Machine learning and edge placement error improve electron beam inspection capture rate

This paper presents a holistic solution for advanced memory manufacturing, addressing challenges in electron beam inspection (EBI) capture rate and patterning defectivity with an integrated approach. Our solution links these aspects to establish a process flow with critical monitoring and detection steps, optimizing process performance and accelerating the yield learning curve. This integration utilizes machine learning to optimize the inspection sampling, enhances the utility of EBI as an early yield indicator and improves inspection time utilization, paving the way towards autonomous defect inspection. We demonstrate a substantial improvement in defect capture rate by using computational guided inspection compared to process of record (POR) EBI sampling, achieving strong correlation with final yield (electrical test). Additionally, the introduction of a dual-layer EPE metric enables the detection of yield-impacting patterning defects several weeks earlier than POR EBI step. This comprehensive approach not only improves defect detection but also supports faster and more efficient yield optimization in the fab.