This work details the construction and evaluation of a low computational cost hybrid multiscale thin film deposition model that couples artificial neural networks (ANNs) with a mechanistic (first-principles) multiscale model. The multiscale model combines continuum differential equations, which describe the transport of the precursor gas phase, with a stochastic partial differential equation (SPDE) that predicts the evolution of the thin film surface. In order to allow the SPDE to accurately predict the thin film growth over a range of system parameters, an ANN is developed and trained to predict the values of the SPDE coefficients. The fully-assembled hybrid multiscale model is validated through comparison against a kinetic Monte Carlo-based thin film multiscale model. The model is subsequently applied to a series of optimization and control studies to test its performance under different scenarios. These studies illustrate the computational efficiency of the proposed hybrid multiscale model for optimization and control applications.

这项工作详细介绍了将人工神经网络（ANN）与机械（第一原理）多尺度模型相结合的低计算成本混合多尺度薄膜沉积模型的构建和评估。多尺度模型将描述前体气相传输的连续微分方程与预测薄膜表面演变的随机偏微分方程（SPDE）相结合。为了使SPDE能够准确预测整个系统参数范围内的薄膜生长，开发了ANN并对其进行了训练，以预测SPDE系数的值。通过与基于动力学蒙特卡洛的薄膜多尺度模型进行比较，验证了完全组装的混合多尺度模型。该模型随后应用于一系列优化和控制研究，以测试其在不同情况下的性能。这些研究说明了所提出的混合多尺度模型在优化和控制应用中的计算效率。