Spectrum reconstruction with filter-free microspectrometers has attracted much attention owing to their promising potential in in situ analysis systems, on-chip spectroscopy characterizations, hyperspectral imaging, etc. Further efforts in this field can be devoted to improving the performance of microspectrometers by employing high-performance photosensitive materials and optimizing the reconstruction algorithms. In this work, we demonstrate spectrum reconstruction with a set of photodetectors based on graded-band-gap perovskite quantum dot (PQD) heterojunctions using both calculation and machine learning algorithms. The photodetectors exhibit good photosensitivities with nonlinear current-voltage curves, and the devices with different PQD band gaps show various spectral responsivities with different cutoff wavelength edges covering the entire visible range. Reconstruction performances of monochromatic spectra with the set of PQD photodetectors using two different algorithms are compared, and the machine learning method achieves relatively better accuracy. Moreover, the nonlinear current-voltage variation of the photodetectors can provide increased data diversity without redundancy, thus further improving the accuracy of the reconstructed spectra for the machine learning algorithm. A spectral resolution of 10 nm and reconstruction of multipeak spectra are also demonstrated with the filter-free photodetectors. Therefore, this study provides PQD photodetectors with the corresponding optimized algorithms for emerging flexible microspectrometer systems.
Keywords: graded band gap; machine learning algorithm; perovskite quantum dots; photodetectors; spectrum reconstruction.