Over the last few years neural networks have been studied for potential applications in plasma processing. The focus of this article will be on two neural network models for complementary metal-oxide-semiconductor production. The models were developed with strict statistical cross-validation and applied to developing a plasma gate etch controller and a plasma model of a contact etch process. For a gate etch controller, the process has been evaluated in a production environment and shown to improve the process variance and throughput. For a model of a contact etch process we demonstrate that the model is limited by the inherent noise in the training data and that the direct current bias and etch time are the key control factors that determine the product quality at the end of the etch.