Carrier Trap Density Reduction at SiO2/4H-Silicon Carbide Interface with Annealing Processes in Phosphoryl Chloride and Nitride Oxide Atmospheres

The electrical and physical properties of the SiC/SiO2 interfaces are critical for the reliability and performance of SiC-based MOSFETs. Optimizing the oxidation and post-oxidation processes is the most promising method of improving oxide quality, channel mobility, and thus the series resistance of the MOSFET. In this work, we analyze the effects of the POCl3 annealing and NO annealing processes on the electrical properties of metal–oxide–semiconductor (MOS) devices formed on 4H-SiC (0001). It is shown that combined annealing processes can result in both low interface trap density (Dit), which is crucial for oxide application in SiC power electronics, and high dielectric breakdown voltage comparable with those obtained via thermal oxidation in pure O2. Comparative results of non-annealed, NO-annealed, and POCl3-annealed oxide–semiconductor structures are shown. POCl3 annealing reduces the interface state density more effectively than the well-established NO annealing processes. The result of 2 × 1011 cm−2 for the interface trap density was attained for a sequence of the two-step annealing process in POCl3 and next in NO atmospheres. The obtained values Dit are comparable to the best results for the SiO2/4H-SiC structures recognized in the literature, while the dielectric critical field was measured at a level ≥9 MVcm−1 with low leakage currents at high fields. Dielectrics, which were developed in this study, have been used to fabricate the 4H-SiC MOSFET transistors successfully.


Introduction
An increasing number of mobile and power electronic devices induces demand for modern, fast chargers and power modules. Silicon carbide is a promising material for producing power electronics elements such as Schottky diodes, bipolar transistors, and metal-oxide field-effect transistors (MOSFETs). These elements are applied to converters for various applications [1], ranging from motor drives for different kinds of vehicles [2], hightemperature [3][4][5][6], and high-frequency applications, to renewable energy generation [7,8]. At the present time, SiC power devices offered on the market are junction field-effect transistors (JFETs), metal-oxide-semiconductor field-effect transistors, bipolar junction transistors (BJT), and bipolar transistors with isolated gate (IGBTs). MOSFETs are particularly important due to their application in high-power and high-frequency switching devices. A major bottleneck in developing these transistors is the low level of channel carrier mobility resulting in high series resistance of a channel region. This is the most important problem in power devices, which often require large currents. The increase in channel resistance is mainly caused by electrically active structural imperfections inside the gate dielectric and at the interface between the dielectric and semiconductor, called carrier traps [9,10]. High trap about the application of POCl 3 annealing and oxidation processes, it is possible to realize dielectrics with very low trap state density and high critical fields by using proper technological treatment and a combination of POCl 3 and NO annealing processes.

Experimental Data
N-type 4H-SiC (0001) 4 • off substrates from Cree Inc. (Durham, NC, USA) (donor concentration n = 5 × 10 18 cm −3 ) with 20 µm thick, n-type-doped epitaxial layer (n = 1.1 × 10 15 cm −3 ) grown by Showa Denko Materials Co., Ltd. (Tokyo, Japan) were used in our experiments. The substrates were cleaned successively in boiling organic solvents of acetone and propanol. Next, they were etched in Piranha solution (H 2 SO 4 :H 2 O 2 :H 2 O, 1:1:1 ratio), rinsed in deionized water, deoxidized in buffered HF solution, and dried in nitrogen blow. Subsequently, for the first series of the samples, high-temperature dry oxidation in an oxygen environment at 1175 • C for 360 or 180 min was conducted. For the second series, wet oxidation in oxygen with water vapors at 1175 • C for 80 min was performed. As a result of these thermal processes, SiO 2 layers with thicknesses of about 50 or 70 nm were formed on the 4H-SiC samples. Then, specific post-oxidation annealing processes in POCl 3 and NO environments were performed, followed by annealing in N 2 for 30 min. POCl 3 vapor was supplied from a bubbler by using nitrogen flow. The POCl 3 annealing was conducted at 1000 • C for 10 min. Next, annealing was completed