An Automatic Impedance Matching Network Based on Neural Network Technique for Inductive RF Plasma Source

A novel adaptive matching network system is proposed in this article which utilizes an efficient neural network to match the RF power generator with the inductively coupled plasma (ICP) source automatically. The proposed method utilizes a multiperceptron neural network with a Bayesian regularization optimization algorithm to identify the impedance of the plasma without measuring the phase of the load. This advantage eliminates extra components of the conventional methods for measuring the phase which improves the design complexity, noise, and cost of the matching network. A dataset from the performance mechanism of the proposed impedance matching network is extracted to analyze the system performance, and a multiperceptron neural network is developed to determine the load impedance (LI). Based on simulations and measurement results, the proposed system can automatically match the RF source to the desired load with less than −10 dB return loss. The success rate of achieving this rate of matching quality is more than 84%.