Deep pattern solution for interpreting systematic defects

AI based, Data-driven solutions are increasingly being applied to risk pattern (RP) analysis and the derivation of design rules in semiconductor design. However, practical methodologies for analyzing RPs while considering process variation remain unclear. This study proposes a neural network-based seamless framework aimed at strengthening design rules. Considering that defects originate from probabilistic processes driven by various sources of variation, we developed a neural network-based prediction methodology and a structural visualization technique that incorporate these variation factors for RP detection. Furthermore, to link the analysis results to the enhancement of design rules, we introduce a method to detect hidden design factors that are not predefined, along with an explanation methodology for identifying vulnerable samples that cause defects. The proposed framework is scalable to handle large datasets and has been applied to open defect analysis in the design layout layers of logic products. By analyzing billions of patterns, it identifies previously unknown RPs and contributes to the reinforcement of design rules.