摘要
Interposers play an important role in integrating multiple dies in a staked-die product. Prebond testing of interposers is an essential process for improving production yield. However, the traditional testing mechanism via probing is not appropriate, since it could destroy the fragile interposers. To this end, we propose a contactless testing methodology for prebond interposers. The testing methodology consists of two stages for detecting surface and inner defects, respectively. It detects a defective interposer according to the thermal images obtained from heating the interposer. Critical features from the thermal images are extracted and used to construct machine learning algorithms for determining whether the interposer is defective. The experimental results show that the proposed testing methodology, which detects both surface and inner defects, improves the production yield to 96.13%, compared with the testing method that reaches a production yield of 84.44% by detecting only surface defects.