TY  - JOUR
T1  - Multiple Defect pattern Recognition in a Wafer Map Using Vector-Representation Based Capsule Network
AU  - Kim, Misun 
AU  - Choi, Ji Hwan 
AU  - Lee, Harim 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2023
DA  - 2023/1/14
DO  - 10.7840/kics.2023.48.9.1179
KW  - Vector representation
KW  - CapsNet
KW  - Learning
KW  - Machine Learning
KW  - Deep Learning
AB  - To satisfy the demand of semiconductor market, the semiconductor manufacturing process should guarantee the production of high-yield and high-quality semiconductors. The fabrication process is complexly compound of several sub-processes, and thus even if an experienced engineer manages the process with precise equipment in a clean environment, it is difficult to make a wafer with no error-free dies. Therefore, the engineer should quickly discover which sub-process is mal-functioning for high yield. Fortunately, error dies make a specific defect pattern, which corresponds to specific abnormal operation of some fabrication sub-processes. Hence, a scheme that recognizes defect patterns in a wafer map can allow fabrication engineers to make the high-quality wafer with few error dies. In this paper, we implement a capsule network based multiple-defect recognition scheme with high precision and recall per each defect pattern. This work is the first to exploit a vector-representation based network for the recognition of defects in a wafer map, and verifies that the network using vector representation shows higher performance compared to the convontional feature-map based networks.