Nonlinear self calibrated spectrometer with single GeSe-InSe heterojunction device
Data files
Apr 24, 2024 version files 2.79 MB
-
Fig2B_S2_S3.csv
-
Fig2C.csv
-
Fig3A.csv
-
Fig3E.csv
-
Fig3F.csv
-
Fig3G.csv
-
Fig3H.csv
-
Fig4B.csv
-
Fig4C.csv
-
Fig4D.csv
-
Fig4E.csv
-
Fig4F.csv
-
Fig5C.csv
-
FigS4.csv
-
FigS5B.csv
-
FigS5C.csv
-
FigS6A.csv
-
FigS6B.csv
-
FigS6C.csv
-
FigS6D.csv
-
FigS6E.csv
-
FigS8.csv
-
Figure3B.csv
-
Figure3C_S7A.xlsx
-
Figure3D_S7F.xlsx
-
FigureS6F.csv
-
FigureS6G.csv
-
FigureS7B.xlsx
-
FigureS7C.xlsx
-
FigureS7D.xlsx
-
FigureS7E.xlsx
-
FigureS7G.xlsx
-
README.md
Abstract
Computational spectrometry is an emerging field that employs photodetection in conjunction with numerical algorithms for spectroscopic measurements. Compact single photodetectors made from layered materials are particularly attractive since they eliminate the need for bulky mechanical and optical components used in traditional spectrometers and can easily be engineered as heterostructures to optimize device performance. However, such photodetectors are typically nonlinear devices, which adds complexity to extracting optical spectra from their response. Here, we train an artificial neural network (ANN) to recover the full nonlinear spectral photoresponse of a single GeSe-InSe p-n heterojunction device. The device has a spectral range of 400-1100 nm, a small footprint of ~25×25 〖μm〗^2, and a mean reconstruction error of 〖2×10〗^(-4) for the power spectrum at 0.35 nm. Using our device, we demonstrate a solution to metamerism, an apparent matching of colors with different power spectral distributions, which is a fundamental problem in optical imaging.
README: Nonlinear Self-Calibrated Spectrometer with Single GeSe-InSe Heterojunction Device
https://doi.org/10.5061/dryad.d7wm37q7t
This datasheet comprises a total of 32 files, encompassing figures 2 to 5 in the manuscript and Figures S2 to S8 in the Supplementary Materials. The naming convention of the datasheet aligns with the respective figure numbers in both the manuscript and Supplementary Materials.
Description of the data and file structure
ANN = artificial neural network
Datasheet Fig 2B, S2, and S3 correspond to the Raman spectra.
Datasheet Fig 2C and S4 correspond to the InSe/GeSe current-voltage output curve while datasheet S4 shows an enlarged view of the device current-voltage output curve, measured over the range of ±1V.
Datasheet Fig. 3 A-H, the training set, and its nonlinear fitting. (Datasheet 3A and 3B) The spectral power density of LED sources from the nonlinear training set, (Datasheet 3C, S7A, and 3D, S7F) corresponding to the voltage-dependent photocurrent, (Datasheet 3E and 3F) with a nonlinear saturation depicted from the measured photocurrents at fixed voltages of 0.1, 0.3 and 0.5 V (black cross, blue diamond, and red dot, respectively, (Datasheet 3G and 3H) fitted to dash, dash-dot and dotted lines) that are translated to a voltage-dependent nonlinear coefficient.
Datasheet Fig. 4 B-F, Nonlinear reconstruction of power spectra. (Datasheet 4B) The reference power spectral density (pink and red lines) and corresponding reconstruction (black dotted and blue dash) of two LED light sources. (Datasheet 4C) The reference spectrum (blue line) and reconstruction (red dotted) of a color-printed transparency, (Datasheet 4D) and the same sample reconstructed at a resolution of 1,000x1 input/output vectors. (Datasheet 4E and 4F) The reconstruction error corresponding to (Datasheet 4C and 4D) (blue line), the error mean (red dashed) and the absolute error mean (green dot-dashed), respectively.
Datasheet Fig. 5C, The reflection spectra corresponding to the two filters as recorded with a reference commercial spectrometer (dashed line) and with our nonlinear spectrometer (solid line).
