Research Article
Vulnerability Testing on the Key Scheduling Algorithm of PRESENT Using Deep Learning
@INPROCEEDINGS{10.1007/978-3-030-96791-8_23, author={Ming Duan and Rui Zhou and Chaohui Fu and Sheng Guo and Qianqiong Wu}, title={Vulnerability Testing on the Key Scheduling Algorithm of PRESENT Using Deep Learning}, proceedings={Security and Privacy in New Computing Environments. 4th EAI International Conference, SPNCE 2021, Virtual Event, December 10-11, 2021, Proceedings}, proceedings_a={SPNCE}, year={2022}, month={3}, keywords={Key scheduling algorithm Deep learning PRESENT}, doi={10.1007/978-3-030-96791-8_23} }- Ming Duan
Rui Zhou
Chaohui Fu
Sheng Guo
Qianqiong Wu
Year: 2022
Vulnerability Testing on the Key Scheduling Algorithm of PRESENT Using Deep Learning
SPNCE
Springer
DOI: 10.1007/978-3-030-96791-8_23
Abstract
PRESENT is a lightweight block cipher developed for extremely constrained environment such as RFID tags and IoT (Internet of Things). In 2020, Pareek et al. suggested a neural network to retrieving the 80-bit key of block cipher PRESENT from the last round subkeys and they arrived at a conclusion that the key scheduling algorithm is strong enough against its neural aided attack. While in this paper, we get a contradict result. We present two different experiments to test the vulnerability of key scheduling algorithm using deep learning. First, we build a 3-depth Fully Connected Neural Network to retrieve the master key bit by bit. The result is that we can predict about 80% bits of a random key with very high accuracy (above 0.9). Furthermore, we train a Residual Neural Network for classification. Compared with Fully Connected Network, we need less networks and the success rate is 100%. Finally, we combine the two networks to retrieve the whole 80-bit key from the last 64-bit round subkey. We think this method can be applied to the key schedule of other block ciphers or other similar cryptographic processes.
