Case studies of defect localization based on software-based fault diagnosis in comparison with PEMS/OBIRCH analysis

• We verified the accuracy of failure localization of the fault diagnosis.
• In same case, the fault diagnosis could not localize the failure accurately.
• It is important to focus on not only score but also mismatch rate of pass patterns.
• The score tends to drop when the short occurs between the input/output nets.
• It is necessary to combine with hardware techniques such as PEMS and OBIRCH analysis.

This study verifies the accuracy of failure localization by a software-based fault diagnosis technique through comparison of the failure localization by photo emission microscope (PEMS) analysis and optical beam induced resistance change (OBIRCH) analysis. To evaluate, the software-based fault diagnosis technique was applied to 14 samples of 0.18 μm-process-node products that failed mainly due to the metal line shorts. We found that this technique was able to accurately localize the failure with a high probability (85.7%). One reason, the diagnosis returned inaccurate results is the influence of the metal line short expanding to several nets in the device simultaneously due to shorts of upper and lower metal lines. The other reason is that the fail log of the failed device itself was inaccurate due to (1) the resistance value at the short point, (2) the driving force of the cells related by the shorted points, and (3) the transition timing. We determined from our study that the rank of score calculation depends more on the mismatch rate of pass patterns than on the match rate of fail patterns. When the mismatch rate of pass patterns is less than 0.04 and the score is more than 70, the software-based fault diagnosis result is reliable. Although software-based fault diagnosis is a powerful tool for failure localization, it is necessary to combine it with hardware techniques such as PEMS analysis and OBIRCH analysis to maintain the accuracy of failure localization.