This document provides metadata for all data provided within Raw Data.zip The following dataset is organised into 3 folders. Fig. 2 (28-02-20) Fig. 4 & 5 (28-02-20) Fig. 7 & 8 (18-03-20) These folders correspond to figures showcased in the publication. This document will provide metadata for each subfolder of the experiments. ------------------------------------------- File(s) Address: Fig.2 (28-02-20), [Contains (2 data files), Corresponding Publication Fig(s): Fig. 2] 'The free running spectra and the subkect to injection spectra were saved. Temperature remained constant. The following devices were measured. DATA137 corresponds to the free running spectra of device serial no. RC 23xxx1ffp-1306 driven with 6.5mA at 293K. DATA138 corresponds to the spectra of device serial no. RC 23xxx1ffp-1306 driven at 6.5mA at 293K when subject to optical injection of 127 microW at -5.64 GHz from the orthogonal mode of the VCSEL. This corresponds to a spiking VCSEL output. Raw spectra were saved using an optical spectrum analyser. Equipment: Optical Spectrum Analyser (Anritsu MS9710B 0.6-1.75um) File(s) Size: 31 KB File(s) format: .txt Processing software used: MATLAB R2018a. Date taken: 28-02-20 Location: TIC 507j, Institute of Photonics, University of Strathclyde. Name of person generating: Joshua Robertson & Yahui Zhang File(s) ID: DATA137 DATA138' ------------------------------------------- File(s) Address: Fig. 4 & 5 (28-02-20)/ AWG Input, [Contains (2 data files), Corresponding Publication Fig(s): Fig. 4 & 5] 'A proof of concept demonstration for convolution is performed using the example image and kernel given in Fig. 3. The convolution operation is performed on a 3x3 pixel image. The convolution was performed using a unique kernel. Both Image and Kernel are generated by an arbitrary waveform generator before being encoded separately into the optical injection of the VCSEL. These .csv fles were used to generate the image and kernel at 12GSa/s repectively. Measurements of the successfully encoded image and kernel were recorded as well as the combined final input injected into the VCSEL. The spiking output of the VCSEL was also recorded (Fig. 4 & 5). File(s) Size: 3 KB File(s) format: .csv Processing software used: MATLAB R2018a Date taken: 28-02-2020 Location: TIC 507j, Institute of Photonics, University of Strathclyde. Name of person generating: Joshua Robertson & Yahui Zhang File(s) ID: Test Image Input Test Kernel Input' ------------------------------------------- File(s) Address: Fig. 4 & 5 (28-02-20)/ AWG Input, [Contains (4 data files), Corresponding Publication Fig(s): Fig. 4 & 5] 'A proof of concept demonstration for convolution is performed using the example image and kernel given in Fig. 3. The convolution operation is performed on a 3x3 pixel image. The convolution was performed using a unique kernel. Both Image and Kernel are generated by an arbitrary waveform generator before being encoded separately into the optical injection of the VCSEL. A 1300 nm tunable master laser signal, encoded with image input from the intial convolution step, was injected into the orthogonal mode of VCSEL device named 'RayCan 23xxx1ffp 1306'. VCSEL Temperature (293K), VCSEL bias (6.5 mA), injection power(127 microW), injection detuning frequency (-5.64 GHz). Time series measurements were taken of the image input (Fig. 4a), kernel input (Fig. 4b), combined kernel and image (Fig. 4c) and VCSEL response (Fig. 4d) and saved in .bin format. MatLab scripts were used to plot the recorded time series and produce the temporal maps (Fig. 5). Equipment: 13 GHz oscilloscope (Agilent/Keysight DSO81304B High Performance Infiniium Oscilloscope) Sample rate: 40GSa/s File(s) Size: 67 KB,74 KB & 16,385 KB File(s) format: .bin Processing software used: MATLAB R2018a Date taken: 28-02-2020 Location: TIC 507j, Institute of Photonics, University of Strathclyde. Name of person generating: Joshua Robertson & Yahui Zhang File(s) ID: Convolved Image Input Image Input Kernel Input VCSEL Output' ------------------------------------------- File(s) Address: Fig. 7 & 8 (18-03-2020)/AWG Inputs, [Contains (8 data files), Corresponding Publication Fig(s): Fig. 7 & 8] 'Convolution is performed using in-series modulators and a VCSEL-Neuron device. 4 convolution operations are performed on the 28x28 pixel 'square' source image. The 4 convolution operations were performed using 4 unique kernel. 4 image files were created such that only the region of the image (Bx) applicable to each kernel was selected. All Images and Kernels were generated by an arbitrary waveform generator before being encoded separately into the optical injection of the VCSEL. These .csv fles were used to generate the images and kernels at 12GSa/s repectively. Measurements of the spiking output of the VCSEL were recorded for each of the 4 convolution operations (Fig. 7). These were then used to calculate image gradient magnitude (Fig. 8). File(s) Size: 416 KB File(s) format: .csv Processing software used: MATLAB R2018a Date taken: 18-03-2020 Location: TIC 507j, Institute of Photonics, University of Strathclyde. Name of person generating: Joshua Robertson & Yahui Zhang File(s) ID: Image Input 1 Image Input 2 Image Input 3 Image Input 4 Kernel Input 1 Kernel Input 2 Kernel Input 3 Kernel Input 4' ----------------------------------- File(s) Address: Fig. 7 & 8 (18-03-2020)/Measurements, [Contains (4 data files), Corresponding Publication Fig(s): Fig. 7 & 8] 'Convolution is performed using in-series modulators and a VCSEL-Neuron device. 4 convolution operations are performed on the 28x28 pixel 'square' source image. The 4 convolution operations were performed using 4 unique kernel. 4 image files were created such that only the region of the image (Bx) applicable to each kernel was selected. All Images and Kernels were generated by an arbitrary waveform generator before being encoded separately into the optical injection of the VCSEL. A 1300 nm tunable master laser signal, encoded with image input from the intial convolution step, was injected into the orthogonal mode of VCSEL device named 'RayCan 23xxx1ffp 1306'. VCSEL Temperature (293K), VCSEL bias (6.5 mA), injection power(127 microW), injection detuning frequency (-5.64 GHz). Time series measurements were taken for each of the 4 convolution operations (Fig. 7) and saved in .bin format. MatLab scripts were used to plot the recorded time series and produce the image gradient magnitude (Fig. 8). Equipment: 13 GHz oscilloscope (Agilent/Keysight DSO81304B High Performance Infiniium Oscilloscope) Sample rate: 40GSa/s File(s) Size: 16,385 KB File(s) format: .bin Processing software used: MATLAB R2018a Date taken: 13-03-2020 Location: TIC 507j, Institute of Photonics, University of Strathclyde. Name of person generating: Joshua Robertson & Yahui Zhang File(s) ID: Convolved Image Input Image 1 Kernel 1 Image 2 Kernel 2 Image 3 Kernel 3 Image 4 Kernel 4' -------------------------------------------- Fig. 9 and 10 showcases results calcuated using the theoretical spin flip model of a VCSEL shared in the paper. The model parameters were as stated in the paper. Convolution was performed on a source image file from the Berkeley Segmentation Data Set. Again the same kernels were applied to the image. The image inputs produced were therefore identical to the experimetal study. MatLab was used to analyse the output form the theoretical model and create the image reconstruction maps. Note: Further questions may be raised to the authors directly using appropriate channels. File(s) ID: N/A ------------------------------------------- -------------------------------------------