A popular technique for ending a plasma etch process is to monitor the optical emissions from reaction species for gross changes. This at first appears to be a simple real-time edge detection task. However, differences in plasma chemistry across runs, reactor chambers, and wafer patterns combine to make this control strategy quite problematic. Moreover, as exposed wafer areas continue to shrink, distinct end points become increasingly harder to identify. This new approach is simple, robust, and flexible. The application user trains elementary pattern detectors, or neurons, on a few representative process end point shapes. Software algorithms process the training end point data to find a range of resolutions for pattern recognition. The application then automatically builds a pattern detection network that tolerates signal noise, adapts to changing end point shape, and precisely registers process end points. The neural network end point detector has been successfully tested on a wide variety of data gathered over a period of many months. The application is currently calling end point in real time on several oxide etch processes at a major wafer fabrication facility.