Study on Artificial Intelligence Non-Destructive Characterization of SiC Single Crystals

SiC wafers are available for high-frequency and high temperature electronic devices, more wafer specifications please see https://www.powerwaywafer.com/sic-wafer.

Silicon carbide (SiC) has been widely used in high-voltage, high-frequency, and high-temperature electronic devices due to its excellent electrical properties. However, traditional SiC crystal PVT growth is considered a “blind box”, and crystal quality and process reliability are easily affected by various factors. Currently, the characterization and quality evaluation of defects in SiC single crystals can only rely on various destructive tests. These traditional methods not only cause loss of crystal materials, time-consuming and labor-intensive, but also make it difficult to trace and present the growth process of SiC crystals in an intuitive and multi-dimensional way. This is also one of the reasons why the yield of SiC single crystals is currently low and the substrate cost remains high. Therefore, there is an urgent need to develop a non-destructive, fast, and multi-dimensional detection method to identify and trace the formation and evolution of internal defects in large-sized ingots, in order to guide the optimization of growth processes.

In response to the above challenges, a research team has successfully developed a method based on micro CT scanning and artificial intelligence (deep learning) enhancement based on the modulation characteristics of defects in SiC crystals on X-rays, achieving fast and non-destructive characterization of large-sized (≥ 6 inches) 4H SiC ingots. The team obtained image features of potential micrometer level defects in SiC crystals, such as microtubules, polytypes, and carbon inclusions, through comprehensive multidimensional scanning of SiC crystals (Fig. 1). Furthermore, by combining traditional optical methods for characterization, the identification relationship between image features and defect types in scanned images was confirmed.

Fig. 1 4H-SiC crystal defect image features by non destructive micro CT scanning

By building an artificial intelligence architecture and utilizing convolutional neural network image recognition technology, the above defect image features were trained and verified, achieving fast and accurate recognition and localization of SiC defects in complex CT images assisted by artificial intelligence (Fig. 2).

Fig. 2 Process and architecture of artificial intelligence recognition of 4H-SiC defects

The experimental results show that the training model can achieve recognition and localization accuracy of over 96% for typical defects at the micrometer scale in 4H-SiC crystals, including polytypes, microtubules, and carbon inclusions. At the same time, this method achieves three-dimensional digital reconstruction of crystals, intuitively displaying the spatial distribution and related evolution process of different micro defects inside the crystal (Fig. 3). The quantitative analysis based on three-dimensional digital reconstruction also provides technical support for the “digital twin” of subsequent SiC substrate preparation.

Fig. 3 Visual evolution process and quantitative analysis of SiC crystal defects obtained through artificial intelligence enhanced characterization

The non-destructive characterization technology of 4H-SiC material enhanced by artificial intelligence has achieved rapid, intuitive, and effective evaluation of its microscopic defects, and revealed the defect evolution during SiC crystal growth process through vivid digital reconstruction, providing important guidance for the optimization of PVT growth process. It is worth mentioning that this method is not only applicable to SiC, but can also be extended to the non-destructive and rapid characterization and growth kinetics reproduction of other high-value semiconductor single crystal materials such as AlN and Ga2O3, improving wafer yield and promoting rapid development in the field of semiconductor material preparation.