The growth of silicon carbide (SiC) crystals is a complex, multiphase process involving the aggregation of Si and C atoms (pictured below). The traditional physical vapor transport (PVT) method for this synthesis can be considered a “black box”, and detailed information about the quality of the as-grown crystal can usually only be obtained afterward using destructive measurements. However, these methods are time-consuming, labor-intensive, and use up raw material. Fast, non-destructive alternative characterization methods would, thus, be useful.
Wenyu Kang, Jun Yin, Xiamen University, China, and colleagues have developed a non-destructive, deep learning-enhanced characterization method for 4H-SiC, a polymorph of silicon carbide. The team’s approach is based on micro-CT scanning (CT = computed tomography). This CT technology can allow the microscopic characterization of SiC crystals without any damage to study the growth process and find defects. To process the resulting images, the team used artificial intelligence (AI) in the form of a deep learning algorithm called convolutional neural network (CNN).
The team grew a 4H-SiC crystal with a size of ca. 15 cm using physical vapor transport and analyzed it using micro-CT, using a tungsten-target reflection for a high yield of X-rays. The sample was moved and rotated within the X-rays to obtain the scanning data. Some of the resulting images were manually classified to identify defects such as carbon inclusions, “micropipes” (a penetrating type of defect), or other polytypes than the desired one. The data, together with pre-existing images, was then used to train the AI model.
The resulting model was able to detect and classify defects with high accuracy and precision. A three-dimensional visual reconstruction then provided an overview of the SiC aggregation process. According to the researchers, this work could be a useful tool to understand and optimize SiC growth technology, and the methodology could be expanded to other semiconductor single crystals.
- Non‐destructive and deep learning‐enhanced characterization of 4H‐SiC material,
Xiaofang Ye, Aizhong Zhang, Jiaxin Huang, Wenyu Kang, Wei Jiang, Xu Li, Jun Yin, Junyong Kang,
Aggregate 2024.
https://doi.org/10.1002/agt2.524
Update (March 14, 2024):
The article previously contained a typographical error in “4H-SiC”. This has been corrected.
