Characterization of Leakage Causing Visible Epitaxial Defects Nucleating from Crystal Defects in the Substrate

Abstract

Silicon Carbide continues to make big strides in adoption in high voltage applications. With many device manufacturers shipping products in high volume and the transition to 150mm substrate wafers, quality and screening defects becomes a very important facet of meeting customer expectations. In our previous work [1] we outlined how to effectively screen out defects causing bipolar degradation. Many devices like Schottky barrier diodes do not have this failure mechanism. Instead major visible defects like downfalls, triangular defects, carrots are known to cause clear electrical fails. Some other classes of defects like partial triangular defects or ‘V’ type defects do not cause outright fails. They however cause increased leakage, parametric shifts and possible reliability failures after prolonged operation time. In this work, we characterize some of these ‘V’ type defects in epi, determine their impact of electrical tests, trace their origin and provide ways of effectively screening them out at their source. After growing epi on every substrate cut from a single 150mm boule, we have observed that the wafers from the start of the crystal growth tend to have a much higher number of partial triangular defects or ‘V’ defects along with the presence of stacking faults when compared to the rest of the boule. The numbers are typically ten times the usual baseline for normal wafers. This effect was observed over multiple 150mm boules. Since we have seen these ‘V’ defects very sporadically in epitaxial layers before, and know that they do not cause an outright yield fail, we fabricated Schottky barrier diodes on these wafers along with some control wafers, to determine the impact of a large population of these epi defects. The wafer sample with the large number of ‘V’ defects exhibited higher leakage and an extended tail for forward voltage (Vf) tests as compared to the control wafers. This was further confirmed by comparing the defect map to the regions of high (non-leaky) Vf. One of the possible mechanisms responsible for the formation of these defects are screw dislocations present in the substrate that act as nucleation centers [2]. Screw dislocations are always present and numerous in the order of 100-500 cm-2in any typical 150mm production grade substrates. They have been detected with some success in low doped epi and substrates using photoluminescence maps in the near infrared photoluminescence (PL) spectrum [3-6]. In our samples, it is evident that not every screw dislocation nucleates the ‘V’ epi defects, since the majority of the epi wafers do not have these defects in spite of there being a normal distribution of screw dislocations in the substrates. High resolution PL scans of various wavelength ranges were performed on substrates with the objective of determining the root cause of these epi defects. It was observed that all the substrates which resulted in a high density of the ‘V’ defects after epi, had very faint scattered PL in the near-UV (NUV) band. This PL activity was scattered across the whole wafer and had a positive contrast to the background. This PL activity was absent in the substrates which did not nucleate the ‘V’ defects after epi, and could be used as a reliable measure to predict the nucleation of the ‘V’ defects after epitaxial growth. Defect delineation etching using molten KOH was performed on both PL affected bare substrates, Epi grown on the PL affected substrates and on control substrates which did not show any PL activity. Every spot of PL activity on the affected substrates was correlated back to a screw dislocation etch pit. It is important to note that there were many other screw dislocations on both the affected substrates and the control wafers that did not exhibit PL activity. This is expected and we conclude that only a subset of the screw dislocations nucleate these ‘V’ defects in the epi. This subset always exhibits the PL signature in the NUV band and can be screened out non-destructively in the substrate itself. The etch pits in the Epi sample confirmed the structure of the ‘V’ defects with a screw dislocation at the vertex. [1] H. Das et al. ECS Transactions, 69 (11) 29-32 (2015) [2] H. Wang et al. Materials Science Forum, Vols. 778-780, 332 (2014) [3] M. Tajima et al. Applied Physics Letters, 86, 061914 (2005) [4] C. Kawahara et al. Japanese Journal of Applied Physics, 53, 020304 (2014) [5] M. Nagano et al. Materials Science Forum, Vols. 778-780, 313 (2014) [6] K. X. Liu et al., Materials Science Forum, Vols. 600-603, 345 (2009)