Plasma Chamber Design Method Combined with Plasma Deep Learning Model and Optimization Algorithm

As plasma process becomes more complex, plasma simulation is becoming important but has several disadvantages such as high computation time. In addition, it is difficult to find a solution that satisfies multiple purposes at the same time. In this study, a deep learning model was implemented based on HPEM (The Hybrid Plasma Equipment Model) plasma simulation to ensure convergence and shorten the calculation time. The neural network consists of two encoders and a decoder, and each encoder distills process recipe and geometry information into a dense vector through fully-connected layers and convolution layers. MOPSO (Multi-Objective Particle Swarm Optimization) algorithm is also applied so that an optimized solution can be derived automatically as the iterations are repeated. The final algorithm can automatically split the selected variables and analyze the results to find optimized conditions for multiple objectives. In addition, the optimizing process which takes dozens of days was reduced to tens of seconds due to the deep learning model, maintaining 95% consistency of HPEM data.