in pure N 2 at 1000 • C for 30 min or in NO at a temperature of 1000 • C, 1100 • C, or 1175 • C. For chosen samples, annealing in NO at 1000 • C was performed directly after oxidation. Identification symbols of the studied samples are collected in Table 1. Characterization of the Si-oxide film by thickness and refractive coefficient was applied using a spectroscopic ellipsometry technique after every annealing process. The film parameters were calculated based on the fitting of measured ellipsometric parameters using the Sellmeier and Cauchy models of the dielectric film. The next step in MOS fabrication was reactive ion plasma etching using CF 4 /O 2 plasma (rf power p = 100 W, pressure p = 50 mT) to remove silicon dioxide from the backside of the samples. Ti/Ni 10/150 nm thick films were deposited via e-beam evaporation technique on the backside for ohmic contact metallization. The ohmic contacts were formed by annealing in the Ar atmosphere at 1050 • C for 3 min. Next, photolithography and a lift-off technique were used to form capacitance structures. A Ti/Al 10/150 nm metallization was fabricated as metallic circles with diameters of 150, 200, 300, and 750 µm. The last annealing of the structures was performed in Ar at 320 • C for 5 min. Electrical methods determining carrier trap density at the oxide/semiconductor interface are based on capacitance versus voltage measurements (C-V) [38][39][40] of the MOS structure. We applied the high-low frequency method (HI-LO), which replaces an ideal characteristic from the high-frequency method (HF) with a low-frequency measured characteristic (LF) or quasi-static characteristic (quasistatic HI-LO method). This further improves the resolution of the method, eliminating a major source of error-a basic HF method uses ideal C-V characteristics for the calculation of (interface trap) MOS capacitance (C it ), which does not include any fixed or near-interfacetrap (NIT) type charges affecting C-V shape during the voltage sweep. The quasi-static HI-LO method depends on applying a frequency with period T higher than for the traps (T > τ it ) so that every interface trap responds fully to the gate voltage. Then, the interface trap capacitance C it is determined by the formula: and D it can be calculated by: where C OX is oxide capacitance, ∆C = C LF − C HF, C LF, and C HF are normalized capacitances measured at LF or HF, respectively [41]. Our oxidation and annealing procedures were applied to fabricate a gate in a planar 4H-SiC MOSFET device. The base was formed by deep implantation of Al + ions with an energy of 450 keV in the 20 µm thick epitaxial n-SiC layer; source and drain regions were donor-doped to the level of 2-5 × 10 19 cm −3 by narrow implantations of P + . After forming a graphite cap layer from a 1 µm thick photoresist film on the implanted samples, thermal activation of the dopants was conducted by annealing in a vacuum at a temperature of 1800 • C for 30 min. The mesa structure was formed with photolithography and ion-reactive ion etching to separate the MOSFETs. Wet oxidation and an optimized sequence of the two-step annealing process in POCl 3 and next in NO atmospheres were applied for gate SiO 2 formation. Ohmic contacts were produced from Ti/Ni and Ti/Al/Ti metallization layers to n-type and p-type regions, respectively. In contrast, ohmic contact formation was performed via annealing at 1050 • C for 2 min in Ar. Ti/Al layers were deposited as the gate electrodes. The channel length (L) and width (W) were designed for 5 and 100 µm, respectively. The field-effect mobility (µ FE ) was determined from the MOSFET transconductance measurements.

Results and Discussion
The obtained structures were characterized by current-voltage (I-V), quasi-static, and high-frequency C-V measurements (f = 100 kHz, amplitude of 30 mV), applying polarization from accumulation to depletion and then returning to the accumulation state. Each measurement was preceded by applying a voltage to the sample and forcing accumulation to ensure full trap occupancy at the beginning of the measurement. The length of this initial voltage stress was determined experimentally for each sample by applying stress with increasing duration. The shortest period not resulting in a further C-V characteristic shift was used during measurements. C-V characteristics for dry-oxidized and N 2 -annealed samples are shown in Figure 1a. One can see that quasi-static C-V is shifted to the left with respect to the theoretical curve.
