I. INTRODUCTION

4H-S_ilicon carbide (S_iC) has attracted attention as a next-generation power semiconductor material owing to its wide band gap (3.26 eV), high critical electric field (EC = 3 MV/cm), and high thermal conductivity (4.9 W/cm·℃). These properties make it suitable for high-power, high-frequency, and high-temperature operation. Moreover, the high EC provides the significant advantages of a low on-resistance and excellent reverse blocking characteristics ^(1,2).

Several edge termination structures for S_iC power devices, such as the field plate ^(3,4), floating guard ring (FGR) ^(5,6), mesa-etched termination ⁽⁷⁾, and junction termination extension (JTE) ^(8-10), have been investigated by various groups. A reduction in breakdown voltage (BV) due to unexpected process conditions has been reported ^(11,12). To obtain stable blocking characteristics regardless of process deviations, new edge termination structures are necessary ^(13-16). It is well known that the surface charge density (Q_surf/q) at the interface between S_iC and a passivation layer affects the operation of power devices ^(17,18). Post-oxidation annealing in NO or N₂O ambient has been widely used to reduce Q_surf/q. In addition, it is a well-validated method of improving the channel mobility of n-channel S_iC metal-oxide-semiconductor field-effect transistors ^(19-21). In both the NO and N₂O processes, a negative surface charge density is formed on the n-type S_iC surface ⁽²²⁾. The presence of the Q_surf/q affects the E-field distribution along the surface at the edge termination structures. The blocking characteristics of ultrahigh-voltage devices are strongly affected by the surface charge density because of the relatively low doping concentration in the drift layer and long termination length.

The JTE is typically used in power devices owing to its simple design and processes. It exhibits a high BV when the implantation dose in the JTE region is close to the cross-sectional concentration in the drift layer ^(8,9). Although the ion implantation dose can be precisely controlled, the profiles of depletion regions in the drift layer are affected by the charge states at the S_iC/passivation layer interface and by the activation efficiency of implanted ions ^(11,23).

The FGR is also widely used as an edge termination structure because it effectively reduces E-field crowding ^(5,24). Also, gradually increasing ring-space structure for FGR was reported to obtain high breakdown voltage ⁽²⁵⁾. However, the design of FGR structures is complicated because various design and process parameters such as the ring space, ring width (W_FGR), doping concentration, and surface charge density should be considered. A narrow space is required near the main junction to transfer potential from anode to edge termination regions. In case of outer regions of termination, a relatively wide space is required for large potential drops to obtain uniform E-field distribution and high breakdown voltage. CS-FGR may not be suitable for uniform E-field distribution because of the constant value of spaces.

The ring spaces are affected by the conditions required for processes such as photolithography, hard mask etching, and high-temperature annealing to diffuse the implanted ions. For these reasons, the process conditions must be accurate and repeatable to provide a consistent ring space and blocking characteristics ⁽²⁶⁾. In addition, photoresist and hard mask layers have a non-perpendicular slope after patterning processes; consequently, the implantation window can differ from the intended pattern width at the photomask ^(27,28). Moreover, the width of the photoresist pattern may be also affected by the exposure time during photolithography processes ⁽²⁹⁾. Note that lateral straggling of implanted ions increases W_FGR by 0.2-0.3 μm, which considerably changes the depletion shape in the drift regions ⁽³⁰⁾.

In this study, we investigated the BV characteristics of edge termination structures, including the JTE, constant space FGR (CS-FGR), and gradually increasing space FGR (GIS-FGR), with a focus on process deviations. The effects of Q_surf/q and the ring space variation on the BV were analyzed, and the electrical stability was evaluated. The BV of a proposed GIS-FGR is sufficiently stable for 6.5 kV S_iC power devices against the Q_surf/q and ring space variation.

II. SIMULATION MODEL AND DEVICE STRUCTURE

Numerical device simulations were performed using the S_ilvaco TCAD simulator. The parallel electric-field dependence, Shockley-Read-Hall recombination, Auger recombination, and band gap narrowing models were used for breakdown analysis. The avalanche breakdown was simulated using the impact ionization model of Selberherr ^(31,32). Fig. 1 schematically illustrates the structures of the JTE, CS-FGR, and GIS-FGR. The thickness and doping concentration of the drift layer were 55 μm and 1 × 1015 cm^-3, respectively. The simulated BV of the ideal parallel-plane junction with these parameters was 12.4 kV.

We assumed that a passivation layer is formed by the following processes: thermal oxidation of 50-nm-thick S_iO2, post-oxidation annealing in NO ambience, and deposition of 1-μm-thick S_iO2. The length and doping concentration of the JTE ranged from 50 to 300 μm and 1.5 × 10¹⁷ to 4.5 × 10¹⁷ cm^-3, respectively. Both the CS-FGR and GIS-FGR had 50 rings and a W_FGR value of 1.5 μm. The nth ring space is determined using Eq. (1).

where S1 is the first ring space, and S_i is the increment of the ring space for the GIS-FGR. S1 of the GIS-FGR was fixed at 1 μm. In addition, S_i was varied from 0.05 to 0.3 μm to analyze the blocking characteristics.

