APPLIED PHYSICS LETTERS

VOLUME 83, NUMBER 24

15 DECEMBER 2003

Core structure and properties of partial dislocations in silicon carbide p - i - n diodes S. Ha, M. Benamara, and M. Skowronskia) Department of Materials Science and Engineering, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, Pennsylvania 15213

H. Lendenmann ABB Corporate Research, SE-721 78, Va¨stera˚s, Sweden

共Received 22 August 2003; accepted 20 October 2003兲 The electroluminescence, mobility, and core nature of partial dislocations bounding stacking faults in 4H silicon carbide p-i-n diodes were investigated using optical emission microscopy and transmission electron microscopy 共TEM兲. The stacking faults developed and expanded in the blocking layer during high current forward biasing. Their bounding partial dislocations showed two distinct characteristics. Bright luminescent segments were mobile while dark invisible ones were stationary during biasing. TEM analysis of their Burgers vectors indicated that the mobile segments were silicon-core 30° partial dislocations while the immobile segments were carbon-core 30° ones. © 2003 American Institute of Physics. 关DOI: 10.1063/1.1633969兴

Silicon carbide 共SiC兲 is the semiconductor of choice for future electronic devices in high voltage applications. Among its various polytypes, 4H receives the most attention due to the high electron mobility and the availability of large diameter substrates. However, the commercialization of SiC high power diodes has been hampered by the degradation of their forward characteristics.1 This effect is due to the formation and expansion of stacking faults 共SFs兲 in the blocking layer. The SFs were identified as the Shockley type which expands through the glide of bounding partial dislocations 共PDs兲.2 There have been several reports on morphology of various SFs in the diodes,3–5 but the driving force for their expansion and properties of bounding PDs are still in debate. In this study, we investigated the properties of PDs and especially the relationship between luminescence, mobility, and atomic structure of the core. This information is essential to identify the driving force of the degradation. The p-i-n diodes examined in this study were fabricated on a 35 mm diameter, 4H-SiC, n-type substrate (n⫽5 ⫻1018 cm⫺3 ) purchased from Cree, Inc. 共Durham, NC兲. The substrate was off-cut by 8° from the 关0001兴 toward the 关1120兴 direction and the diodes were processed on the silicon face. The low-doped (n⬃1015 cm⫺3 ) n ⫺ blocking layer 共30– 40 ␮m兲 and the p-type anode were deposited by horizontal hot-wall chemical vapor deposition method. Standard metal contacts were formed on the surface and backside. Diode mesas were defined by reactive ion etching. The detailed design and fabrication procedures of the diodes have been described elsewhere.1 The top contact layer was formed in a grid pattern with windows making it possible to observe dislocations in the active layer by electroluminescence. The optical emission microscopy 共OEM兲 was performed using a liquid-nitrogen cooled 共⫺100 °C兲, UV sensitive camera and an optical microscope mounted on a probe station. The OEM images were a兲

Electronic mail: [email protected]

collected during the forward biasing of diodes at the current density of 10 mA/cm2 with the exposure time of 20 min. High current biasing was performed between collections of images by applying a current density between 0.1 and 10 A/cm2 . Under these conditions, the evolution of defects was recorded for later comparison with transmission electron microscopy 共TEM兲 results. The diodes for TEM observation were biased in small increments such that SFs remained small 共width⬍100 ␮m兲 and well-separated from each other. The selected diodes were cut from the wafer after biasing for plan-view TEM and metal contact layers were removed in boiling HCl:H2 O2 (30%):H2 O⫽1.5:1.5:5 共volume fraction兲. The samples were lapped and dimpled from the substrate side in order to analyze the segments of PDs close to the diode surface. The electron transparency was finally obtained by ion milling. The locations of dislocation segments could be traced throughout the sample preparation by comparing the diode’s OEM images with the optical micrographs of the sample taken in each step of preparation. Using this technique, we could correlate each PD and SF observed in TEM with those recorded in OEM. It is well-established that dislocations in the p-i-n diode structure can be observed due to electroluminescence during forward biasing.4,5 Electrons and holes recombine preferentially at dislocations and a part of their energy is released in the form of light. Figure 1 shows four OEM images of dislocations, 共a兲, 共b兲, 共d兲, and 共e兲 where they appear as bright lines and dots. Since the images are top views of the diodes, the bright lines correspond to basal plane dislocations while the bright dots to threading dislocations. The black areas around the edges of the figures are the top contacts while the gray areas in the middle correspond to SiC seen through the metallization windows. Figure 1共a兲 is the first OEM image of a diode taken before any high current biasing. At the current density used for imaging (10 mA/cm2 ), no dislocation motion was observed. Figure 1共b兲 was taken from the same location after 10 min of biasing at 1.0 A/cm2 . A new wedge-

0003-6951/2003/83(24)/4957/3/$20.00 4957 © 2003 American Institute of Physics Downloaded 11 Dec 2003 to 128.2.52.235. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/aplo/aplcr.jsp

4958

Appl. Phys. Lett., Vol. 83, No. 24, 15 December 2003

Ha et al.