The C-V results for samples after post-oxidation annealing in POCl 3, and NO are shown in Figure 1b. The difference between theoretical and measured C-V curves is significantly smaller than for the samples annealed only in N 2 . Such figures correspond with the interfacial trap state density in the samples, especially in the energy range near the Fermi level situated close to the conduction band. The calculated D it vs. energy can be seen in Figure 1c. The D it values obtained for the d-N2 sample are consistent with the previously reported D it values for dry oxidation [36,42,43] and, near the conduction band edge, considerably exceed 1 × 10 12 eV −1 cm −2 . Since the sample was not treated with any processes aimed at reducing trap density, it can be used as a reference sample for comparison with others. Figure 1. C−V characteristics of MOS SiO 2 /4H-SiC structures formed by using oxide formation processes: dry oxidation and N 2 annealing-(a), successive POCl 3 and NO annealing-(b); interface trap state density D it versus energy distance from conductance band edge calculated for the SiO 2 /4H-SiC structures formed by using dry or wet oxidation, and post-oxidation annealing processes in N 2 , POCl 3 , and/or NO environments-(c).
It is interesting to compare the dry and wet oxidation in terms of trap density. The kinetics of oxidation suggest that wet oxidation is preferable due to the much shorter time required to form gate oxide [44]. Some results suggest that this process can result in lower interface state densities than conventional dry oxidation [45,46]. Recently, it has been shown that, during oxidation, the wet oxidation process generates lower interface stresses [45,47]. Since hydrogen has been shown to passivate dangling bonds at the SiO 2 /SiC interface [48], and due to the amorphous structure of the oxide, those defects appear in a variety of trap energies, mostly located close to the conduction band, so wet oxidation can be used as an efficient and better process for SiO 2 formation during the oxidation of SiC. Even though our study does not fully confirm this claim, we have been focused on wet oxidation because this process consumes less energy. The wet oxidized sample is characterized by D it (at 0.2 eV below E C ) with an approximately 30% lower value than that for d-POC.
The interface trap densities D it for the samples after post-oxidation processes are presented in Figure 1c. These are compared for the energy of 0.2 eV below the conduction band and included in Table 2. The trap density D it behavior of the NO-annealed sample is consistent with previous data [20,24,49]. However, the results for the POCl 3 -annealed sample must be discussed. Although it is generally agreed that POCl 3 post-oxidation annealing improves channel mobility of the MOSFET by reducing density of interface trap states [50], the energy distribution of trap density affected by this process is unclear. Different studies provide dissimilar D it profiles [51][52][53][54][55][56][57] for POCl 3 -treated samples. Pascu et al. showed a little reduction in the trap density near the conduction band edge and almost flat D it profile for deeper levels in the bandgap. However, Okamoto et al. observed that the trap state density near the conduction band edge is reduced by approximately one order of magnitude as compared with a dry-oxidized SiC reference, and it is almost constant for energies deeper in the bandgap. The present study is roughly concurrent with our previously reported findings [27]. This result is also consistent with the theoretical research of Kobayashi et al. [58], which shows via DFT simulations that, at the upper band, close to the conduction band, carbon-related defects play a dominant role in the interface trap density profile [17,[59][60][61], and these can be reduced due to the creation of the −O 3 PCO 2 in the oxide film near the interface during POCl 3 treatment. A recent study [54], also based on the DFT computation, suggests that phosphorus incorporated near the interface oxide region destabilizes carbon defects, especially carbon dimmers, and some Si defects through structural reconstruction. It was concluded that phosphorus affects mostly traps with energies approximately 0.15-0.4 eV below the conduction band. This is consistent with the curve line shown in Figure 1c for the dry-oxidized sample after the POCl 3 treatment, designated as d-POC. The D it for the traps with energies below 0.4 eV is reasonably lower than for the reference, and almost unchanged at higher energy. The best results were obtained by combining the post-oxidation processes in the sequence of POCl 3 and NO annealing, as seen in Figure 1c. A pronounced slope in the D it profile is visible here for the samples after wet oxidation, POCl 3 annealing, and successive NO annealing. Moreover, the influence of annealing temperature on the D it profile can be observed, and 1175 • C seems to be the optimal temperature for NO annealing (w-POC-NO-3). The idea behind the sequence was to combine the processes of (1) destabilization of C defects and (2) incorporation of nitrogen into the interface via thermal diffusion. It has been shown [9,16,19] that NO effectively introduces nitrogen into the interface. The mechanism behind improving oxide quality with the annealing process with nitrogen compounds has been theoretically investigated using DFT towards shifting the trap energy associated with a given defect outside the bandgap. In these simulations, carbon-type defects and the dominant defect contributing to the high trap density close to the conduction band edge were considered [62][63][64]. Older as well as newer studies have shown that other possible types of Si-Si bond defects can be removed with nitrogen passivation [65][66][67]. Particularly, NO annealing efficiently removes active states of Si-related defects such as Si-Si bonded atoms, Si dangling bonds, and silicon vacancy [68]. These defects are responsible for introducing trap states nearer the middle of the gap, as was shown by Nakanuma et al. [68]. Numerous Si dangling bonds can be repaired under optimal thermal conditions thanks to the nitrogen reaction with Si bonds at the interface on both sides of the SiC/SiO 2 interface.
