Wafer defects have a well-known impact on device yield, and timely identification of their root cause is crucial for high fabrication yield and accelerated process development. There are two complementary systems for detecting defects: inspection and review. Optical Patterned Wafer Inspection (OPWI) systems are used to identify potential defects on the wafer. Subsequently, Scanning Electron Microscope (SEM) review systems provide the higher resolution necessary for classifying different defect types. However, OPWI systems face challenges as semiconductor designs become increasingly complex and involve the fabrication of 3D structures. Additionally, the extremely small feature sizes of leading-edge transistors make yield-killing defects undetectable by the optical resolution of OPWI systems. Therefore, advanced sampling and filtering algorithms are critical to solve the most challenging aspect of inspection, which is distinguishing yield-killing defects from noise, particularly for advanced design rules and novel architectures. In this paper, we present the results of a defect metrology study conducted by implementing the innovative ExtractAI technology on imec's unique Gate-all-around (GAA) Forksheet process. The ExtractAI utilizes Machine Learning and statistical processing algorithms to establish a connection between a state-of-the-art Deep-Ultraviolet (DUV) brightfield wafer inspection system and a leading SEM review and classification system. The ExtractAI employs an iterative and adaptive sampling process that maximizes the extraction of Defects of Interest (DOI) while maintaining a low review budget. The end result is a yield-killing oriented classified wafer defect map. The ExtractAI value is most significant when utilized during the Research and Development (R&D) of new processes. This is primarily due to the fact that R&D wafers typically exhibit extensive defect count, yet only a few of these defects are of interest, thereby making the identification of yield-killing defects more challenging. Moreover, the ExtractAI facilitates swift recipe creation and expedited time to obtain results, which are crucial during the R&D phase. For this study, a post Silicon-Germanium (SiGe) Nanosheet (NS) etch wafer was selected, and the inspection recipe was focused on ‘single-bridge’ defects between the GAA lines, and ‘single-hole’ defects in the dielectric wall which are evident in this GAA Forksheet architecture. The study results using the ExtractAI technology showcased a 100% increase in the detection of yield-killing defects. Additionally, it reduced the number of nuisance defects by 30%. Alternatively, achieving the benchmark capture rate is possible with only 60% of the review budget. An offline study conducted with the next generation of ExtractAI demonstrated an additional 20% reduction in the review budget. These findings emphasize the potential value of the ExtractAI-enabled synergy between optical inspectors and SEM review systems for advanced nodes and novel 3D architectures. This synergy offers a higher capture rate and improved separation between yield-killing and nuisance defects, resulting in reduced FAB cost of ownership, accelerated yield ramp, and faster time to market.
This project has received funding from the ECSEL Joint Undertaking (JU) under grant agreement No 101007254. The JU receives support from the European Union’s Horizon 2020 research and innovation program.