Development of Artificial Intelligence-Based Forcefields To Model Tunnel Barriers In Superconducting Qubits

In this study, we developed inter-atomic potential models to predict surface reactions between atomic configurations of water and TMA (Tri-Methyl-Aluminum) within the context of Atomic Layer Deposition (ALD) using Artificial Intelligence (AI). The main goal of the ALD process is to generate a thin alumina layered less than 1 nanometer tunnel barrier as a part of the building block of the Josephson Junctions for superconducting qubits. Depositions of water that deliver Oxygen and TMA that carries Aluminum are alternated during deposition; each layer should be as atomically conforming as possible to guarantee full coverage and thus no leakage to enhance the quantum circuit's fidelity. To ensure a high-quality Alumina tunnel barrier, understanding the atomic interactions between water and TMA is crucial, and the Machine Learning Interatomic Potential (MLIP) using AI can be employed to reduce the time and cost while guaranteeing the accuracy of quantum mechanics as well. We systematically evaluated the efficacy of the MLIP by assessing the resulting surface reactions and dynamics during each half-cycle of ALD process. The computing works have been performed using NERSC supercomputing facility and Missouri State University' sAI workstation.