The obtained values of oxide thickness t ox , trap surface density D it , effective static charge Q eff , and flat-band potential V FB parameters of the MOS structures are collected in Table 2. Different film thicknesses are the results of different thermal and gas conditions. Ellipsometry measurements of the wet-oxidized and POCl 3 -annealed films are characterized by a thickness in the range of 73.9-86 nm and refractive index n = 1.489, which is increased in relation to the n-index of the model of thermal SiO 2 layer on Si. An increase in thickness of approximately 20 nm or a bit above was registered for the dry or wet oxide films after the POCl 3 annealing process. Such a rapid oxidation process may be partially responsible for the effective removal of carbon defects from the interface, as previously reported in [27,31,50]. This can be explained by the transformation of the oxide layer into phosphosilicate glass (PSG) during annealing. In our experiment, the process temperature was kept at a relatively low level (1000 • C), so the transition rate to PSG was slower than reported in other works. The oxygen-SiC reaction at this temperature is too slow for efficient SiO 2 formation in the timescale used in the experiment. However, it is worth mentioning that the SiC oxidation reaction in the POCl 3 ambient has been reported to have lower activation energy than in the dry oxygen [27,69].
The interface trap density D it at E C − E T = 0.2 eV obtained for the SiO 2 /4H-SiC samples annealed at different atmospheres and temperatures is illustrated in Figure 2a, while an effective surface charge density (Q eff ) is shown in Figure 2b. The most important observation that can be drawn from these results is a change in the value and sign of the effective charge value and in the sign for the sample annealed only in POCl 3 (d-POC). These can be attributed to the formation of phosphosilicate glass (PSG) in the volume of the oxide layer, where P-oxide is built into the layer, and such formation has been described earlier [53]. The increased positive effective surface charge suggests a high phosphorus concentration near the interface region, most likely achieved due to the annealing process at a temperature similar to those in experiments described by others [50][51][52][53]. The chemical reaction of POCl 3 with SiC at 1000 • C is too slow for quick oxide formation in the timescale used in our experiment. Therefore, a higher phosphorus concentration can be achieved through the diffusion process [57]. The POCl 3 interaction with SiC at the interface can be the cause of a significant reduction in the interface trap state density close to the conduction band. Moreover, the samples after POCl 3 treatment and subsequently annealed in NO at a higher temperature are characterized by the effective surface charge density (Figure 2b) close to that observed for the reference sample annealed only in N 2 . On the samples annealed in POCl 3 , it was also observed that after annealing in the NO atmosphere, the thickness of the oxide layers slightly decreased. The effect was measured with ellipsometry on additional wet-oxidized SiC samples. Results are presented in Table 3. The difference in thickness (∆t ox ) increases with NO annealing temperature, while the n-index is still high. The higher temperature causes an amplified effect on the process. Such decrease in thickness may be caused by the evaporation of some P-oxide compounds from the dielectric layer at the applied temperature. Table 3. The thickness t ox of oxide films formed by wet oxidation and POCl 3 annealing, just before and after NO annealing at different temperatures T NO ; refractive index η (at λ = 650 nm) and thickness difference ∆t ox . The t ox and η-index calculated based on ellipsometry measurements. Based on the C-V measurements, calculation of the interface trap density, and oxide thickness measurements, we suggest that the subsequent NO annealing process performed after the POCl 3 process can reverse some of the changes in the oxide structure initially created by phosphorus compounds. It leads to a decrease in P concentration in the layer and, consequently, to a decrease in the positive charge accumulated in this layer during POCl 3 annealing. Formation of the PSG-like layer during POCl 3 annealing can be the most critical step in the application of such a layer to a gate dielectric in MOSFETs, which has been studied in the last decade [25][26][27][28]. Such considerations are mainly due to two reasons: (1) it was shown that introducing phosphorus can decrease the energy barrier for the Fowler-Nordheim current that dominates the conduction mechanism at the high electric field leading to the premature dielectric breakdown [56], and (2) a high concentration of phosphorus ions in the oxide film can lead to thermal instability of the device, as some of the ions in the oxide volume can be mobile under the strong electric fields present in the MOS device during its operation, modulating the device's threshold voltage.

