Applications Information - Verity Instruments, Inc. - #9

Text version of the page

In North America, call 1-972-446-9990 13 Neural PCA Multi-Wavelength Algorithm Verity’s endpoint-detection computations can employ robust algorithms such as the patent pending Neural PCA for multivariate, full-spectrum analysis. Data represented at left apply to a 0.5% exposed area contact etch using Verity’s Neural PCA algorithm. Within the SpectraView™ endpoint software application, the Neural PCA endpoint trace can be processed using Neural Network or threshold-based methods. Neural Network Algorithm The Neural Network algorithm is used to analyze endpoint traces. The Neural Network uses proprietary techniques to recognize characteristic endpoint shapes in the trend line. This is performed in real time and the pattern recognition algorithm adapts to expected amplitude and duration changes in the endpoint trace during successive runs. Unlike other types of neural networks, Verity’s algorithm can be set up with only a few training runs. If a false positive or negative is found, it can easily be added to the training set for improved robustness. Process engineers using Verity’s Neural Net software are freed from the burden associated with developing and testing threshold-based algorithms. In addition, data can be analyzed “on-the-fly” or replayed, reviewed, or reprocessed with SpectraView™. Threshold-Based Algorithm Using threshold-based algorithms, endpoint recognition is based upon the output rising above or below a preset level for a predetermined length of time. However, for demanding applications, the Neural Network algorithm is commonly selected over the threshold-based algorithm. Algorithms for Endpoint Detection Verity Instruments provides a powerful suite of endpoint algorithms, including the multivariate Neural PCA algorithm, which can be processed with Verity’s proprietary Neural Network pattern recognition software.