There are 33 data files here.

The "TrainingAccuracyVsHiddenNeurons.csv" file contains Training Accuracy for hidden neurons of 16, 32, 48, 
  and 64 in training phase for 3 cases of using Software, Simulated Arrays, and Experimental using partial software ReLUs.

The "TestAccuracyVsHiddenNeurons.csv" file contains Test Accuracy for hidden neurons of 16, 32, 48, 
  and 64 in inference phase for 4 cases of using Software, Simulated Arrays, Experimental using partial software ReLUs, 
  and Experimental using hardware ReLUs.
  
The "G_slow.csv" file contains coductance of all devices on the array through 39 cycles by gradually increase of Vgate of the corresponding cell.


The "Accuracy_NonLinearity.csv" file contains accuracy measured in the inference phase in the case of using different non linearity factors.

The "mnist_test_8x8_0d2Labels.csv" file contains corresponding test labels from downsampled MNIST dataset.
The "mnist_training_8x8_0d2Labels.csv" file contains corresponding train labels from downsampled MNIST dataset.

The .csv files "AC_100K.csv, AC_1Meg.csv, AC_5Meg.csv, and AC_10Meg.csv" show the results of our ReLUS for different frequencies of 100k, 1Meg, 5Meg, and 10Meg
 and each of them has 3 different columns (first column is time, second one is input voltage, and third one is output voltage). 
 To achieve the AC characteristic for input current, the input voltage should be devided by 1K since the input resistor in our experiments was 1K.
 
Also, there is a file named "AC_delay.csv" which has the same columns as others and can be used to calculate delay of ReLUs in the frequency of 1MHz.

The "DC_Gain100_25Times.csv" file contains output voltage values measured 25 times per input current for the gain of 100.
The "DC_Gain100.csv" file contains output voltage values measured 5 times per input current for the gain of 100.
The "DC_Gain1K.csv" file contains output voltage values measured for different input currents for the gain of 1K.
The "DC_Gain10K.csv" file contains output voltage values measured for different input currents for the gain of 10K.
The "DC_Gain100K.csv" file contains output voltage values measured for different input currents for the gain of 100K.

The "Accuracy_vs_ReLUs_Gain_Error.csv" file contains accuracy measured in the inference phase in the case of using different gain errors for ReLUs 
 from 0% to 30% with the step of 1%.
 
The "AccuracyVsNoise.csv" file contains the recognition accuracy vs. different values of noise on the output of ReLUs.
The "AccuracyVsNoiseZoomIn.csv" file contains the recognition accuracy vs. more zoomed in values of noise on the output of ReLUs.

The "Single_Device_Forming.csv" file contains the result of forming a device for 10 cycles by sweeping a positive
 voltage across the memristor in an order of 0 → 2 V → 0 with a current compliance of 100 μA. 
 
The "Single_Device_IV.csv" file contains the result of the DC voltage sweeping loops in which reset and set processes
 were performed 10 times. The reset process was done by negative voltage sweeping in an order of 0 → -1.5 V → 0 without 
 current compliance while a set process was done by a positive voltage sweeping in an order of 0 → 1.5 V → 0 with a current 
 compliance of 500 μA.
 
The "Accuracy_ReLUs_Mismatch.csv" file contains accuracy measured in the inference phase in the case of using 
 different percentages of ReLUs mismatch for 20 times for each.
 
The "Software.csv", "SimulatedArray.csv", and "Experimental.csv" files contain the accuracy 
 during training for different no. of weight updates using 2400 weight updates in total.