Microscopic and data-driven modeling and operation of thermal atomic layer etching of aluminum oxide thin films

With increasing demands for microchips and increasing needs in the nano-scale semiconductor manufacturing industry, atomic layer etching (ALE) has been developing into a critical etching process. Unlike its counterpart in the film deposition domain, atomic layer deposition (ALD), which has been extensively studied, ALE has not been fully studied yet from a modeling and operation point of view. Therefore, this work develops microscopic models to characterize the thermal ALE process of aluminum oxide thin films with two precursors (hydrogen fluoride and trimethylaluminum). First, the reaction mechanisms for the two half-cycles for the thermal ALE process are established. Electronically predicted geometries of the Al2O3 structure with two precursors are optimized. Along with the optimized geometries, possible reaction pathways are proposed and calculated by density functional theory (DFT)-based electronic structure calculations. The proposed reaction paths and their kinetic parameters are used in a kinetic Monte Carlo (kMC) algorithm, which is capable of capturing the features of the thermal ALE of aluminum oxide. The kMC simulation provides an etch time for the given steady-state operating conditions, which are validated via comparison with available experimental results. Finally, data sets collected from the kMC simulation are used to train a feed-forward artificial neural network (FNN) model. The trained FNN model accurately predicts an etch time and dramatically reduces the computation time compared to the kMC simulation, thereby making it possible to carry out real-time, model-based operational parameter calculations. In addition, the trained FNN model can be used to establish a feasible range of operating conditions without demanding experimental work.