(389b) Neural Network-Based Modeling and Operation for ALD of SiO2 Thin Films Using Data from a Multiscale CFD Simulation

Atomic layer deposition (ALD) plays an important role in the manufacturing of highly conformal thin-film materials, which are used extensively in the semiconductor industry. In our previous study, a multiscale computational fluid dynamic (CFD) model has been developed for the SiO2 ALD process in an optimized showerhead ALD reactor geometry. The optimized reactor design involves a volume adjusting horn in the upstream region and a radially adaptive sizing of the showerhead holes [1], [2] . The results from the developed macroscopic CFD model show that those two design features enable quick achievement of steady-state in the gas-phase domain of the reactor, and provide a relatively uniform surface pressure and temperature profile for the microscopic kinetic Monte-Carlo (kMC) modeling of the surface deposition, which authentically simulates the film deposition process [3].

However, the multiscale CFD model is computationally demanding and is not suitable for applying real-time control schemes like model-based controllers (MPC). Motivated by this, in this work, we investigate the transient operation of the full reactor in details, using the model developed previously. The impact of operation parameters, including inlet process gas flowrate, temperature and pressure, on the deposition process is discussed and a corresponding database with temporal information of the process is constructed. Also, the appropriate operating boundary accounting for both microscopic and macroscopic operations is determined to ensure process feasibility [3]. Subsequently, a recurrent neural network (RNN) is constructed and trained using the model-generated database, which enables the prediction on the transient process dynamics based on the inputs of operating condition [4]. Moreover, utilizing the RNN model, a control scheme is designed based on the developed RNN to reduce the impact of potential disturbances on the deposition process and enhance the film uniformity. Finally, an operation example is shown to demonstrate the practicability of the proposed method.

[1] Lee, C.S., Oh, M.S., Park, H.S. US Patent No. US7138336B2. 2003. Retrieved from

https://patentimages.storage.googleapis.com/01/e8/3e/5c751ffe023e4d/US7138336.pdf

[2] Crose, M., Zhang, W., Tran, A., Christofides, P.D. Multiscale three-dimensional CFD modeling for PECVD of amorphous silicon thin films. Computers & Chemical Engineering. 2018, 113, 184 – 195.

[3] Ding, Y., Zhang, Y., Kim, K., Tran, A., Wu, Z., Christofides, P.D. Microscopic Modeling and Optimal Operation of Thermal Atomic Layer Deposition. Chemical Engineering Research and Design. 145, 159-172, 2019.

[4] Connor, J.T., Martin, R.D., Atlas, L.E. Recurrent neural networks and robust time series prediction. IEEE Transactions on Neural Networks. 1994, 5, 240-254.