Dependency of lightweight block ciphers over S-boxes: A deep learning based analysis

Abstract

Lightweight ciphers have been proposed from time to time to tackle various usage issues in IoT (Internet of Things) scenario. PRESENT and DoT are two such ciphers. PRESENT has been a benchmark cipher to establish and validate the security features of any new lightweight block cipher. This paper presents deep learning based analysis of a widely known lightweight cipher PRESENT and relatively less known lightweight cipher DoT. Our observations show that both the ciphers are extremely sensitive to any intentional or unintentional error occurring in their S-boxes. In our experiments, these two ciphers maintain the desired immunity to such errors when either of the randomly generated ten non-linear S-boxes are used in their corresponding algorithms. However, the ciphers fail to withstand our deep learning based attack when either no S-box is used or a reverse S-box is used in their algorithms. This work validates the fact that S-box should always be carefully chosen while designing a block cipher. The observations from our work suggest the designers to be observant towards vulnerabilities arising out of any intentional fault attacks from the adversary.