Silicon Carbide power devices are seeing a tremendous increase in demand and adoption that had been forecast at least a decade back. This is fueled to a great degree by the adoption of electric cars and power electronics needed both inside and supporting the charging infrastructure. Along with the automotive industry comes clear spotlight on costs, quality and efficiency. This in turn results in intense pressure and scrutiny to improve yields and reduce defects in every part of the manufacturing chain. Even with all the recent advances in substrate and epitaxial quality, a substantial portion of device failures still come from extended defects in the epitaxial layers, which are typically nucleated from dislocations in the substrates. The reality is that every SiC wafer is scanned multiple times; before epi, after epi, inline during fabrication, along with each device receiving extensive stress tests to ensure quality by design [1].Killer extended defects (KD) in the epitaxial layers are well understood [1]. With newer epi processes getting close to 0.1 cm-2 – 0.2 cm-2, the focus shifts to other propagated or nucleated extended defects in the epitaxial layers. The current production substrate scans do not reveal enough data on the underlying dislocations to provide a comprehensive picture of the formation of these extended defects. Techniques like molten KOH-etching and X-ray diffraction imaging (XRDI), also known as X-ray topography (XRT) can provide an insight into the substrate dislocations. The use of XRT has recently accelerated [2-8] to study the dislocations, albeit in limited R&D settings. This is due to the large amount of time needed for both Synchrotron X-ray scans [2,3,4,5] and dedicated development systems [6,7,8]. In this work, we analyze SiC wafers with an XRT system with high throughput, designed for compatibility with SiC production environments, while comparing dislocation detection to molten KOH etching and statistically tracking the nucleation / propagation of extended defects in grown epitaxial layers.Each type of scan of a SiC wafer whether of the surface, photoluminescence, XRT, Etch Pits, etc. gives a different view of the defects in the wafer. This often provides complementary information, which has to be interpreted and analyzed in an integrated way to understand the complete picture. SiC wafers from different vendors were scanned using XRT along with automated classification of all the dislocations detected. Some of the wafers were etched by molten KOH to correlate these two views. Further, many of the wafers were used to grow epitaxial layers and the extended defects detected were correlated back to the detected dislocations. As this is a probabilistic process, and not a 1:1 conversion, a framework was developed to analyze these nucleation and propagation processes statistically over a large dataset.A review of all these results will be presented. The applicability of such a production tool will be discussed along with directions where such data can be used and have the maximum impact.
Read full abstract