Machine learning enabled optimization of showerhead design for semiconductor deposition process

In semiconductor fabrication, the deposition process generates layers of materials to realize insulating and conducting functionality. The uniformity of the deposited thin film layers’ thickness is crucial to create high-performance semiconductor devices. Tuning fabrication process parameters (e.g., for evenly distributed gas flow on the semiconductor wafer) is one of the dominant factors that affect film uniformity, as evidenced by both experimental and numerical studies. Conventional trial and error methods employed to change and test a range of fabrication conditions are time-consuming, and few studies have explored the effect of changing the geometry of hardware components, such as the showerhead. Here, we present a design optimization of the showerhead for flow uniformity based on numerical simulation data using machine learning surrogate models. Accurate machine learning models and optimization algorithms are developed and implemented to achieve 10% more flow uniformity compared to a benchmark traditional showerhead design. Moreover, the developed Bayesian optimization method saves 10-fold computational cost in reaching the optimal showerhead designs compared to conventional approaches. This machine learning enabled optimization platform shows promising results which could be implemented for other optimization problems in various manufacturing systems such as semiconductor fabrication and additive manufacturing.