Implementing Machine Learning Techniques for Optimizing Atomic Layer Deposition in Thin Films Growth
Session Number
PHYS(ai) 03
Advisor(s)
Dr. Angel Yanguas-Gil, Argonne National Laboratory
Discipline
Physical Science
Start Date
17-4-2025 2:45 PM
End Date
17-4-2025 3:00 PM
Abstract
Perfecting the development of materials synthesis conditions in nanotechnology remains a significant challenge, particularly in atomic layer deposition (ALD), a versatile process used in the development of semiconductors. ALD involves repeated dosing and purging steps with chemical material and requires incredible precision. Optimizing time intervals for dosing and purging steps is essential to avoid incomplete surface coverage or poor-quality deposition. To accomplish this, good growth must be maximized while minimizing unwanted reactions at the atomic level and overall time for the ALD process. This study utilizes machine learning techniques to accelerate the optimization of ALD processes by developing a Gaussian Process-based (GP) algorithm trained across 100 simulated reaction environments. These simulations mimic real-world experimental conditions, enabling the model to predict optimal deposition parameters efficiently and accurately. To validate the approach, the trained model is tested in an experimental ALD setup equipped with in-situ characterization techniques that provide real-time feedback, allowing the model to tune the reaction. By eliminating human intervention, this work creates a completely autonomous platform for self-optimizing materials synthesis. This framework can extend to alternative deposition techniques such as plug-flow, fluidized bed, and spatial ALD, paving the way for broader applications in thin film development.
Implementing Machine Learning Techniques for Optimizing Atomic Layer Deposition in Thin Films Growth
Perfecting the development of materials synthesis conditions in nanotechnology remains a significant challenge, particularly in atomic layer deposition (ALD), a versatile process used in the development of semiconductors. ALD involves repeated dosing and purging steps with chemical material and requires incredible precision. Optimizing time intervals for dosing and purging steps is essential to avoid incomplete surface coverage or poor-quality deposition. To accomplish this, good growth must be maximized while minimizing unwanted reactions at the atomic level and overall time for the ALD process. This study utilizes machine learning techniques to accelerate the optimization of ALD processes by developing a Gaussian Process-based (GP) algorithm trained across 100 simulated reaction environments. These simulations mimic real-world experimental conditions, enabling the model to predict optimal deposition parameters efficiently and accurately. To validate the approach, the trained model is tested in an experimental ALD setup equipped with in-situ characterization techniques that provide real-time feedback, allowing the model to tune the reaction. By eliminating human intervention, this work creates a completely autonomous platform for self-optimizing materials synthesis. This framework can extend to alternative deposition techniques such as plug-flow, fluidized bed, and spatial ALD, paving the way for broader applications in thin film development.