Linear regression combined KNN algorithm to identify latent defects for imbalance data of ICs

During the manufacturing test process, researchers often overlook those latent defects (induce failure) which, similar to process variation (PV, will not induce fatal failure in early life stage), can seriously affect the test results of early-life failure (ELF) and therefore must be removed. Theoretically, machine learning (ML) classification method can be used to identify these latent defects. In fact, when significant data imbalance occur, classifiers perform poorly. Therefore, this paper proposes a new type data processing method, which can extract latent defect characteristic by linear regression function, in this way, latent defects and PV can be successfully distinguished by K-Nearest Neighbor (KNN). Experimental results demonstrate that the predictive accuracy of this data processing method is 32% higher than other ML method.