Overcoming Data Scarcity in Semiconductor Anomaly Detection via Zero-Shot and Few-Shot Models.

Accurate quality control is critical in semiconductor manufacturing, but training traditional anomaly detection models requires massive and manually labeled data. This thesis addresses data scarcity by evaluating zero-shot and few-shot learning frameworks for semiconductor anomaly detection. Utilizing specialized datasets, which captures micrometer-scale flaws in semiconductors, this research benchmarks how effectively cutting-edge vision models identify and localize anomalies with minimal to zero prior training exposure.