Defect Density Minimization in ALD Passivation Layers by Bayesian Optimization

Atomic layer deposition (ALD) is a highly efficient and effective method for providing conformal coating to sensitive materials and related surfaces, such as integrated circuits. While ALD has demonstrated its applicability across a wide range of materials and geometries, it is also crucial to achieve defect-free layers for better corrosion protection. This objective is heavily influenced by a multitude of fabrication parameters, resulting in a high-dimensional and cross-correlated parameter space that renders conventional optimization techniques impractical.<br/>To address this challenge and optimize the process parameters, we propose a probabilistic machine learning approach utilizing Bayesian Optimization (BO) with Gaussian Processes (GPs). Our approach leverages BO's data-efficient learning scheme to identify the optimal deposition parameters in just three iterations, leading to defect minimization in an ALD-Al₂O₃ passivation layer. Furthermore, our methodology enables the analysis of the impact of each process parameter on defect density, thereby providing insights into which parameters should be tuned for improved performance under varying conditions.<br/>The presented study showcases the time and cost efficiency of optimizing ALD layers using our proposed approach. Our findings highlight the effectiveness of Bayesian Optimization in material science and fabrication technologies, offering a valuable tool for optimizing diverse materials and processes.