Automatic recognition of circuit patterns on semiconductor wafers from multiple scanning electron microscope images

A technique is proposed for high-precision, automatic recognition of circuit patterns on a semiconductor wafer from multiple scanning electron microscope (SEM) images. This technique uses multiple SEM images obtained by selective detection of secondary and backscattered electrons emitted from a wafer surface irradiated with primary electrons. It automatically detects circuit patterns in these images. The appearances of circuit patterns in SEM images vary widely depending on the structure, the material and the pattern layout. The proposed technique can cope with such a large variation in pattern appearance by adaptively selecting two recognition methods based on pattern structure and pattern density. Other information, such as the images to be processed and the contrast between pattern and non-pattern regions, is also utilized for recognition. The technique provides effective preprocessing for automating defect classification. It is expected to improve the efficacy of process monitoring and yield management in semiconductor device fabrication. Experimental results for five wafers (from which 421 circuit pattern images were obtained) demonstrate that the proposed technique can automatically recognize circuit patterns with an accuracy of 99.8%