after 1000 C and 1150 C annealing for 30 minutes had not yet reached saturation point. In addition, although the half value width is almost 3.4 nm in the SIMS measurement results shown in Fig. 1(a) and (b), the N distribution area is expected to be narrower than 1 nm because of SIMS resolution limitations and previously reported information [16]. B. HAXPES and XPS measurements Figure 2 shows the Si 1s, O 1s, C 1s, and N 1s photoelectron spectra obtained by HAXPES. With respect to the binding energy axis, the SiC peak position of the 1000 C sample was determined to be ev by fitting the SiO 2 peak of the Si 1s spectrum to the value quoted in the literature [17]. The SiC peaks of all the other samples were assumed to have the same binding energy. Consequently, the spectra of the other elements were shifted in accordance with the difference between the above value and the measured values. After carrying out this calibration, the SiC peak positions in the C 1s spectra of every sample matched, thereby confirming the accuracy of this analysis. The peak detected at ev in the N 1s spectra was assumed to be Si N bonding [8, 17, 18] and all samples had the same Si N bonding status. In these graphs, each bonding energy peak value is set as 1. The noise in the N 1s spectrum of the 1000 C sample is due to the greater thickness of the remaining SiO 2 compared to the other samples, as evident from the SiO 2 peak height of the Si 1s signal. In addition, the N density itself is smaller than the other samples, as identified from the SIMS measurements. These results confirmed that Si N bonding is dominant for N. Figure 3 shows the N 1s HAXPES photoelectron spectra of the 1000, 1150, and 1300 C samples. Because the ev spectrum indicates that N forms three chemical bonds with Si, it is assumed that most of the N at the SiO 2 /SiC interface replaced C. Furthermore, since all the N 1s HAXPES spectra sweep toward the higher energy side, the N 1s spectra were fitted using a set of three Gaussian representations FIG. 2. Si 1s, O 1s, C 1s, and N 1s photoelectron spectra obtained by HAXPES. Samples treated under the conditions of NO-POA at 1000 C (colored in red), 1150 C (colored in blue) and 1300 C (colored in green) for 30 minutes. with different bonding configurations known as main Si N bonding (397.8 ev), Si 2 -N-O bonding (398.3 ev), and Si-N-O 2 or C N bonding (399.4 ev) [19]. Samples at all the NO annealing temperature showed a Si 2 -N-O bonding signal strength of around 10% to 20% of the main Si N signal value, and the existence of Si 2 -N-O bonding with a similar ratio may be assumed on the interface. Figure 4 shows the N 1s XPS photoelectron spectra of the 1000, 1150, and 1300 C samples. These spectra were also fitted using the same three peak values as adopted for the HAXPES analysis. Regardless of the bonding states, most of the N is expected to exist at the SiC crystal side on the SiO 2 /SiC interface because the XPS spectra are similar to the HAXPES spectra at each annealing temperature. Furthermore, the XPS spectra, which do not include information from the SiO 2 film side, showed less temperature dependence. However, the HAXPES spectra showed a slight temperature dependence at high energy bonding states. This result indicates that the N bonding states in the SiO 2 bulk were more affected by temperature than the bonding states at the SiO 2 /SiC interface within a temperature range from 1000 C to 1300 C. C. XAFS measurements In Fig. 5, the lower three lines are the XAFS experimental results and the upper six lines are the XAFS spectra of several assumed N statuses obtained by first-principle calculation as references to identify the XAFS experimental results. The XAFS spectra of samples with a NO-POA temperature of 1150 and 1300 C closely resemble the calculated XAFS spectra of Si 3 N 4 in which N replaces the C in SiC crystals. The XAFS spectrum of the 1000 C annealing sample has only a distant resemblance to the calculated references. This is probably due to the lower N interface density, which is not sufficient to absorb the characteristic X-ray energies of N bonding. In contrast, characteristics such as the replacement of Si in the SiC (J-Stage: 111

