Inline Inspection Improvement using Machine Learning on Broadband Plasma Inspector in an Advanced Foundry Fab

Abstract

For inline defect inspection it is important to achieve a high capture rate of defects of interest (DOI) at low nuisance rate to increase production efficiency. A broadband plasma (BBP) wafer defect inspection system with Inline Defect Organizer™ (iDO) can separate DOI and nuisance defects into different bins.However, high expertise is required to set up an effective iDO™ classifier. Traditional iDO setup complexity increases as design rules shrink. A novel approach is developed by adopting machine learning algorithms and SEM-classified defect data to create a new iDO classifier (a.k.a. iDO 2.0). The results are promising, showing that iDO 2.0 classifier outperforms the iDO in sensitivity, nuisance rate, ease of use, time to results and cross- device portability.