Non-destructive Dislocation Defect Inspection Technology Assists The Study of The Growth Mechanism of Dislocation Defects in SiC Crystal

Defects in SiC wafers (including surface morphology, stacking faults, and dislocation defects) are key factors affecting the yield and long-term reliability of SiC power chips. For conductive 4H-SiC wafers, reducing defect density by properly controlling ingot growth and fabrication processes from ingot growth, substrate processing to epitaxial growth, has been a long-standing goal for SiC manufacturers. Over the past decade, despite significant improvements in SiC ingot growth and fabrication techniques, the defect density in SiC wafers remains 3-4 orders of magnitude higher than that in traditional silicon-based wafers, owing to the complexity of their crystal structure and growth methods. The primary limiting factor stems from insufficient scientific understanding of SiC ingot growth mechanisms, and the influencing factors. Therefore, observing the formation and evolution of defects in SiC ingots and analyzing their generation mechanisms are particularly crucial for producing high-quality wafers and improving chip yield.

Research on SiC wafer defects relies on diverse inspection equipment. Current non-destructive inspection techniques, including optical bright-dark field surface scanning, fluorescence imaging, and Raman spectroscopy, can effectively identify defects such as pits, protrusions, scratches, and stacking faults in SiC substrates and epitaxial wafers. However, for dislocation defects in substrate wafers (Threading Screw Dislocations (TSD), Threading Edge Dislocations (TED), and Basal Plane Dislocations (BPD)), the conventional method for determination has long relied on KOH destructive etching. Research on their growth processes and transformation mechanisms has been challenging due to limitations in such destructive inspection techniques. Although X‑ray topography (XRT) can non‑destructively image dislocation defects in SiC wafers, its widespread adoption in research and industry is limited by its high equipment cost, operational complexity, and long measurement time. Consequently, critical questions about how TSD, TED, and BPD defects persist and transform during SiC crystal growth, how they migrate from the substrate to the epitaxial layer, and their underlying mechanisms remain unclear in both academic and industrial circles. These unresolved issues significantly constrain improvements in SiC ingot growth processes and wafer quality enhancement.

In early 2024, Time-Tech Spectra (TTS) developed a fast and non-destructive inspection tool (DISPEC 9000/8000) for detecting dislocation defects in conductive SiC substrate wafers. By using transient absorption spectroscopy coupled with AI recognition algorithm, the tool achieves rapid, precise, and non-contact optical inspection of defects such as TSD, TED, and BPD, and holds a great promise to replace the traditional KOH etching methods. Collaborating with Prof. Wang at ZJU-Hangzhou Global Scientific and Technological Innovation Center, TTS has successfully monitored the emergence, elimination, and transformation of various defects--including polytype coexistence defects, point defects, dislocations, and stacking faults--in conductive SiC ingot samples. This advancement offers essential experimental insights into the growth mechanisms and process optimization of SiC ingots, as evidenced by recent research on the chemical vapor reaction (CVR) method and synthesis process optimization. This article presents selected experimental results from our study.

 Figure 1. Dislocation defect inspection imaging of N-type 4H-SiC ingot longitudinal sections. The SiC ingot was cut along its growth direction, and the cut sections underwent precision surface grinding. Non-destructive optical imaging of dislocation defects was performed using the DISPEC9000 system. In the image, dislocation defects appear as "dark" signals relative to the background. The extensions of TD defects (TSD and TED) run roughly parallel to the growth direction, while the extensions of base plane dislocations (BPD, stacking faults, etc.) are oriented roughly perpendicular to the growth direction. The identification of TSD and TED was achieved through TTS’s proprietary AI image categorization model.

To observe the growth processes of TSD, TED defects, and BPD or stacking faults in SiC ingots, we fabricated N-type 4H-SiC ingots and subsequently cut longitudinal sections approximately 500μm thick along the growth direction. After precision surface grinding, the samples were imaged using the DISPEC9000 by TTS. Figure 1 illustrates localized imaging of dislocation defects within the longitudinal section, as demonstrated in the context of semiconductor materials. Due to the rapid non-radiative recombination of photogenerated carriers at defect sites the transient absorption signal intensity of excited-state carriers in these regions significantly differs from that in non-defective areas. This contrast manifests as distinct black signals in transient absorption imaging, highlighting the influence of defects on carrier dynamics. In the longitudinal section, the growth direction of TSD and TED defects aligns with the ingot's growth direction, manifesting as vertical linear signals in the imaging. The growth direction of BPD defects is nearly perpendicular to that of the ingot, forming a ~4 degree angle with the ingot surface and appearing as horizontal linear signals in the imaging. We achieved classification and recognition of TSD and TED defects in the longitudinal section through a big data AI recognition model.

