Principal Component Analysis Based Support Vector Machine for the End Point Detection of the Metal Etch Process

Estimation and Control in Semiconductor Etch: Practice and Possibilities

“…In [60], a PCA-based support vector machine (SVM) algorithm, which could loosely be described as a kernel PCA method [61], has been employed for endpoint detection, based on OES measurement. The use of an SVM to facilitate nonlinear PLS in a classification/fault detection setting has also been considered [62], though no applications in plasma etch have yet been reported.…”

Spectral clustering algorithm for real‐time endpoint detection of silicon nitride plasma etching

“…[15][16][17][18] Researchers have attempted to improve sensitivity using whole spectra by adopting various machinelearning techniques that utilize the signals from thousands of optical channels in a short time. [19][20][21][22][23][24][25][26][27][28][29][30][31][32] Dimension reduction methods, such as principal component analysis (PCA) [20][21][22] and non-negative matrix factorization (NMF), [23] have been developed to reduce the OES dataset dimensionality and extract patterns from the endpoint. Classification methods, such as the support vector machine (SVM), [24][25][26][27] hidden Markov model (HMM), [28,29] and convolutional neural network (CNN), [30] have been developed for EPD to classify the state of etching processes before and after the endpoint.…”

“…[19][20][21][22][23][24][25][26][27][28][29][30][31][32] Dimension reduction methods, such as principal component analysis (PCA) [20][21][22] and non-negative matrix factorization (NMF), [23] have been developed to reduce the OES dataset dimensionality and extract patterns from the endpoint. Classification methods, such as the support vector machine (SVM), [24][25][26][27] hidden Markov model (HMM), [28,29] and convolutional neural network (CNN), [30] have been developed for EPD to classify the state of etching processes before and after the endpoint. Both dimension reduction and classification methods require training for real-time EPD, and the trained models are limited to specific target materials and process conditions.…”