Two applications of data-mining in production test are cost reduction and quality improvement. In production, the most costly test step is burn-in. Hence, eliminating or reducing the burn-in test is highly desirable. Test quality is essential for products in the automotive market that require a zero defective parts per million (DPPM) rate. In this thesis, we propose a test data-mining framework consisting of two phases. The first phase is for developing an understanding of test data and uncovering its important aspects such as systematic variations and anomalies. The knowledge learned in the first phase is applied in the second phase, where learning tools and methodologies are used to build test models that are applied in production. These models are similar to derived tests that can be used to replace expensive test steps or as complementary tests to improve the test quality. For the first phase, we develop three pattern mining methodologies for inter-wafer abnormality analysis.

Given a set of wafers, the first methodology identifies wafers with abnormal failing patterns based on a test or a group of tests. Given a wafer of interest, the second methodology searches for a test perspective that reveals the abnormality of the wafer. Given a particular pattern of interest, the third methodology implements a monitor to detect wafers containing similar patterns. We address the key elements for implementing each methodology and demonstrate the potential based on experiments performed on a high-quality SoC product line. For the second phase, we study applications for burn-in time reduction and customer return screening based on two product lines designed for the automotive market. For burn-in time reduction, the experiment focuses on developing multivariate test models based on parametric test data collected after 10 hours of burn-in to predict parts likely-to-fail after 24 and 48 hours of burn-in.

Applying these models will identify a large portion of all parts that do not require longer burn-in time, potentially providing significant cost saving. For screening potential customer returns, preemptive and reactive model building approaches are developed to identify potential customer returns during wafer probe testing. The preemptive approach selects correlated tests to construct multivariate test models to screen outliers. A reactive approach selects tests relevant to a given customer return and builds an outlier model specific to the return. This model is applied to capture future parts similar to the return. The experiment shows that each approach can capture returns not captured by the other. We demonstrate that both approaches can have a significant impact on reducing customer return rates especially during the later period of the production.