Abstract
3D integration and miniaturization techniques get more widely used in conventional integrated circuits but also represent crucial ingredients for future quantum computing devices. This consolidates the need for efficiently detecting increasingly small defects on wafer size. Here we present a time-efficient and accurate way of measuring, localizing and statistically classifying defects down to the micrometer regime, utilizing a combination of scanning acoustic microscopy (SAM), You Only Look Once object detection, semantic segmentation and a machine learning super-resolution (ML-SR) approach. In particular, we test the capabilities of different ML-SR approaches to enable self-supervised quality enhancement of the measured image data. We reveal that the developed AI-powered workflow enhances time-efficiency by a factor of around 4x and 6x for the TSV and delamination analysis, respectively. Yet, the mentioned approach is not limited to SAM image data but presents a general way for speeding-up failure analysis in various fields.