Selection of Test Patterns and Primary Outputs for Volume Diagnosis

In the early stage of manufacturing new semiconductor devices, many chips are found to be faulty during test. To increase profit for manufacturing industries, design and manufacture process must be improved, and this is achieved by diagnosing failed chips. This paper considers volume diagnosis, where failed chips are diagnosed by test equipment with limited memory capacity. We propose three methods for selecting test patterns and primary outputs (POs) for which output responses are collected to store in the memory of test equipment. In the methods, following metrics are considered: the number of detecting patterns for each fault, the number of faults whose effects are propagated to each PO, the number of fault pairs that are distinguished by each test pattern and at each PO. We aim to limit the degradation of fault diagnosis ability of the test set. The effectiveness of the methods is investigated through experiments for benchmark circuits.