Universal super-resolution framework for imaging of quantum dots

TL;DR

Diffraction-limited imaging hampers nanoscale characterization of quantum emitters; the authors introduce a universal, calibration-free deep-learning framework trained on physics-based synthetic data to reconstruct super-resolved emitter maps from a single camera frame. The method generalizes across diverse PSFs, aberrations, and noise without system-specific retraining, and is validated on sparse and dense In(Ga)As quantum dots as well as strain-induced emitters in WSe2, resolving overlapping emitters under low SNR and inhomogeneous backgrounds. Key contributions include a scalable CNN trained via incremental learning on synthetic data, a loss function combining a Gaussian-filtered MSE with entropic regularization, and demonstration of sub-Rayleigh localization in real experiments. The approach enables rapid, calibration-free nanoscale metrology suitable for quantum photonic device fabrication and potentially broader solid-state emitter classes.

Abstract

We present a universal deep-learning method that reconstructs super-resolved images of quantum emitters from a single camera frame measurement. Trained on physics-based synthetic data spanning diverse point-spread functions, aberrations, and noise, the network generalizes across experimental conditions without system-specific retraining. We validate the approach on low- and high-density In(Ga)As quantum dots and strain-induced dots in 2D monolayer WSe$_2$, resolving overlapping emitters even under low signal-to-noise and inhomogeneous backgrounds. By eliminating calibration and iterative acquisitions, this single-shot strategy enables rapid, robust super-resolution for nanoscale characterization and quantum photonic device fabrication.

Figures (4)

• Figure 1: The experimental setup for imaging quantum dots with the following components. Either a collimated pulsed excitation laser (810 nm) or a continuous-wave laser (638 nm) is reflected off a 90:10 pellicle and focused onto the sample of quantum dots using an aspheric lens ($L_1$). This lens collects the photons emitted from the quantum dots. Scattered laser light from the sample is filtered out by a long-pass filter that transmits wavelengths above 850 nm, along with polarization filtering. A second lens ($L_2$) focuses the collected photons either onto a CMOS camera or an EMCCD camera, the latter providing higher sensitivity. The focal lengths of $L_1$ and $L_2$ are chosen depending on the desired magnification and the characteristics of the sample being imaged.
• Figure 2: Demonstration of deep-learning model performance on a sparse quantum dot image. a A low-resolution intensity image containing four well-separated quantum dots with a low signal-to-noise ratio. b The super-resolved reconstruction provided by the model with no prior information on the imaging system. The red rectangles indicate regions of interest that are magnified as insets in panel c. The zoomed-in view of the selected areas, with red circles denoting the positions and uncertainties estimated by a localization technique, i.e., the 2D Gaussian fitting. The predictions of our deep learning model and the fitting-based localization agree perfectly, considering that the discrepancy is 72 times smaller than the Rayleigh optical resolution limit of 0.79 $\mu$m.
• Figure 3: Super-resolved reconstructions of two high-density images of overlapping quantum dots with the 0.81 $\mu$m Rayleigh resolution limit. Both panels a and b depict the same sample region but were captured at different time points. While the overall structural features remain consistent between the frames, a subtle change occurs in the lower area of the panel b, where one quantum dot ceases to emit light. The reconstructions are virtually identical outside this region, which demonstrates the stability of the model. The areas highlighted with red squares indicate the change and are shown in detail in the zoomed-in insets I and II, where the disappearance of the emitter between frames is clearly resolved.
• Figure 4: a Wide-field image of a TMD monolayer quantum dot sample. The WSe$_2$ monolayer exfoliated from a nanoflake contains several etched, 100 nm deep square holes. The blue highlight indicates the region where the quantum dot image was taken. b Quantum dot emission image. Strongly localized excitonic states formed along the edge of an etched square hole behave as quantum dots. This low-resolution image poses a challenging case due to the combined effects of low signal-to-noise ratio and background inhomogeneity. c, d Zoomed-in view of the quantum dots and their super-resolution reconstruction. Despite the challenging conditions, the model successfully reconstructs a high-resolution image, revealing multiple quantum dots aligned along the edge. These reconstructed quantum dots are separated by approximately 0.85 µm (left to middle) and 0.59 µm (middle to right). Moreover, they form a straight line with an average deviation of 9 nm, which is significantly below the Rayleigh resolution limit of 0.65 µm.