IMPROVEMENT ON THE PROCESS CONTROL AND EFFICIENCY OF ALD THROUGH MACHINE LEARNING

Abstract
This research aims to improve the efficiency of ALD process by developing a Machine Learning program to assist engineers in their work. From an equipment perspective, better parameter calibration can extend the maintenance cycle and reduce manufacturing costs. From the perspective of process parameter setting, automation can save time spent on parameter calibration. The challenge with semiconductor processing equipment is that the environment inside the chamber gradually changes during continuous use. Deposits affect the supply of raw materials, and fluctuations in the working efficiency of mechanical equipment affect the vacuum environment and gas flow field. All of the above make changes to the process window. The current practice is manual calibration and maintenance, which consumes a lot of time and manpower. Parameter optimization of semiconductor manufacturing processes can be a serious problem due to the complexity of the process itself and the dependencies between each tuning parameter. The difficulty of parameter optimization lies in finding parameter combinations to compensate for the trade-off relationship between certain objectives. Machine Learning (ML) can be a powerful tool for parameter calibration. Artificial neural networks have long been explored as a way to model complex nonlinear relationships between different variables. By modeling the relationship between parameters, it is possible to understand the appropriate parameter combination under different chamber environments. In addition to saving the time spent on calibrating parameters, the improvement of parameter calibrating technology has the potential to improve the efficiency of raw material usage and prolong the cycle of equipment maintenance.