FIG. 1. Optical emission microscopy images of stacking fault development in 4H-SiC p-i-n diodes: 共a兲, 共d兲 The ‘‘virgin’’ diode images before high current biasing, 共arrow⫽off-cut direction兲; 共b兲, 共e兲 After high current biasing 共scale bar⫽50 ␮m兲; 共c兲, 共f兲 Schematic diagrams of the rhombic 共a兲, 共b兲 and the rectangular 共d兲, 共e兲 stacking fault development.

shape bright line appeared in the middle next to the bright spot. Its morphology is illustrated in 共c兲. The solid line represents the new bright line. It started from the bright spot as a small wedge and grew in the direction marked by an arrow. This direction is approximately parallel to the long diagonal of the rhombus made by the wedge and the spot. It is known that the rhombic area swept by the bright wedge is occupied by a SF which causes a degradation of the diode’s forward characteristics.1 A rhombic SF is bounded by a PD loop with a single Burgers vector.4 The dislocation segments represented by the solid and broken lines in 共c兲 have different luminescence and mobility even though they constitute a single loop. The solid line segments moved and emitted light while the broken ones were immobile and nonradiative. Figures 1共d兲, 1共e兲, and 1共f兲 show another example of a currentinduced SF. The fault originated from a straight bright line along the off-cut direction, seen in Fig. 1共d兲 obtained before high current biasing. After high current biasing, the bright line shifted downward leaving a trail on its right end as seen in 共e兲. As is illustrated in 共f兲, the shaded area swept by the bright line is a rectangular SF. The luminescent right edge is a PD segment near the intersection between the basal plane containing the SF and the diode surface. Twigg et al.5 reported that SFs stop their expansion at about 100 nm from the diode surface. Similar rectangular SF was also reported by Jacobson et al.3 They showed that such faults form by dissociation of a pure screw dislocation and are bounded by two PDs with different Burgers vectors. In Fig. 1共f兲, the solid and broken lines correspond to two PDs with different luminescence and mobility. Similarly as in Fig. 1共c兲, the solid line denotes the mobile and optically active PD while the dashed line denotes the immobile and nonradiative one. In order to determine the core structure of PDs, we applied plan-view conventional TEM. Figures 2 and 3 show an example of such analysis. A SF is shown in Fig. 2共a兲 as a plane consisting of alternating black and white contrast lines. The lower left and upper right edges of the SF plane are the intersections of the SF with the sample surfaces. The top edge is bounded by a PD. The fault contrast is extinguished in 兵11-20其 reflections where only the bounding PD along the

关11-20兴 direction can be seen. The diffraction conditions used for the bright-field images are shown in the insets, 关0001兴 zone axis for 共a兲 and g⫽共1-210兲, 共2-1-10兲, 共11-20兲, 共-12-10兲 for 共b兲, 共c兲, 共d兲, and 共e兲, respectively. The Burgers vector of the PD was determined by applying both g"b and oscillating contrast analysis. A mixed or a pure edge dislocation becomes invisible when both conditions, g"b⫽0 and g"共bÃu兲⫽0 are satisfied where g, b, and u are the diffraction vector, Burgers vector, and line sense vector of the dislocation, respectively. For the PDs considered here, the g"共bÃu兲⫽0 for any g in the basal plane and the invisibility criteria reduces to one condition, g"b⫽0. This condition is satisfied in 共c兲. Thus, the Burgers vector is either a/3关01-10兴 or a/3关0-110兴 among six possible vectors of the a/3具1-100典 type. As a consequence, the PD is a 30° dislocation. In order to differentiate these two possibilities, we applied oscillating contrast analysis proposed by Marukawa.6 Figures 3共a兲 and 3共c兲 are enlarged images of the right end of the PD in Figs. 2共b兲 and 2共e兲, respectively. The dislocation line is composed of regions of dark and bright contrast, which curve and oscillate along the line. For example, in Fig. 3共a兲, a region of bright contrast is situated above a region of dark contrast at the dislocation end, which is reversed in the next portion of the line. This oscillating contrast is asymmetric 共bright versus dark兲 across the dislocation line and symmetric between the line ends in the diffraction conditions of Fig. 3. When following a dislocation along its positive line sense 共u兲, if the dark contrast appears on the left side of the line at its ends, then g"b⬎0 in that diffraction condition. If the dark contrast is on the right side, then g"b⬍0. The analyses of Figs. 3共a兲 and 3共c兲 are depicted schematically in Figs. 3共b兲 and 3共d兲, respectively. The Burgers vector as determined from each image was a/3关0-110兴. The definition of the Burgers vector used here is the same as that used by Hirth and Lothe.7 The vector product uÃb gives the direction of the extra half plane, which is related to the edge component of the dislocation. The cross-product points into the plane of the figure for the PD in Fig. 3, which is toward the carbon face of the sample. Since the PDs discussed here should belong to glide set,7 which has the glide plane between the closely

Downloaded 11 Dec 2003 to 128.2.52.235. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/aplo/aplcr.jsp

Ha et al.