In previous understanding, TD-type defects (TSD and TED) in SiC ingots were typically defined as "threading" defects. Through analysis of dislocation defect transient optical imaging results from longitudinal sections, we identified two primary growth patterns for TD defects. The first type is the long-range threading defect (Figure 2a), characterized by continuous growth over several millimeters, exhibiting enhanced signal intensity. These defects are accompanied by bending and directional changes during their propagation, which is indicative of their extended nature and the potential impact on the material's properties. The AI model preliminarily classifies these defects as TSD-type. The second characteristic TD defect pattern, termed the growth-annihilation-growth alternating type (Figure 2b), displays dash-line-like signals along specific directions in the image. Statistical analysis reveals that many of this type of defects grow for 100-500μm before undergoing an annihilation, which likely occurs randomly during the growth. The AI model tentatively categorizes these defects as TED-type.

Figure 2. Imaging of the growth process of TD (TSD and TED) dislocation defects. (a) Long-range threading TD defects, mostly TSD. (b) Growth-annihilation alternating TD defects, mostly TED.

Base plane defects (including stacking faults and BPD) appear as horizontally oriented signals in transient spectral imaging. Since these defects exhibit multiple growth directions, their signal length in the imaging does not reflect the actual growth distance (i.e., the defect's projection on the image). Through extensive data analysis, we have discovered that TSD and stacking faults, as well as TED and BPD, exhibit coexistence and transformation phenomena, as illustrated in Figure 3. This coexistence process manifests as the formation of multiple horizontally oriented base plane defects during the growth along the TD, which appears as fishbone-shaped signal features in the imaging. Additionally, we observed that the annihilation of numerous TD defects can be accompanied by the formation of a single base plane defect, indicating that TD defects can transform into base plane defects. These findings indicate that TD formation in SiC ingots may result in base plane defect formation.

Figure 3. The interaction and transformation process of TD defects and base plane defects.

The formation of dislocation defects in SiC ingots produced via vapor-phase synthesis is influenced by multiple factors during ingot growth, including mass transfer, energy transfer, temperature, and raw material distribution. These factors contribute to the creation of various types of defects such as micropipes, BPDs, TEDs, and TSDs, which can significantly impact the performance of SiC-based power devices. Although the design and technology of ingot growth furnaces have been significantly optimized, controlling and improving the growth environment at the microscopic scale remains a major challenge. Dislocation imaging of longitudinal cross-sections reveals that macroscopically, defect density in ingots exhibits certain patterns from the start to the end of growth, but microscopically, the density and distribution of defects vary significantly and randomly across different growth positions, as evidenced by studies on the distribution of shrinkage porosity and cavities in steel ingots and the microstructure of silicon and Sm₂(Co,Cu,Fe,Zr)₁₇ alloy ingots. The dislocation density and distribution on wafer surfaces produced by substrate wafer cutting at various positions within the ingot may undergo substantial changes at the microscopic level. Therefore, the current method of evaluating the dislocation defect density of the substrate by selecting the "head and tail slices" of the ingot for KOH etching cannot accurately predict and represent the defect density and distribution of each wafer produced from the whole ingot.

Figure 4. TTS DISPEC9000: Non-destructive optical SiC substrate dislocation defect inspection system. Based on the principle of semiconductor transient absorption spectroscopy, this equipment directly observes and identifies dislocations and various ingot defects through transient spectral imaging of the substrate wafers, completely replacing the traditional KOH etching method.

The DISPEC 9000, as detailed in this article, utilizes state-of-the-art non-linear optical technology to perform full-surface scans on SiC substrates, effectively and nondestructively identifies critical crystal defects. This innovative approach replaces the conventional and costly KOH etching method, significantly reducing inspection time and substrate costs, and thereby enhancing production efficiency and yield. Non-destructive spectral inspection, as a critical component in semiconductor manufacturing, facilitates 'wafer-by-wafer' inspection, offering robust support for defect management in subsequent epitaxial layer growth and chip fabrication processes. Utilizing AI image categorization algorithm, the system offers comprehensive in-situ defect data traceability, which is crucial for chip failure analysis and yield control. We believe that the widespread adoption of non-destructive dislocation defect inspection technology for SiC substrates will significantly advance research on SiC ingot growth mechanisms and improve quality control in industrial wafer manufacturing, thereby promoting the high‑quality development of third‑generation semiconductor materials and devices.