Carrier dynamics through V-defects in long-wavelength InGaN LEDs and efficient defect region density recognition based on multilayer perceptron

Micro-light-emitting diodes (micro-LEDs) have profoundly advanced the next-generation display technology, with their miniaturization and reliability advantages. InGaN red micro-LEDs are regarded as promising candidates for ultrahigh-resolution full-color microdisplays with dimensions of less than 5 μm, featuring superior efficiency and stability. In this work, carrier dynamics in V-defects of InGaN red LEDs grown on the silicon (111) substrate have been directly demonstrated by means of microscopic cathodoluminescence (CL) imaging. By screening dislocations and modulating carrier distribution through the sidewall of V-defects, recombination efficiency can be locally enhanced. However, reliable methodologies for accurately identifying the spatial distribution of V-defect regions in long-wavelength InGaN LEDs remain underexplored, which is critical for local carrier injection modulation. The framework trained with a multilayer perceptron (MLP) was adopted to correlate CL features with pixel-scale prediction of the defect region. This machine learning strategy effectively complements the limitations inherent in conventional analyses and enables efficient evaluation of the defect region, with an average prediction deviation below 15%.