Abstract Semiconductor nanocrystals with uniform morphology and composition are expected to show consistent responses during light-matter interactions. However, microscopy reveals significant variations in their photoluminescence blinking patterns, even under identical experimental conditions. This discrepancy arises from differences in crystal defects and nonradiative trap states. As a result, heterogeneous blinking patterns serve as valuable indicator of material quality, uncovering several concealed features through statistical analysis of large datasets. Nonetheless, efficient segregation and analysis of numerous blinking trajectories remain a challenge due to laborious calculations, computational bottlenecks, and manual intervention. In this study, we introduce a robust unsupervised machine learning (UML) assisted module to cluster high-dimensional blinking patterns in near-real-time, while calculating category-wise power spectral densities (PSD) to investigate active traps. Furthermore, we explore the impact of data preprocessing on clustering performance. The ‘clustering-segregation-analysis’ (UML-PSD) methodology demonstrates versatility, paving a way to advance contemporary (micro)spectroscopy, specifically for rapid and cost-effective optical characterization of semiconductor nanocrystals.
Mukherjee et al. (2026) studied this question.