海归学者发起的公益学术平台
分享信息，整合资源
交流学术，偶尔风月
近年来出现的自主合成平台，包括多种原位或自动诊断和表征技术，促使人们开发能够利用这种多模态合成数据的机器学习模型。原位诊断和表征被应用于自主材料合成中，用以得到优化指标或者将生长动力学与实验控制相关联。此外，它们能够在合成前预测材料性能，并检测在长时间、无监督的实验中可能会出现的异常条件。脉冲激光沉积（PLD）是一种前景广阔的物理气相沉积技术，由于其能够兼容各种光学、电学、电子的诊断测量，因而被广泛应用于自主合成多种材料系统。因此，亟需开发机器学习方法，以利用PLD的诊断优势。在PLD合成过程中，人们曾使用多种原位诊断技术来监测薄膜生长或等离子体羽流。虽然机器学习应用于材料合成已有很长一段时间，但它在原位诊断数据中的应用却相当有限。
Fig.1 | Schematic diagram of the neural networks used for multi-output regressionwith intensified-CCD (ICCD) image sequences and/or growth parameters as inputs.
来自美国橡树岭国家实验室纳米材料科学中心的Sumner B. Harris团队，证明了PLD过程中产生的等离子体羽流的增强式CCD图像序列可用于异常检测和薄膜生长动力学预测。他们开发了多输出（2+1）维卷积神经网络回归模型，从羽流动力学中提取深层特征。这些特征不仅与所测量的腔室压力和入射激光能量相关，而且能够预测由原位激光反射率实验得到的自催化薄膜生长模型的参数。
Fig.2 | Training results for the anomaly detection model which predicts the chamberpressure P and laser energies E1 and E2 from intensified-CCD image sequencesusing a (2 + 1)D convolutional neural network.
他们的研究结果展示了如何利用PLD中的原位羽流诊断数据进行机器学习来保持沉积条件处于最佳状态。此外，羽流动力学对于沉积前薄膜生长动力学或薄膜其他性能的预测能力，为非专业人员快速预筛选生长条件提供了一种方案。
Fig.4 | Activations for select feature maps (FM) of the convolutional layers in the(2 + 1)D convolutional neural network model for growth kinetics highlight thedeep features learned from the plumes generated during pulsed laser deposition.
Deep learning with plasma plume image sequences for anomaly detection and prediction of growth kinetics during pulsed laser deposition
Sumner B. Harris, Christopher M. Rouleau, Kai Xiao & Rama K. Vasudevan
Materials synthesis platforms that are designed for autonomous experimentation are capable of collecting multimodal diagnostic data that can be utilized for feedback to optimize material properties. Pulsed laser deposition (PLD) is emerging as a viable autonomous synthesis tool, and so the need arises to develop machine learning (ML) techniques that are capable of extracting information from in situ diagnostics. Here, we demonstrate that intensified-CCD image sequences of the plasma plume generated during PLD can be used for anomaly detection and the prediction of thin film growth kinetics. We develop multi-output (2 + 1)D convolutional neural network regression models that extract deep features from plume dynamics that not only correlate with the measured chamber pressure and incident laser energy, but more importantly, predict parameters of an auto-catalytic film growth model derived from in situ laser reflectivity experiments. Our results demonstrate how ML with in situ plume diagnostics data in PLD can be utilized to maintain deposition conditions in an optimal regime. Further, the predictive capabilities of plume dynamics on the kinetics of film growth or other film properties prior to deposition provides a means for rapid pre-screening of growth conditions for the non-expert, which promises to accelerate materials optimization with PLD.