A Progressive Transfer Learning Method for Real-Time Detection of Wafer Test-Induced Defects

Abstract

Wafer-level testing is essential for detecting defects caused by fabrication process deviations. However, test failures may introduce misjudgment and even physical damage to defect free wafers, resulting in wafer test-induced defects (WTIDs) with significant impact on yield management. Although convolutional neural networks (CNNs) have been widely adopted for defect detection, they face difficulties for WTID detection due to limited samples and real-time constraints. To address this issue, this paper proposes a transfer learning (TL)-based WTID detection method, where a pretrained CNN is progressively fine-tuned using quarter, half, and full wafer maps in sequence. A corresponding progressive early-exit procedure is further developed, enabling defective wafers to be identified at early stages. The proposed method is implemented on the ST20 automatic test equipment (ATE) platform. Experimental results show that the detection accuracy on quarter, half, and full wafer maps all exceeds 95%, while the overall detection time is reduced by 37.5% compared with the conventional detection flow based on the full wafer maps.