While Cadmium Telluride (CdTe) excels in terms of photon radiation absorption properties and outperforms silicon (Si) in this respect, the crystal growth, characterization and processing into a radiation detector is much more complicated. Additionally, large concentrations of extended crystallographic defects, such as grain boundaries, twins, and tellurium (Te) inclusions, vary from crystal to crystal and can reduce the spectroscopic performance of the processed detector. A quality assessment of the material prior to the complex fabrication process is therefore crucial. To locate the Te-defects, we scan the crystals with infrared microscopy (IRM) in different layers, obtaining a 3D view of the defect distribution. This provides us with important information on the defect density and locations of Te inclusions, and thus a handle to assess the quality of the material. For the classification of defects in the large amount of IRM image data, a convolutional neural network is employed. From the post-processed and analysed IRM data, 3D defect maps of the CdTe crystals are created, which make different patterns of defect agglomerations inside the crystals visible. In total, more than 100 crystals were scanned with the current IRM setup. In this paper, we compare two crystal batches, each consisting of 12 samples. We find significant differences in the defect distributions of the crystals.

虽然碲化镉 (CdTe) 在光子辐射吸收特性方面优于硅 (Si)，但晶体生长、表征和加工成辐射探测器要复杂得多。此外，大浓度的扩展晶体缺陷，如晶界、孪晶和碲 (Te) 夹杂物，因晶体而异，会降低处理后的探测器的光谱性能。因此，在复杂的制造过程之前对材料进行质量评估至关重要。为了定位 Te 缺陷，我们用红外显微镜 (IRM) 扫描不同层的晶体，获得缺陷分布的 3D 视图。这为我们提供了关于 Te 夹杂物的缺陷密度和位置的重要信息，从而成为评估材料质量的句柄。对于大量IRM图像数据中的缺陷分类，采用了卷积神经网络。根据经过后处理和分析的 IRM 数据，创建了 CdTe 晶体的 3D 缺陷图，这使得晶体内不同的缺陷聚集模式可见。使用当前的 IRM 设置总共扫描了 100 多个晶体。在本文中，我们比较了两个晶体批次，每个批次由 12 个样品组成。我们发现晶体的缺陷分布存在显着差异。这使得晶体内部的缺陷聚集的不同模式可见。使用当前的 IRM 设置总共扫描了 100 多个晶体。在本文中，我们比较了两个晶体批次，每个批次由 12 个样品组成。我们发现晶体的缺陷分布存在显着差异。这使得晶体内部的缺陷聚集的不同模式可见。使用当前的 IRM 设置总共扫描了 100 多个晶体。在本文中，我们比较了两个晶体批次，每个批次由 12 个样品组成。我们发现晶体的缺陷分布存在显着差异。