Prediction of plasma etch process by using actinometry-based optical emission spectroscopy data and neural network

Optical emission spectroscopy (OES) data were used to construct neural network models of plasma etch process. According to a statistical experiment, actinomeric OES data were collected from the etching of oxide thin films in a CHF3–CF4 magnetically enhanced reactive ion etching system. The etch responses modeled include an etch rate, a profile angle, and an etch rate-nonuniformity. Principal component analysis was applied to reduce the dimensionality of OES data. Three data variances adopted are 98, 99, and 100%. For each data variance, backpropagation neural network models were constructed. The training factors optimized by genetic algorithm include the training tolerance, magnitude of initial weight distribution, number of hidden neurons, and two gradients of activation functions in the hidden and output layers. The presented models demonstrated much improved predictions over the previous ones. The improvements were 43, 61, and 17% for the etch rate, profile angle, and etch rate-nonuniformity models, respectively.