Challenging Conditions Datasets: an Extension to MOSID Dataset

This repository collects the data from the temperature and internal voltage reference sensors of 5 STM32L-DISCOVERY devices, powered by the HAMEG HMP4040 equipment, during the execution of 5 different workloads under challenging conditions. The workloads used are as follows:

20x20 Long-type matrix product.

20x20 Float-type matrix product.

Algorithm for ascending sorting, Bubble Sort.

Algorithm for 2D-point clustering, Convex Hull.

AES 128-bit encryption algorithm.

The 5 devices used, labeled here from 1 to 5, correspond respectively to the device IDs 18, 16, 19, 17, and 20 found in the MOSID dataset (https://doi.org/10.5281/zenodo.10042177).

Four scenarios were selected for the experiments, described as follows:

High Temperature Conditions Scenario: this experiment was conducted in a thermal chamber at 50 degrees Celsius with external 5V power supply. A total of 13,648,000 T-V pairs were collected.

Cold Temperature Conditions Scenario: this experiment was conducted in a thermal chamber at 5 degrees Celsius with external 5V power supply. A total of 13,648,000 T-V pairs were collected.

Undervolting Conditions Scenario: this experiment was conducted with external power supply set to 4V (the limit before communication failure begins) under normal conditions. A total of 13,648,000 T-V pairs were collected.

Accelerated Aging Scenario: for this experiment, the devices were subjected to 3 cycles of electrical and thermal stress (5.5V and 100°C) for 48 hours (equivalent to 5-8 years of accelerated aging according to Arrhenius model estimations) before performing the acquisitions under normal conditions. A total of 13,648,000 T-V pairs were collected.

The data has been structured into different .mat files, each corresponding to one of the proposed scenarios. The generated files are presented below in the order of the scenarios described:

HighTemp_Dataset.rar, containing HighTemp_Dataset.mat

LowTemp_Dataset.rar, containing LowTemp_Dataset.mat

Undervolt_Dataset.rar, containing Undervolt_Dataset.mat

AccAging_Dataset.rar, containing AccAging_Dataset.mat

Each of these files is organized in the same way: the matrices X1, X2, X3, X4, and X5, each with dimensions 5000000x2, contain the T-V samples for the algorithms [1,5] respectively, and matrix Y (with dimensions 5000000x1) contains the labels of the corresponding T-V sample pairs.