III. SIMULATION RESULTS AND DISCUSSION

Fig. 2(a) shows the BVs of JTEs with various doping concentrations (N_JTE) without considering surface charges. We obtained a maximum BV of 10,481 V at N_JTE = 2.5 × 10¹⁷ cm^-3. The BVs for various JTE lengths (L_JTE) are shown in Fig. 2(b), and the inset shows the E-field distribution for L_JTE = 200 μm. No significant improvement was observed at L_JTE values exceeding 200 μm. Fig. 2(c) shows the Q_surf/q dependence of the BV for N_JTE = 2.5 × 10¹⁷ cm^-3 and L_JTE > 200 μm. The BV changes when the Q_surf/q value on the S_iC surface is negative because the surface charges affect the N_JTE value at the maximum BV.

Table 1-1 and 1-2 present the blocking characteristics of S_iC devices using a JTE, including the average BV and its standard deviation against N_JTE and Q_surf/q variations, respectively. The JTE structure with N_JTE = 2.5 × 10¹⁷ cm^-3 and L_JTE = 200 μm exhibited a high average BV of 10,040 V. The average BVs of the JTE structures with different N_JTE values are lower than those of the device with N_JTE = 2.5 × 10¹⁷ cm^-3. In addition, the BVs of JTE structures with various N_JTE values showed very large standard deviations of 45-57%, indicating that the BV of the JTE structure depends strongly on N_JTE.

Fig. 3 shows the BVs of CS-FGR structures with ring spaces of 1.1 to 3.8 μm. For a ring space of 2 μm, the maximum BV was 6,840 V; the corresponding E-field distribution is shown in the inset. When the ring space is narrower or wider than 2 μm, the BV of the CS-FGR structures is much lower. The ring space can change not only during the patterning process but also during the implantation processes because lateral straggling of Al ions can result in a narrow ring space ⁽¹³⁾.

Fig. 4 shows the BV versus the ring space variation (−0.3 to +0.3 μm) at Q_surf/q = 0, −1 × 10¹¹, −5 × 10¹¹, and −1 × 10¹² cm^-2. The ring space and Q_surf/q affect the BV because variation of these parameters changes the optimal point of space. However, the CS-FGR structures have a nonuniform E-field distribution, as shown in the inset, because a potential drop occurs along the CS-FGR structure.

As shown in Table 2-1 and 2-2, the blocking characteristics of the CS-FGR structures may not provide the required BVs for 6.5 kV devices.

A ring space adjacent to the main junction must be narrow to deliver potential from the main junction to edge termination structures. However, other spaces that are further from the main junction must be wide enough to afford an effective potential drop across the edge termination structures; consequently, a gradually increasing space can provide stable blocking characteristics ⁽²⁵⁾. Fig. 5 shows the BVs of simulated GIS-FGR structures with various S_i. The insets show the E-field distributions for S_i = 0.05 and 0.15 μm. When S_i = 0.15 μm, the E-field is crowded in front of the GIS-FGR structure. The GIS-FGR with S_i = 0.05 μm has a wide E-field distribution and high BV (>10 kV).

Fig. 6 shows the BVs of GIS-FGR structures versus ring space variation for various Q_surf/q values. The GIS- FGR structure has an excellent E-field distribution even when the ring space is unintentionally varied during patterning. The BV for S_i = 0.15 μm tends to increase for negative Q_surf/q values because for negative Q_surf/q, the depletion regions are extended along the surface ⁽²³⁾. By contrast, BV decreases when S_i = 0.05 μm because the E-field is distributed uniformly; consequently, it is rather sensitive than that at S_i = 0.15 μm. For a large Q_surf/q and narrow ring space, the depletion regions exhibit rapid lateral expansion, and E-field crowding occurs at the end of the final ring.

Table 3-1 and 3-2 show the average and standard deviation of the BV of the GIS-FGR structures when S_i = 0.05 or 0.15 μm for Q_surf/q and ring space variations, respectively. We obtained fairly a small standard deviation of 2% against Q_surf/q variation from 0 to -1 × 10¹² cm^-2. Despite the surface charge density and ring space variation, we obtained a high BV (>8 kV) for the GIS-FGR with S_i = 0.05 μm. We verified that the GIS-FGR is sufficiently stable for use in 6.5 kV S_iC power devices despite process deviations during photolithography, etching, and ion implantation.

IV. CONCLUSIONS

The effects of process deviations on the blocking characteristics of 6.5 kV S_iC power devices were investigated to obtain devices that are unaffected by these variations. S_imulations of a JTE structure showed that the standard deviation of the BV was 45-57% for various N_JTE values. Moreover, the presence of Q_surf/q changed the optimal N_JTE value. A CS-FGR showed standard deviations of 4-18% and an average BV of 6,750 V owing to a nonuniform E-field. For a GIS-FGR with S_i = 0.05 μm, the standard deviation was 2-8%. In addition, we obtained a maximum average BV of 10,180 V, and a high BV (>8 kV) was obtained for various ring space variation and Q_surf/q values. Our results, including the electrical characteristics and validation of the effects of process deviations, provide stable blocking capability for 6.5 kV S_iC power devices.