------------------- GENERAL INFORMATION ------------------- READ ME File For 'Supporting Data for Thomas_Radford_PhD_Thesis' Dataset DOI: 10.5258/SOTON/D3717 ReadMe Author: Thomas W Radford, University of Southampton - https://orcid.org/0000-0002-3410-3206 This dataset supports the thesis entitled "Programmable photonic matrices using hybrid silicon - phase change devices" AWARDED BY: University of Southampton DATE OF AWARD: 2025 Dataset created: September, 2025 Date of data collection: between 2024/07/19 and 2024/12/16 Information about geographic location of data collection: University of Southampton, Southampton, Uk, SO17 1BJ Related projects/Funders: PhD project funded by EPSRC Associated Publications: - Blundell, Sophie, et al. "Ultracompact programmable silicon photonics using layers of low-loss phase-change material Sb2Se3 of increasing thickness." ACS photonics 12.3 (2025): 1382-1391. - Radford, Thomas W., et al. "Inverse design of unitary transmission matrices in silicon photonic coupled waveguide arrays using a neural adjoint model." ACS photonics 12.3 (2025): 1480-1493. -------------------------- SHARING/ACCESS INFORMATION -------------------------- Licenses/restrictions placed on the data, or limitations of reuse: CC BY Recommended citation for the data: Supporting data for thesis entitled: "Programmable photonic matrices using hybrid silicon - phase change devices" T.Radford, University of Southampton. -------------------- DATA & FILE OVERVIEW -------------------- This dataset contains: [Folder/File list] - PCM_material_data_files: -> "Amorphous_GST.txt" - Elipsometry data for real and imaginary refractive index -> "Amorphous_Sb2S3.txt" - Elipsometry data for real and imaginary refractive index -> "Amorphous_Sb2Se3.txt" - Elipsometry data for real and imaginary refractive index -> "Crystalline_GST.txt" - Elipsometry data for real and imaginary refractive index -> "Crystalline_Sb2S3.txt" - Elipsometry data for real and imaginary refractive index -> "Crystalline_Sb2Se3.txt" - Elipsometry data for real and imaginary refractive index - Chapter 4: -> "All_matricies_raw_data_from_images.py" - Python file containing all matrix elements for transmission matricies presented in the thesis. Corresponding figures 4.7, 4.9, 4.10 labelled accordingly. -> "Figure_4_11_spectral_sweep.py" - Data used to generate figure spectral sweep in figure 4.11 - Chapter 5: -> "3x3_CWG_random_spiked_data_network_predictions_and_resimulation.zip" Neural network generated pixel patterns and the asociated electric field predictions for 1000 random target transmission matrices. Also contained is the manually resimulated electric field for validation. -> "Thermal_flucuations.zip" Pixel pattern file predicted by the network to implement the H3 Hadamard transmission matrix. The resimualted electric field is recorded for a range of temperatures ranging from 100K to 395K -> "3x3_cwg_training_dataset.h5" - Training dataset for neural network to predict geometries of a 3x3 coupled waveguide array. -> "3x3_cwg_validation_dataset.h5" - Validation dataset for neural network to predict geometries of a 3x3 coupled waveguide array. -> "4x4_cwg_training_dataset.h5" - Training dataset for neural network to predict geometries of a 4x4 coupled waveguide array. -> "4x4_cwg_validation_dataset.h5" - Validation dataset for neural network to predict geometries of a 4x4 coupled waveguide array. -> "5x5_cwg_training_dataset.h5" - Training dataset for neural network to predict geometries of a 5x5 coupled waveguide array. -> "5x5_cwg_validation_dataset.h5" - Validation dataset for neural network to predict geometries of a 5x5 coupled waveguide array. -------------------------- METHODOLOGICAL INFORMATION -------------------------- Results presented in chapted 4 are measured using an IR camera, allowing measurement of multiple device ports at once. Image analysis allows integration of the intensity across the laser spot corresponding to each output grating coupler, results of which are presented in the supporting python file. Simulations are carried out using Lumerical MODE's VarFDTD simualtion engine as discussed in the text contained within this thesis. Matrix entried are extracted using power and frequency monitors to assess the electric field at each output waveguide, again presented in the supporting python file. Presented matricies are normalised during data processing for each figure so that the summation of each row = 1. Supporting neural network datasets are created using Lumerical MODE as outlined in the thesis using an iterative optimization process. The preidctions within "3x3_CWG_random_spiked_data_network_predictions_and_resimulation.zip" are made by the netwrok architecture defined in chapter 5 of the thesis. The training dataset for this was spiked with approximatley 5% randomly genearted patterns as discussed in the text. The neural network is presented with 1000 random unitary targets, for which it predicts a pixel pattern to implement the given transmission matrix. These pixel patterns are then manually loaded into Lumerical MODE and resimulated to generate the resimulation files. Electric fields are measured using identical field monitors in Lumerical, each monitor produces 14 datapoints. Transmission matrices are created taking the peak value of the field for both the real and imaginary part of the complex electric field, normalised to the input field. The simulated device structure is outlined in the text body and supplementary information. Pixel pattern files are constructed from -1 corresponding to crystalline Sb2Se3 and 1 corresponding to amorphous Sb2Se3. -------------------------- DATA-SPECIFIC INFORMATION -------------------------- files named "3x3_2_5_percent_supp_random_U_resim_0XXX_E_y_output_field_total.txt" contain 126 electric field datapoints, these are recorded in the file corresponding to the input (I) and output (O) as follows: I3O1 (14 datapoints) I3O2 (14 datapoints) I3O3 (14 datapoints) I2O1 (14 datapoints) I2O2 (14 datapoints) I2O3 (14 datapoints) I1O1 (14 datapoints) I1O2 (14 datapoints) I1O3 (14 datapoints) Files named "3x3_2_5_percent_supp_random_U_resim_0XXX_E_y_input_field_total.txt" contain 42 electric field datapoints whcih are recorded as followsx: I3 (14 datapoints) I2 (14 datapoints) I1 (14 datapoints)