4 Volume 15 (2017) Hamada, et al. (a)1300 C (b)1150 C (c)1000 C FIG. 3. HAXPES N 1s spectra (colored in black) is divided into possible peak assignments of Si N bonding (397.8 ev, colored in green), Si 2 -N-O bonding (398.3 ev, colored in blue) and Si-N-O 2 or C N bonding (399.4 ev, colored in red). (a) 1300 C, (b) 1150 C, and (c) 1000 C (a)1300 C (b)1150 C (c)1000 C FIG. 4. XPS N 1s spectra (colored in black) is divided into possible peak assignments of Si N bonding (397.8 ev, colored in green), Si 2-N-O bonding (398.3 ev, colored in blue) and Si-N-O 2 or C N bonding (399.4 ev, colored in red) for comparison to HAXPES spectra (a) 1300 C, (b) 1150 C, and (c) 1000 C IV. DISCUSSION FIG. 5. Calculated XAFS spectra (upper six lines) and experimental spectra (lower three lines). crystal by N or interstitial N are not observed at all. Incidentally, the XAFS N spectrum overlaps the XAFS C spectrum and the peak at around ev is in the XAFS spectrum, which does not appear in the obtained XAFS spectra. The XAFS analysis results show that NO-POA process conditions of 1000 C and 30 minutes are not enough to complete nitridation of the SiO 2 /SiC interface. N on the interface forms chemical bonds mainly with Si, and N is likely to replace C in the SiC crystal. These results agree closely with the results of SIMS, HAXPES, and XPS. This research investigated the N state of a SiO 2 film deposited by LP-CVD on the 4H-SiC m-face after NO- POA. It was found that most of the N segregates only on the SiO 2 /SiC MOS interface, most of the N combines with Si, Si-N bonding is dominant for the N on the interface, N is likely to replace C in the SiC crystal, and the N density saturates. The segregated N distribution area is very thin, and is expected to be narrower than 1 nm. These m-face characteristics are similar to the Si-face, C- face, and a-face. Each face has a unique saturation value, the saturation value of the m-face was identified for the first time as cm 2. The N saturation densities of NO-POA at 1 atm were previously reported to be cm 2 [15] and cm 2 on the Si-face and cm 2 on the C- face [16], which can be compared to the amount of monolayer C on each face. The calculated amount of monolayer C on each face is shown in Table II. The ratio of N to C on the Si-face is almost one-third and the ratio on the C-face is almost one. In the case of the a-face, the saturation N density was reported to be cm 2 [16] and the ratio of N to C on the a-face is also almost one. Pennington et al. reported that the energy per surface cell on the Si-face became smallest when the amount of N near the surface is one-third of the amount of the C mono-layer [20]. This report demonstrated the possibility that the saturation N density is decided by the existence probability of N atoms. In contrast, neither the saturation amount nor saturation mechanism of N on the m-face (J-Stage:

5 e-journal of Surface Science and Nanotechnology Volume 15 (2017) TABLE II. Measured N density, mono-layer C density, and the ratio of N to C on each face. 4H-SiC N density(cm 2 ) C density(cm 2 ) Ratio of N to C crystal face experimental result on top of surface (N/C) Si-face (0001) 4.8 / / 0.29 C-face (000 1) a-face (11 20) m-face (1 100) each plane that includes C atoms as the 1st, 2nd, and 3rd internal plane from the top surface. Each plane is parallel with the m-face and exists at equal intervals in the vertical direction of the m-face. Based on the experimental results, C atoms on both the 1st and 2nd planes are likely to be replaced by N simultaneously during NO-POA of the m-face. FIG. 6. 