Datasheet Fig. S6 A-G, Dataset of the LED spectra used to train the ANN. The reference spectra of seven LEDs covering the spectral range of the device. The LEDs are centered at wavelengths of (Datasheet S6A) 505 nm, (Datasheet S6B) 565 nm, (Datasheet S6C) 595 nm, (Datasheet S6D) 660 nm, (Datasheet S6E) 730 nm, (Datasheet S6F) 850 nm, and (Datasheet S6G) 940 nm. Each LED was measured at ten intensities, highest in blue to lowest in brown.,
Datasheet Fig. S7A-G, Dataset of the LED photocurrents vs. voltage used to train the ANN. Photocurrent as function of voltage measured for seven LEDs covering the spectral range of the device. The LEDs are centered at wavelengths of (Datasheet S7A) 505 nm, (Datasheet S7B) 565 nm, (Datasheet S7C) 595 nm, (Datasheet S7D) 660 nm, (Datasheet S7E) 730 nm, (Datasheet S7F) 850 nm, and (Datasheet S7G) 940 nm. Each LED was measured at ten intensities, highest in blue to lowest in brown.
Datasheet Fig. S8A-P, Dataset of the spectra of Laser Driven Light Source (LDLS) with various filters used to train the ANN. The spectra were measured using (Thorlabs) bandpass spectral filters with Full Width at Half Maximum (FWHM) of 40 nm, depict wavelengths centered at (Datasheet S8A) 400 (Datasheet S8B) 450 (Datasheet S8C) 500 (Datasheet S8D) 550 (Datasheet S8E) 600 (Datasheet S8F) 650 (Datasheet S8G) 700 (Datasheet S8H) 750 (Datasheet S8I) 850 nm. Additionally, spectral data were acquired utilizing (Thorlabs) bandpass filters of (Datasheet S8J) DG10-600-B (Datasheet S8K) DG10-1500-B (Datasheet S8L) FGB25 (Datasheet S8M) FGB39 (Datasheet S8N) FGV9. Furthermore, spectral measurements also incorporated in house color printed polymer filters with color codes (Datasheet S8O) 97_100_15_40 and (Datasheet S8P) 100_0_76_0.
Code/Software
The dataset comprising (i) 10 power spectra acquired from seven LED light sources; (ii) Laser-driven white light source (LDLS) with filter sets: (a) a bandpass filter set (at width of 25 nm) in the spectral range of 400-1100 nm; (b) a bandpass filter set (at width of 40 nm) in the spectral range of 400-1100 nm; (c) a set of homemade filters produced by printing transparencies. All filter sets were spectrally characterized. Overall, 50 filters were used with the white-light source. The data were split randomly into training (80%) and testing (20%) sets. The reconstruction code utilized TensorFlow, sklearn, numpy and scipy software packages.
Note: Handling of Missing Values
In this dataset, empty cells in certain columns do not signify missing or unavailable information. Instead of infilling empty cells with a specific placeholder value like "N/A" or "null," they have been intentionally left blank due to the nature of the Excel file."
This revised version clarifies that empty cells are not indicative of missing data and explains the reason for leaving them blank.
Methods
Photoresponse Characterization: All measurements of photocurrent as a function of bias voltage were performed at room temperature ( 25 ± 1 °C) under vacuum conditions at ~10-5 Torr. The photocurrent was measured with incident light modulated by a mechanical chopper at frequency of 1 kHz, and with low noise current pre-amplifier (Femto DLPCA-200) and lock-in amplifier (Model SR830). In this photocurrent measurement, the heterojunction was illuminated by seven light-emitting diodes and a Laser Driven Light Source (LDLS) as a white-light source combined with a set of bandpass filters and with transparency printed filters (see supplementary information for details). The reference spectrum of each light source was measured with a Thermo Fisher Scientific Nicolet-iS50R Fourier Transform Infrared (FTIR) spectrometer connected to an external silicon detector (Thorlabs FDS100) and the spectra were normalized to the silicon detector’s calibrated responsivity.
Computational Spectrum Reconstruction: The dataset (see Supplementary materials) comprising (i) 10 power spectra acquired from seven LED light sources; (ii) Laser-driven white light source (LDLS) with filter sets: (a) a bandpass filter set (at width of 25 nm) in the spectral range of 400-1100 nm; (b) a bandpass filter set (at width of 40 nm) in the spectral range of 400-1100 nm; (c) a set of homemade filters produced by printing transparencies. All filter sets were spectrally characterized. Overall, 50 filters were used with the white-light source. The data were split randomly into training (80%) and testing (20%) sets. The reconstruction code utilized TensorFlow, sklearn, numpy and scipy software packages.