Skip to Main Content
A method of fault detection and classification (FDC) for semiconductor manufacturing equipment e-diagnostics using equipment data is presented. Detecting faulty processes, identifying any anomaly at their onsets, and rapidly classifying the root cause of the fault are crucial for maximizing equipment utilization in current semiconductor manufacturing; however, tool data acquired from production equipment contains much information that is often challenging to analyze due to its sheer volume and complexity. In this paper, modular neural network (MNN) modeling is presented as a method for fault detection modeling in plasma etching. Based on the result from the MNN modeling, a tool data set is grouped according to its related subsystems, and FDC is performed using Dempster-Shafer (D-S) theory to address the uncertainty associated with fault diagnosis. Subsystem level fault detections, such as radio frequency (RF) power source module, RF power bias module, gas delivery module, and process chamber module, are presented by combining related parameters, and successful fault detection is achieved. The evidential reasoning of RF probe is also beneficial for the detection of chamber leak simulation, and the classification of fault is made by further investigating voltage signal of RF probe. Successful fault detection in subsystem level with zero missed alarms was demonstrated using D-S theory of evidential reasoning, and the classification for finding root cause of the fault is presented in the chamber leak fault simulation. We realized that successful FDC can be accomplished by combining various related information and by incorporating engineering expert knowledge.