Appl. Phys. Lett., Vol. 83, No. 24, 15 December 2003

4959

FIG. 3. Oscillating contrast analysis of a partial dislocation: 共a兲, 共c兲 Enlargements of the right end of the dislocation line in Figs. 2共b兲 and 2共e兲, respectively 共scale bar⫽0.5 ␮m兲. 共b兲, 共d兲 Analysis results of the images, 共a兲 and 共c兲. The Burgers vector 共b兲 direction along its axis 共dashed line兲 is determined based on the selected line sense 共u兲, diffraction vector 共g兲 and the corresponding sign of g"b.

FIG. 2. TEM analysis of a stacking fault and a bounding partial dislocation 共bright-field images兲: The inset of each image is the diffraction pattern. 共a兲 The stacking fault image taken along the 关0001兴 zone axis; 共b兲, 共c兲, 共d兲, 共e兲 The partial dislocation in different diffraction conditions, g⫽共1-210兲, 共2-110兲, 共11-20兲, 共-12-10兲, respectively. The g vectors point from the center to the outer spot in the insets. g"b⫽0 in 共c兲.

spaced silicon and carbon atom layers, the core is composed of only carbon atoms if the extra half plane extends toward the carbon face of the crystal. This dislocation is denoted as C(g). 4 We have analyzed more than 20 PDs in TEM experiments and determined their Burgers vectors and cores 关 Si(g) or C(g)] by the method described earlier. All mobile PDs with bright luminescence were 30° Si(g) and all dark immobile ones were 30° C(g). We also have applied the large angle convergent beam electron diffraction method and obtained the same conclusion.8 This result agrees with the calculations of Blumenau et al.9 Using a density-functionalbased tight-binding total energy calculation and a pseudopotential methods, these authors concluded that Si–Si bonds along the core of 30° Si(g) PD have energy states in

the band gap and serve as electron-hole recombination sites while C–C bonds of 30° C(g) PD do not. The Si(g) PD could, therefore, emit light and become mobile under forward bias. On the contrary, Twigg et al.5 reported no correlation between PD core structure and its mobility and optical activity. For example, they designated different stationary PDs as Si(g) and C(g). It also appears that they used a definition of Si(g) and C(g) different from ours. For example, they observed a PD with the extra half plane toward the carbon face, which they designated as a Si(g). This is the same configuration as the PD in Fig. 2 which we call a C(g). Our nomenclature is consistent with that used by Ning et al.10 and Blumenau et al.9 This work was supported in part by ONR Contract No. N00014-02-C-0302 monitored by Dr. Harry Dietrich. 1

H. Lendenmann, F. Dahlquist, N. Johansson, R. So¨derholm, P. A. Nilsson, J. P. Bergman, and P. Skytt, Mater. Sci. Forum 353–356, 727 共2001兲. 2 J. Q. Liu, M. Skowronski, C. Hallin, R. So¨derholm, and H. Lendenmann, Appl. Phys. Lett. 80, 749 共2002兲. 3 H. Jacobson, J. Birch, R. Yakimova, M. Syva¨ja¨rvi, J. P. Bergman, A. Ellison, T. Tuomi, and E. Janze´n, J. Appl. Phys. 91, 6354 共2002兲. 4 M. Skowronski, J. Q. Liu, W. M. Vetter, M. Dudley, C. Hallin, and H. Lendenmann, J. Appl. Phys. 92, 4699 共2002兲. 5 M. E. Twigg, R. E. Stahlbush, M. Fatemi, S. D. Arthur, J. B. Fedison, J. B. Tucker, and S. Wang, Appl. Phys. Lett. 82, 2410 共2003兲. 6 K. Marukawa, Philos. Mag. A 40, 303 共1979兲. 7 J. P. Hirth and J. Lothe, Theory of Dislocations, 2nd ed. 共Krieger, Malabar, 1982兲, pp. 17–24; pp. 378 –382. 8 C. T. Chou, A. R. Preston, and J. W. Steeds, Philos. Mag. A 65, 863 共1992兲. 9 ¨ berg, T. Frauenheim, and P. R. A. T. Blumenau, C. J. Fall, R. Jones, S. O Briddon, Phys. Rev. B 共to be published兲. 10 X. J. Ning and P. Pirouz, J. Mater. Res. 11, 884 共1996兲.

Downloaded 11 Dec 2003 to 128.2.52.235. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/aplo/aplcr.jsp