Before NO
The low-temperature NO annealing step developed here can reduce both problems by maintaining the desired low interfacial trap density. To support this claim, we have measured the breakdown electric field for our samples, and the results are shown in Figure 3. The critical field for the dry oxides annealed successively in N 2 or NO was near 10 MVcm −1 . A 30% lower critical field characterized the wet-oxidized structures. This drift has been previously observed [19]. On the other hand, the critical field for oxides annealed in the phosphoryl chloride atmosphere was above 10 MVcm −1 . This result is better than reported for high-temperature annealing [56], and it is concurrent with the results reported by Pascu et al. [70], taken on SiO 2 /SiC samples after the POCl 3 annealing process at a similar temperature. The processes applied for reducing the density of the trap states do not significantly affect the breakdown voltage resistance of the oxide films. Annealing processes in the phosphoryl chloride atmosphere followed by a nitric oxide atmosphere at 1175 • C allow obtaining the critical field of~9 MVcm −1 . Moreover, the MOS structures with this two-step annealed dielectric are characterized by a very low leakage current density, <3 × 10 −7 Acm −2 , at an applied stress voltage of −100 V. The successful application of the dielectric layer annealed in accordance with the developed procedure was confirmed by electrical measurements of the manufactured planar MOSFET transistors with a gate width of 100 µm, which showed the field effect mobility µ FE of the channel electrons at the level of 30 cm 2 V −1 s −1 . Such a high mobility is consistent with the results published in the review work for MOSFETs with the dielectric gate passivated via NO annealing at high temperature. This confirms the usefulness of the proposed procedures for the thermal formation of SiO 2 dielectrics in SiC MOSFETs. Some papers present µ FE mobility with a peak value close to 100 cm 2 V −1 s −1 [25,26,29], suggesting the possibility of further improvement of the channel mobility with additional optimization of technological processes during fabrication of our MOSFET structure, for example by better recovering the mono-crystalline structure of the implanted p-well region. The obtained results are consistent with other studies in this matter, which have been successively published over the last 20 years [18], until now [68,71].

Conclusions
MOS SiO 2 /4H-SiC structures were characterized using C-V measurements and ellipsometry techniques. It was pointed out that two-stage annealing of SiO 2 /4H-SiC structures in POCl 3 atmosphere at 1000 • C for 10 min and subsequently in NO atmosphere at 1100 • C or 1175 • C reduced the interface trap state density D it down to the level of 2 × 10 11 cm −2 in the vicinity of the conduction band edge. The reduction in the trap state density is related to the specific phosphorous impact on carbon-associated trap states at the SiO 2 /SiC interface and to nitrogen attaining and passivating the interface traps. The nitridation process of NO annealing at a temperature of 1100 • C or higher is more effective in combination with phosphorous inclusions in the oxide film. The results show that, by performing two-step annealing at a temperature lower than the threshold value for efficient SiC oxidation in POCl 3 and subsequently in NO, one can obtain the dielectric layer that has both low interface state density and high critical electric field, such as the one obtained via simple dry oxidation (the critical field of~9 MVcm −1 ). Additionally, it was observed that those layers have very low leakage current (<3 × 10 −7 Acm −2 at an applied stress voltage of −100 V), making them suitable for application in SiC device technology. Such layers can be used in high field conditions, where high breaking voltage and small leakage currents are required, for example, as surface passivation for junction termination extension structures, in passivation of SiC IGBT structures, or possibly as gate oxides for MOSFET and IGBT devices.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data is not publicly available due to internal regulations.

Conflicts of Interest:
The authors declare no conflict of interest.