USING NEURAL NETWORKS AND DEMPSTER-SHAFER THEORY FOR FAILUREDETECTION AND DIAGNOSIS OF EXCIMER LASER ABLATION

摘要

With the continuing advancement of the use ofexcimer laser systems in microsystems packaginghave come an increasing need to offset the highcapital equipment investment and lower equipmentdowntime. This paper presents a methodology for in- inlineline failure detection and diagnosis of the excimerlaser ablation process. Our methodology employsresponse data originating directly from the tool andcharacterization of microvias formed by the ablationprocess. Neural networks (NNs) are trained andvalidated based on this data to generate evidentialbelief for potential sources of deviations in theresponses. Dempster-Shafer (D-S) theory is adoptedfor evidential reasoning. Successful failure detectionis achieved: 100% failure detection out of 19 possiblefailure scenarios. Moreover, successful failurediagnosis is also achieved: only a single false alarmand a single missed alarm occurred in 19 possiblefailure scenarios.