4H SiC m-face ball and stick model with the 1st, 2nd, and 3rd internal planes within one unit cell. Yellow balls show carbon atoms and gray balls show silicon atoms. The 1st plane is a usual m-face. has been reported. In this experiment, the nitrogen density saturation value on the m-face measured by SIMS was cm 2. Since the C density on the top surface of the m-face is cm 2, the ratio of N to C is approximately 1.5. Therefore, it is likely that the saturation N density of the m-face is determined by a different mechanism than the other faces. Figure 6 shows a 4H SiC m-face ball and stick model. Four C and Si atoms are included in one 4H SiC unit cell and, looking at the layout of the C atoms on the m-face, C atoms repeatedly appear in a 2:1:1 rotation. For convenience, this paper defines V. CONCLUSIONS The N states on the SiO 2 /SiC interface after NO-POA were investigated in detail by SIMS, HAXPES, XPS, and XAFS. It was found that most of the N combines only with Si at the SiC crystal side, which may be established by replacing C on the interface with N. The saturation N density of the m-face is around cm 2, 1.5 times that of the C density on the m-face top surface. This is the first report to identify that the Si-face, C-face, a-face, and m-face have different saturation N densities and a different N to C ratio. Furthermore, this paper also showed that the m-face has three internal planes that appear at equal intervals in the vertical direction of the m-face, which include a C ratio of 2:1:1. It was suggested that the unique saturation N density of the m-face may be decided by the concurrent nitridation reaction of the 1st and 2nd planes during NO-POA. [1] H. Morkoç, S. Strite, G. B. Gao, M. E. Lin, B. Sverdlov, and M. Burns, J. Appl. Phys. 76, 1363 (1994). [2] T. Kimoto, Jpn. J. Appl. Phys. 54, (2015). [3] H. Yoshioka, J. Senzaki, A. Shimozato, Y. Tanaka, and H. Okumura, Appl. Phys. Lett. 104, (2014). [4] G. Y. Chung, C. C. Tin, J. R. Williams, K. McDonald, R. K. Chanana, R. A. Weller, S. T. Pantelides, L. C. Feldman, O. W. Holland, M. K. Das, and J. W. Palmour, IEEE Electron Device Lett. 22, 176 (2001). [5] J. Rozen, S. Dhar, M. E. Zvanut, J. R. Williams, and L. C. Feldman, J. Appl. Phys. 105, (2009). [6] J. Rozen, X. G. Zhu, A. C. Ahyi, J. R. Williams, and L. C. Feldman, Materials Science Forum 645, 693 (2010). [7] J. Rozen, A. C. Ahyi, X. Zhu, J. R. Williams, and L. C. Feldman, IEEE Trans. Electron Devices 58, 3808 (2011). [8] R. Kosugi, T. Umeda, and Y. Sakuma, Appl. Phys. Lett. 99, (2011). [9] H. Yano, H. Nakao, H. Mikami, T. Hatayama, Y. Uraoka, and T. Fuyuki, Appl. Phys. Lett. 90, (2007). [10] T. Nakamura, Y. Nakano, M. Aketa, R. Nakamura, S. Mitani, H. Sakairi, and Y. Yokotsuji, 2011 IEEE International Electron Devices Meeting (IEDM), (2011). [11] H. Takaya, J. Morimoto, K. Hamada, T. Yamamoto, J. Sakakibara, Y. Watanabe, and N. Soejima, th International Symposium on Power Semiconductor Devices and IC s (ISPSD), 43 (2013). [12] D. Peters, R. Siemieniec, T. Aichinger, T. Basler, R. Esteve, W. Bergner, and D. Kueck, th International Symposium on Power Semiconductor Devices and IC s (ISPSD), 239 (2017). [13] S. Nakazawa, T. Okuda, J. Suda, T. Nakamura, and T. Kimoto, IEEE Trans. Electron Devices 62, 309 (2015). [14] J. Rozen, in: Physics and Technology of Silicon Carbide Devices, Ed. Y. Hijikata (InTec, 2012), p [15] K. McDonald, L. C. Feldman, R. A. Weller, G. Y. Chung, C. C. Tin, and J. R. Williams, J. Appl. Phys. 93, 2257 (2003). (J-Stage: 113

6 Volume 15 (2017) Hamada, et al. [16] S. Dhar, L. C. Feldman, K.-C. Chang, Y. Cao, L. M. Porter, J. Bentley, and J. R. Williams, J. Appl. Phys. 97, (2005). [17] Y. Xu, X. Zhu, H. D. Lee, C. Xu, S. M. Shubeita, A. C. Ahyi, Y. Sharma, J. R. Williams, W. Lu, S. Ceesay, B. R. Tuttle, A. Wan, S. T. Pantelides, T. Gustafsson, E. L. Garfunkel, and L. C. Feldman, J. Appl. Phys. 115, (2014). [18] R. I. Hegde, P. J. Tobin, K. G. Reid, B. Maiti, and S. A. Ajuria, Appl. Phys. Lett. 66, 2882 (1995). [19] K. Y. Cheong, S. Dimitrijev, J. Han, and H. B. Harrison, J. Appl. Phys. 93, 5682 (2003). [20] G. Pennington and C. R. Ashman, Appl. Phys. Lett. 91, (2007) (J-Stage: