We propose a complete Artificial Intelligence (AI)/Deep Learning (DL)-based image analysis framework for hardware assurance of digital integrated circuits (ICs). Our aim is to examine and verify various hardware information by analyzing the Scanning Electron Microscope (SEM) images of an IC. In our proposed framework, we make use of DL-based methods at all essential steps of the analysis. To the best of our knowledge, this is the first such framework that makes heavy use of DL-based methods at all essential analysis steps. For image analysis tasks such as stitching misalignment detection and stacking movement regression that were previously performed mainly manually, we propose novel DL-based method and novel DL model architecture to automate these tasks. One of the salient features of our proposed framework is the heavy use of automated and semi-automated methods in preparing training data and the use of synthetic data to train a DL model. We also propose to train a preliminary DL model for training data preparation in scenarios where the noise level of the image set is high. Further, to maximally encourage model re-use, we propose various DL models that can operate on feature images thus applicable to new image sets without model re-training. By applying our proposed framework to analyzing a set of SEM images of a large digital IC, we prove its efficacy. Our DL-based methods are fast, accurate, robust against noise, and can automate tasks that were previously performed mainly manually. Overall, we show that, by applying our proposed various DL-based methods, we can largely increase the level of automation in hardware assurance of digital ICs and improve its accuracy.

我们提出了一个完整的基于人工智能 (AI)/深度学习 (DL) 的图像分析框架，用于数字集成电路 (IC) 的硬件保证。我们的目标是通过分析 IC 的扫描电子显微镜 (SEM) 图像来检查和验证各种硬件信息。在我们提出的框架中，我们在分析的所有基本步骤中都使用了基于 DL 的方法。据我们所知，这是第一个在所有基本分析步骤中大量使用基于深度学习的方法的此类框架。对于以前主要手动执行的拼接错位检测和堆叠运动回归等图像分析任务，我们提出了新颖的基于深度学习的方法和新颖的深度学习模型架构来自动化这些任务。我们提出的框架的显着特征之一是在准备训练数据时大量使用自动化和半自动化方法，以及使用合成数据来训练深度学习模型。我们还建议训练一个初步的 DL 模型，用于在图像集的噪声水平较高的情况下训练数据准备。此外，为了最大限度地鼓励模型重用，我们提出了各种可以对特征图像进行操作的 DL 模型，从而适用于无需模型重新训练的新图像集。通过应用我们提出的框架来分析大型数字 IC 的一组 SEM 图像，我们证明了其有效性。我们基于 DL 的方法快速、准确、抗噪性强，并且可以自动执行以前主要手动执行的任务。总的来说，我们表明，通过应用我们提出的各种基于 DL 的方法，