Siliconware Precision Industries (SPIL) has achieved satisfactory results from applying AI to detect defects on packaging lines following nine months of efforts, according to KH Wan, director of technology development at the Taiwan-based semiconductor backend service provider.
At a recent Nvidia-hosted AI Strategy Summit in Taipei, Wan said that AOI (automated optical inspection) system can hardly judge the "true defect" that may affect the quality of wafers, as only one out of every 100 defects spotted by AOI proves a true one.
Wan said that in order to improve AOI efficiency and reduce manual double-check, SPIL moved in the second half of 2018 to enforce an ADC (automatic defect classification) project incorporating the use of AI, joining forces with AOI equipment suppliers and ADC system developers to evaluate and build deep learning patterns with the support of Nvidia.
Wan revealed SPIL has significantly improved defect detection performance with the assistance of AI, and hopes to totally drop manual double-check on packaging lines.
SPIL expects to work out 100 training AI patterns by the end of the third quarter 2019, and will incorporate the patterns into the firm's three backend process plants, Wan said, adding that his company also hopes to utilize AI to help classify defects beyond detection.