基于数字孪生辅助电致发光分析的无损微腔显示器结构识别

In microcavity displays, different device structures can produce nearly identical normal-incidence electroluminescence spectra, making non-destructive identification difficult. This is especially relevant when narrow-band QLED emission is compared with cavity-narrowed OLED emission. Angle-resolved measurements can distinguish these cases, but the need for mechanical rotation limits measurement throughput. Here, we developed a digital twin-assisted method that uses a single normal-incidence spectrum for structural identification. The optical model was parameterized with measured material properties and checked against measured electroluminescence spectra. It was then used to generate spectra with ±1 nm electrode-thickness variations, and measured spectra were also included during training. Four machine-learning classifiers were compared for eight QLED/OLED device structures. The Tanh-activated multilayer perceptron gave the highest testing accuracy of 93.94%, compared with 84.85% for logistic regression. These results show that small differences in the full spectral shape can support structural identification when peak wavelength and linewidth alone are ambiguous. The method provides a practical basis for rotation-free optical screening of microcavity display structures.