Automatic defect localization in VLSI circuits: A fusion approach based on the Dempster-Shafer theory

Abstract

Defect localization in Very Large Integration Circuits (VLSI) requires to use multi-sensor information such as electrical waveforms, emission microscopy images and frequency mapping in order to detect, localize and identify the failure. Each sensor provides a specific kind of feature modeling the evidence. Thus, the defect localization in VLSI can be summarized as a problem of data fusion with heterogeneous and imprecise information. This study illustrates how to reproduce the human decision for modeling and fusing the different multi-sensor features by using the Demspter-Shafer theory. We propose not only an automatic decision rule for mass functions computing but also confidence intervals to quantify the final decision and to bring a decision help for the analysts expertise. Finally, a case of study is reported to attest the expert decision reproducibility.