Research data supporting "Self-supervised deep learning for tracking degradation of perovskite light-emitting diodes with multispectral imaging"

Contents of the dataset: Fig 3a-e.zip: Zip file containing a raw hyperspectral photoluminescence (PL) 3D cube of a self-assembled CsPbBr3 perovskite nanoplatelet film in Hierarchical Data Format (HDF5). Image pixel size is 66 nm. The wavelength range is 420-550 nm with a step size of 2 nm. The sample was unencapsulated and measured in air under a 405 nm continuous-wave (CW) laser with an excitation intensity of 100 mW/cm2. SourceData_Fig1.xlsx: Non-blind denoising results on a 3D airborne hyperspectral remote sensing image of the Washington DC Mall (public dataset link: https://engineering.purdue.edu/~biehl/MultiSpec/hyperspectral.html). SourceData_Fig2.xlsx: Blind denoising results on a 3D hyperspectral microscopy image of organic mCBP-doped 4CzIPN films. SourceData_Fig3.xlsx: 1) Statistics on signal-to-noise ratio and PL peak prediction for PL mapping of self-assembled CsPbBr3 perovskite nanoplatelet film from 440 to 540 nm (dataset in Fig 3a-e.zip). 2) Raw PL spectrum over time of a thermally evaporated wide-gap FA0.7Cs0.3Pb(I0.6Br0.4)3 perovskite film for 20 min of laser exposure. The data was collected from a sub-micron region (330 x 330 nm) of an unencapsulated film in ambient condition. SourceData_Fig4.xlsx: 1) Raw electroluminescence (EL) spectrum over time on mixed Br/Cl blue-emitting perovskite LEDs at a bias voltage of 6 V. 2) Statistics of local-EL peak over time on 14,044 local points (330 x 330 nm spatially resolved). SourceData_ExData_Fig2.xlsx: Raw activation data from the first 32 channels of the first batch-normalization layer output of PA-Net and PA-CNN using noisy image inputs of various noise levels. SourceData_ExData_Fig4.xlsx: Statistics of 119,931 local-PL peak predictions of CsPbBr3 perovskite nanoplatelet film (dataset in Fig 3a-e.zip). SourceData_ExData_Fig5.xlsx: 1) Raw local-PL data of Cs0.05FA0.78MA0.17Pb(I0.83Br0.17)3 perovskite film on SnO2/ITO/glass substrates. 2) Statistics on signal-to-noise ratio against wavelength from 700 to 900 nm. SourceData_ExData_Fig6.xlsx: EL peak prediction results across 500-pixel line scan (330 nm pixel size) of mixed Br/Cl blue-emitting perovskite LEDs at 0 min and 10 min of device operation. Abstract of paper that the dataset supports: Emerging functional materials such as halide perovskites are intrinsically unstable, causing long-term instability in optoelectronic devices made from these materials. This leads to difficulty in capturing useful information on device degradation through time-consuming optical characterisation in their operating environments. Despite these challenges, understanding the degradation mechanism is crucial for advancing the technology towards commercialisation. Here we present a self-supervised machine learning model that utilises a multi-channel correlation and blind denoising to recover images without high-quality references, enabling fast and low-dose measurements. We perform operando luminescence mapping of various emerging optoelectronic semiconductors, including organic and halide perovskite photovoltaic and light-emitting devices. By tracking the spatially resolved degradation in electroluminescence of mixed-halide perovskite blue light-emitting diodes, we discovered that lateral ion migration (perpendicular to the external electric field) during device operation triggers the formation of chloride-rich defective regions that emit poorly – a mechanism which would not be resolvable with conventional imaging approaches.

数据集内容如下：
1. Fig 3a-e.zip：为层级数据格式（Hierarchical Data Format, HDF5）存储的自组装CsPbBr₃钙钛矿纳米片薄膜的原始高光谱光致发光（Photoluminescence, PL）三维数据立方体压缩包。图像像素尺寸为66 nm，波长范围为420~550 nm，步长为2 nm。样品未封装，在空气中采用405 nm连续波（Continuous-wave, CW）激光器，以100 mW/cm²的激发强度进行测量。
2. SourceData_Fig1.xlsx：针对华盛顿特区购物中心三维机载高光谱遥感影像的非盲去噪结果（公开数据集链接：https://engineering.purdue.edu/~biehl/MultiSpec/hyperspectral.html）。
3. SourceData_Fig2.xlsx：针对有机mCBP掺杂4CzIPN薄膜的三维高光谱显微图像的盲去噪结果。
4. SourceData_Fig3.xlsx：
(1) 针对440~540 nm范围内自组装CsPbBr₃钙钛矿纳米片薄膜的PL成像的信噪比与PL峰预测统计（数据集位于Fig 3a-e.zip中）；
(2) 热蒸发制备的宽间隙FA₀.₇Cs₀.₃Pb(I₀.₆Br₀.₄)₃钙钛矿薄膜在20分钟激光辐照下的原始PL光谱随时间变化数据。该数据采集自环境条件下未封装薄膜的亚微米区域（330×330 nm）。
5. SourceData_Fig4.xlsx：
(1) 偏置电压6 V下混合Br/Cl蓝色发射钙钛矿发光二极管的原始电致发光（Electroluminescence, EL）光谱随时间变化数据；
(2) 14044个局部点位（空间分辨330×330 nm）的局部EL峰值随时间变化统计数据。
6. SourceData_ExData_Fig2.xlsx：使用不同噪声水平的带噪图像作为输入时，PA-Net与PA-CNN的首个批归一化层输出的前32通道原始激活数据。
7. SourceData_ExData_Fig4.xlsx：CsPbBr₃钙钛矿纳米片薄膜的119931个局部PL峰预测统计（数据集位于Fig 3a-e.zip中）。
8. SourceData_ExData_Fig5.xlsx：
(1) SnO₂/ITO/玻璃衬底上Cs₀.₀₅FA₀.₇₈MA₀.₁₇Pb(I₀.₈₃Br₀.₁₇)₃钙钛矿薄膜的原始局部PL数据；
(2) 700~900 nm范围内的信噪比随波长变化统计数据。
9. SourceData_ExData_Fig6.xlsx：混合Br/Cl蓝色发射钙钛矿发光二极管在器件运行0 min和10 min时的500像素线扫描（像素尺寸330 nm）的EL峰值预测结果。

本数据集支持的论文摘要：卤化物钙钛矿等新兴功能材料本征稳定性较差，导致基于此类材料制备的光电器件长期稳定性不足，进而难以通过耗时的光学表征在器件工作环境中获取其退化过程的有效信息。尽管面临上述挑战，阐明退化机制对于推动该技术走向商业化仍至关重要。本文提出一种自监督机器学习模型，该模型利用多通道相关性与盲去噪技术，无需高质量参考样本即可实现图像恢复，从而支持快速低剂量测量。我们针对多种新兴光电子半导体——包括有机与卤化物钙钛矿光伏及发光器件——开展了运行状态下的原位发光成像表征。通过追踪混合卤化物钙钛矿蓝色发光二极管电致发光的空间分辨退化过程，我们发现器件运行过程中的横向离子迁移（垂直于外电场方向）会触发富氯缺陷区域的形成，此类区域发光效率极低——这一机制是传统成像方法无法分辨的。

搜集汇总

背景与挑战

背景概述

该数据集包含钙钛矿发光二极管降解过程的多光谱成像数据，支持自监督深度学习模型的研究。数据集提供了原始光谱数据、统计分析结果及相关的Python代码链接，用于追踪和分析材料降解机制。