Benchmarking Classical and Deep Learning Approaches for Defect Detection in High-Resolution Wafer Inspection Imaging

Abstract

High-resolution wafer inspection is a critical step in semiconductor manufacturing, directly impacting yield and reliability. This paper benchmarks classical image processing techniques against recent deep learning models for defect detection in wafer images. We compare accuracy, computational performance, and robustness across various defect types using a standardized dataset. Results reveal that while deep learning offers superior accuracy, classical methods still excel in specific, low-variance detection scenarios. The study highlights the trade-offs involved in real-time inspection systems and offers guidance on model selection based on practical deployment requirements.