SETS-THIRAL-AES128-EM-Protected-TI-FPGA

About Dataset

The traces were obtained by measuring the real-time EM emission of an AES128 encryption process. These measurements are highly suitable for Side-Channel Analysis (SCA), enabling cryptographic key recovery through both statistical techniques and deep learning–based approaches. EM traces were collected using KeySight DSOS204A Oscilloscope, with the bandwidth of 2GHz and Sampling rate of 20GS/s connected to an Kintex-7 FPGA. Trigger-based synchronization is used during acquisition process to align the captured traces with encryption operations. After collection, the traces were formatted and annotated with appropriate labels to support side-channel analysis experiments and AI model development. The dataset is provided in HDF5 (.h5) format and is divided into Profiling_traces and Attack_traces groups. The Profiling_traces group consists of 100,000 EM traces, each containing 99,999 sample points along with a metadata structured array containing the corresponding plaintext, key, and ciphertext values and a label array. The Attack_traces group contains 20,000 EM traces together with the corresponding metadata structured array. Labels are generated from the first-round S-box output of the AES-128 encryption algorithm. For each trace, plaintext byte 2 (PT[2]) is XORed with byte 2 of the secret key (K[2]). The resulting value is passed through the AES S-box, and the resulting 8-bit S-box output is used to derive the label. The labels correspond to the complete S-box output value, resulting in 256 possible classes ranging from 0 to 255. Threshold Implementation (TI) is a masking-based countermeasure that provides resistance against side-channel attacks by decomposing sensitive intermediate values into multiple shares while satisfying the properties of correctness, non-completeness and uniformity. By ensuring that no individual share reveals information about the secret, TI reduces exploitable leakage from cryptographic computations. In the SETS-THIRAL repository, TI-protected implementation was incorporated to generate datasets corresponding to masked cryptographic executions. The resulting datasets enable the evaluation of leakage assessment methodologies using conventional side-channel attacks and AI-assisted attack techniques against masking-based countermeasures.