Non-Profiled Partial Nibble Recovery on Power Attack Resilient Adiabatic PRESENT Block Cipher Through AI Vulnerability Assessment

Abstract

Cipher implementations in embedded devices are often vulnerable to Side-Channel Attacks (SCA), particularly power analysis attacks. Circuit-level countermeasures aim to enhance security against SCA at the most fundamental abstraction of VLSI design. Artificial Intelligence (AI) techniques, by eliminating the need for prior knowledge of cryptographic algorithms, have emerged as powerful tools for executing effective SCAs on secure implementations. This study investigates the resilience of encryption keys in a circuit-level implementation by analyzing the complexity of nibble-based attacks. Specifically, Multi-Layer Perceptron (MLP) models are employed to target the round structure of the PRESENT lightweight block cipher. The cipher is implemented using Charge Balancing Symmetric Pre-resolve Adiabatic Logic (CBSPAL), a secure adiabatic logic style, and its susceptibility to power analysis is rigorously evaluated. This research represents a pioneering effort in circuit-level analysis focusing on partial nibble recovery. The CBSPAL-SCA dataset captured at the circuit-level with the 4-bit substitution box (sBox) as the Point of Interest (PoI) is used to conduct non-profiled power analysis attacks across two distinct encryption keys, leveraging various activation functions. The results show that the AI-based nibble attack fails in fully recovering the secret key, thereby demonstrating the effectiveness and power analysis attack resistance of the secure CBSPAL based VLSI IP.

Keywords

• Artificial intelligence methodologies
• nibble attack
• power analysis attack
• PRESENT
• secure adiabatic logic