Statistical outlier screening for latent defects

This study analyzes parametric wafer probe test measurements from high quality SoCs for automotive market. This product is a safety critical part that must have a near zero Defective Parts per Million (DPPM) rate. In order to achieve the required quality standard, a comprehensive parametric test set is performed on each part. In very rare occasions, a part with latent defect is identified. The latency of the defect is established through failure analysis after the part is deemed failing. In this paper, we study the possibility of screening such latent defective parts during wafer sort based on its early signature shown on parametric wafer tests. In earlier works, it is shown that multivariate outlier analysis can be used for capturing the rare defective parts (or returns) for a high quality product line [1]. Using parametric wafer probe test measurements, multivariate outlier models are created and applied to preemptively predict potential returns. This paper analyzes three particular returns, starting from its failure analysis report to suggesting a statistical outlier methodology to screen this part. In this full paper, multiple returns with latent defects will be analyzed.