Using neural networks and dempster-shafer theory for failure detection and diagnosis of excimer laser ablation

Abstract

With the continuing advancement of the use of excimer laser systems in microsystems packaging have come an increasing need to offset the high capital equipment investment and lower equipment downtime. This paper presents a methodology for in-line failure detection and diagnosis of the excimer laser ablation process. Our methodology employs response data originating directly from the tool and characterization of microvias formed by the ablation process. Neural networks (NNs) are trained and validated based on this data to generate evidential belief for potential sources of deviations in the responses. Dempster-Shafer (D-S) theory is adopted for evidential reasoning. Successful failure detection is achieved: 100% failure detection out of 19 possible failure scenarios. Moreover, successful failure diagnosis is also achieved: only a single false alarm and a single missed alarm occurred in 19 possible failure scenarios.With the continuing advancement of the use of excimer laser systems in microsystems packaging have come an increasing need to offset the high capital equipment investment and lower equipment downtime. This paper presents a methodology for in-line failure detection and diagnosis of the excimer laser ablation process. Our methodology employs response data originating directly from the tool and characterization of microvias formed by the ablation process. Neural networks (NNs) are trained and validated based on this data to generate evidential belief for potential sources of deviations in the responses. Dempster-Shafer (D-S) theory is adopted for evidential reasoning. Successful failure detection is achieved: 100% failure detection out of 19 possible failure scenarios. Moreover, successful failure diagnosis is also achieved: only a single false alarm and a single missed alarm occurred in 19 possible failure scenarios.