Controlling IC Quality through Diagnosis Based Adaptive Test

The main purpose of manufacturing test has been and continues to be to separate good integrated circuits (ICs) from ones that do not meet the desired operational characteristics. An ideal testing procedure should pass all good chips and fail all those that do not deliver the required functionality. However, a defective chip may pass all the tests, which leads to reduction in product quality. Product quality is commonly measured using defect level, which is the ratio of defective chips in all the chips that pass all tests. Defect level is typically expressed in units of defective parts per million shipped (DPPM). A variety of test sets derived using different test metrics are usually applied to ensure an acceptable quality level. However, various experiments involving actual ICs have shown that a significant portion of the test patterns applied do not contribute to the improvement in product quality, which indicates the possibility of significantly reducing the cost of test without affecting product quality, or alternatively improving product quality without incurring additional test cost. In this dissertation, a diagnosis-assisted adaptive test (DAT) method for improving test quality is described, developed, and validated. DAT uses diagnosis to capture the failure characteristics on a per-design basis and improves test quality by changing the test set to match the current failure characteristics. DAT uses (i) a multiple defect diagnosis method, referred to as MDX, which is flexible in handling various defect behaviors, and (ii) a failing behavior identification (FBI) method that captures defect behavior by identifying signal lines related to defect activation. MDX is evaluated using logic simulation experiments. Results from over 5,700 simulated and 131 failing ICs show that MDX can effectively diagnose circuits that are affected by a large (>20) or small number of defects of various types. Specifically, when diagnosing circuits with more than 20 defects, MDX identifies 74% of them. For circuits affected by two defects, the number changes to 87%. This level of diagnosis accuracy for ICs of this type exceeds all claims currently found in the literature. The efficacy of FBI is validated using circuitand logic-level simulation experiments. The