Optimizing Niobium-Oxide Film Deposition via Pulsed-DC Sputtering by Dual-Stage Deep Neural Network-Gaussian Process Regression Framework

This work presents a neural network-based method to predict process conditions for fabricating NbOxfilms with target oxygen stoichiometry via pulsed-DC magnetron sputtering (PDC). By correlating O2flow and target voltage with film hysteresis behavior, we develop a hybrid model incorporating Deep Neural Network-Gaussian Process Regression (DNN-GPR). This approach significantly reduces experimental cost in process optimization while maintaining strong transferability to other metal oxide systems, and exhibits robust performance under limited data conditions, demonstrating substantial potential for AI-driven optimization in integrated circuit manufacturing processes.